《花旗银行（Citi GPS）：2024金融领域的人工智能研究报告（英文版）（124页）.pdf》由会员分享，可在线阅读，更多相关《花旗银行（Citi GPS）：2024金融领域的人工智能研究报告（英文版）（124页）.pdf（124页珍藏版）》请在三个皮匠报告上搜索。

1、Citi is one of the worlds largest financial institutions,operating in all major established and emerging markets.Across these world markets,our employees conduct an ongoing multi-disciplinary conversation accessing information,analyzing data,developing insights,and formulating advice.As our premier

2、thought leadership product,Citi GPS is designed to help our readers navigate the global economys most demanding challenges and to anticipate future themes and trends in a fast-changing and interconnected world.Citi GPS accesses the best elements of our global conversation and harvests the thought le

3、adership of a wide range of senior professionals across our firm.This is not a research report and does not constitute advice on investments or a solicitations to buy or sell any financial instruments.For more information on Citi GPS,please visit our website at GPS:Global Perspectives&Solutions June

4、 2024AI in FinanceBot,Bank&Beyond Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 Ronit Ghose,CFA Head of Future of Finance Citi Global Insights Sophia Bantanidis Future of Finance Analyst Citi Global Insights Kaiwan Master Future of Finance Analyst Citi Global Insights Ronak S Shah Future

5、 of Finance Analyst Citi Global Insights Yirou Yu FinTech&Blockchain Citi Business Advisory Services Prag Sharma Global Head Artificial Intelligence Centre of Excellence Citigroup Katie Dilaj Global Head of Product Marketing,Treasury and Trade Barry Van Kooij Global Head of TWCS Product Marketing Ci

6、ti Aldona Zajac Product Marketing,Treasury and Trade Solutions Citi Expert Contributors Bin Ren Founder and CEO SigTech David G.W.Birch Principal at 15Mb Ltd.Global Ambassador for Consult Hyperion Dessislava Savova Partner and Head of the Continental Europe Tech Group Clifford Chance Devika Kornbach

7、er Partner and Co-Chair,Global Tech Group Clifford Chance Eric Chang Co-Founder and CEO Claira Erkin Adylov Founder and CEO Behavox Ken Sena Co-Founder and CEO Aiera Kirsty Rutter FinTech Investment Director Lloyds Banking Group Matthew Van Buskirk Founder and CEO Hummingbird RegTech Sandeep Alur Di

8、rector Microsoft Technology Center Shameek Kundu Chief Data Officer and AI Entrepreneur Shelby Austin Founder and CEO Arteria AI Inc Spencer Reich Founder and CEO The SER Group Vijay Balasubramaniyan Co-Founder and CEO Pindrop Zor Gorelov Co-Founder and CEO Kasisto With thanks to:Damian Kowalski,Fai

9、zan Khwaja,Jiten Mehta,Arash Eskandari,Santiago Dinelli and Mike Lu June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 3 AI IN FINANCE Bot,Bank&Beyond Artificial Intelligence(AI)could be the General-Purpose Technology(GPT)of the 2020s-2030s.And it will profoundly change finance

10、 and money.GPTs have the potential to transform entire economies,changing the way we live and work.They create new opportunities for growth and innovation,often improving our overall quality of life.They also destroy existing ways of doing things.And as such they also create losers.Especially in the

11、 short term.The steam engine commoditized production and physical movement,powering the industrial revolution.More recently,the Internet revolutionized communication and ushered in the age of information.Similarly,AI may commoditize human intelligence,including analysis,decision making and content c

12、reation.Finance will be at the forefront of the changes.Existing jobs have disappeared in prior cycles,to be replaced by new ones.So have firms.What a bank or financial firm looks like in the mid-2020s,be it retail or wholesale finance,looks very different to the mid-1980s,or the mid-1940s!AI will r

13、epeat this cycle,possibly speeding it up.AI itself has gone through many waves of hype and disillusionment from the 1950s onwards(see Appendix for a summary history,especially Figure 40).We wrote about AI in Finance in our 2018 GPS report(The Bank of the Future The ABCs of Digital Disruption in Fina

14、nce),but interest in the topic is much higher now.Advances in generative AI(GenAI),including the release of ChatGPT in November 2022,marked a turning point.GenAI brought a user interface(UI)to AI and placed it literally in the palm of our hands.The advent of GenAI brought AI to the masses,sparking i

15、nterest among consumers and key decision makers alike.Generative AI,a subset of AI,refers to models that can generate high-quality text,images,videos,and other data.For now,GenAI in finance is largely at a proof of concept(POC)stage,but it is rapidly transitioning.In this report,we discuss what use

16、cases are likely in 2024-25,and also look further ahead.Based on the results of a recent Citi TTS Client Survey,we estimate the global banking sector 2028E profit pool could increase 9%or$170 billion from the adoption of AI,rising from just over$1.7 trillion to close to$2 trillion.The tech adoption

17、strategy of most incumbents involves adding it on top of existing products or using the new technology to improve productivity.Startups,by contrast,use new technology to unbundle what incumbents do.Disruptive startups create new products and services that are native to the new technology.1 1 Benedic

18、t Evans in Slush Conference in Helsinki,AI and Everything Else,December 2023 AI will transform money and finance.Artificial intelligence may commoditize human intelligence.Most incumbents technology adoption strategy involves adding it on top of existing products.Disruptive startups create new produ

19、cts/services that are native to the new technology.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 4 As AI-powered agents,bots and beyond,become increasingly prevalent,how will money and finance change?How will the underlying concepts and structures of finance be reshaped?In

20、 a bot-to-bot world,where machines transact with minimal human intervention,what does the world of money look like?Most of us have already been wowed by AI.Many of us now want to know how we will use it.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 5 Contents(A.)Finance:Wh

21、at Will Bots Mean?8 Lets Talk A-bot AI and Finance.9 Non-Human Customers Expert Interview.11 Identifying Potential Leaders and Laggards.13 Unlocking the True Promise of AI Expert Interview.17 Role of AI in Transforming Finance.18 1 Does AI Create Value for Shareholders?.18 2 How Does AI Impact the F

22、inance Workforce?.22 3 Does AI Create Value for Banks Clients?.26 The Bot Frontier Expert Interview.28(B.)How Will AI Be Used in Finance?32 AI Scepticism to FOMO Expert Interview.34 Document&Data Management with AI Expert Interview.35 1 Coding&Software.36 Measuring Developer Productivity Expert Inte

23、rview.38 2 Transaction Monitoring and Compliance.39 Scaling Compliance with AI Expert Interview.41 3 Customer Services&Chatbot 2.0.43 Chatbots Gallop From Horses to Cars Expert Interview.45 4 Credit Risk&Underwriting.47 5 Investment Research.50 Next-Gen Financial Research with AI Expert Interview.53

24、 6 Asset&Portfolio Management.55 Supercharging Capital Markets with AI Expert Interview.58 Putting AI on Portfolio Managers Team Expert Interview.60 Decentralized AI:The Next Frontier?.62 Centralized vs Decentralized AI Stack.62 Tokens in a Bot World.63 AI Agents and Tokenized Money.63(C.)Navigating

25、 AIs Potential Pitfalls 64 Managing Risks&Building Trust Expert Interview.66 Bias and Discrimination.68 Lack of Transparency&Explicability.69 Misinformation,Manipulation&Hallucination.73 Grappling with AI-Led Deepfake Proliferation.76 Detecting Giraffe Mans Voice Expert Interview.82 AI Amplifies Soc

26、ial Engineering.84 AI An Energy Guzzler.86(D.)Real Regulations for Artificial Intelligence 88 Global AI Regulatory Developments.90 International Guiding Principles for Organizations Developing Advanced AI System.92 Common Themes in Regulating the Application of AI.99 The European Unions AI Act.100 R

27、egulating AI Robots Expert Interview.103 Appendix 1:AI&General Purpose Technologies 106 Appendix 2:A to Z of Key Terminologies 114 2024 CitigroupAI In Finance:Bot,Bank&BeyondIn 5 years time,will AI increase or decrease profits?Impact of Adopting AI on Banks ProfitabilityImprove profits93%Reduce prof

28、its7%Range of profit impact 0 10%10 20%20%AI Could Add$170 billion or 9%to Global Banks Sector Profit Pool by 2028ESource:SNL Financial,IMF World Economic Outlook,Citi Treasury and Trade Solutions Survey 2024,Citi Global InsightsSource:Citi Treasury&Trade Solutions Survey 2024Number of respondents=9

29、0 including banks,insurers,and asset managers.Banking sector profit,2023ROW:15%UK:7%US/Canada 15%Asia(x-CN):18%Europe:22%China:24%Banking sector baseline profit,2028EProfit growth in-line with nominal GDPIncrease in banks profit from adoption of AI+9%increase in baseline profitsAI-led banking sector

30、 profits,2028E 0 10%increase 10 20%increase 20%increase 0 10%decline 10 20%decline 20%decline1%Respondents-$5 bn6%Respondents-$5 bn30%Respondents$82 bn52%Respondents$48 bn11%Respondents$51 bnBreakdown of banks profit increase from AI adoption$170 bn$384 bn$1,439 bn$1,822 bn$1,992 bn$170 bnSource:Cit

31、i Global InsightsPotential for AI-Led Job DisplacementWork time distribution by industry and potential AI impact(Based on their employment levels in the US in 2021)BankingInsuranceSoftwarae and platformsCapital marketsEnergyCommunications and mediaRetailIndustry averageHealthPublic serviceAerospace

32、and defenceAutomotiveHigh techTravelUtilitiesLife sciencesIndustrialComputer goods and servcieChemicalsNatural resourcesPaving the Way for AI in Financial ServicesSource:Accenture Research Report“A New Era of Generative AI for Everyone”,2023.54%48%36%40%43%33%34%31%28%30%26%30%26%38%27%25%26%24%24%2

33、0%12%14%21%14%9%13%7%9%11%9%13%6%8%6%6%8%6%6%5%5%24%26%28%29%14%21%12%22%33%35%20%13%16%15%15%17%14%13%14%11%10%12%15%17%34%33%47%38%28%26%41%51%50%41%52%50%54%57%57%64%High Strategy and ExecutionLaggardsSlower incumbent banksFast followersAgile incumbent banksDark horsesWill new players emerge?Lead

34、ersBigTechs and FinTechsHigh AI CapabilitiesLow AI CapabilitiesLow Strategy and Execution Higher potential for automation Higher potential for augmentation Lower potential for automation and augmentation Non-language tasks Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 8(A.

35、)Finance:What Will Bots Mean?Technology is a strategic priority for both the Board and the C-suite.Its become commonplace to hear bank leaders referring to banks as technology companies with a banking license.Financial firms increasingly use sophisticated technology to deliver their services and man

36、age their operations,and that tech is often a vector of competitive differentiation.Banks were early movers in the first technology wave-investing in the first mainframe computers as far back as the 1950s-before falling behind in the recent Internet and mobile era.We believe generative AI(GenAI)has

37、revolutionary potential in financial services because the sector is information rich.Data is its raw material.In many respects,finance is the perfect sector for the application of AI.Generative AI has the potential to change the world in ways that we cant even imagine.It has the power to create new

38、ideas,products,and services that will make our lives easier,more productive,and more creative BILL GATES,CO-FOUNDER,MICROSOFT2 Finance,in a world of AI-powered agents,bots and beyond,will likely face change across the board:in terms of market share,employment,and client experience.In the medium term

39、,by 2030 or before,AI-powered bots will play an increasing role in banking and finance.This will challenge many existing ways of doing business.David Griffiths,Citis Chief Technology Officer,said:“The pace of adoption and impact of Gen AI across industries has been astounding as it becomes clear tha

40、t it has the potential to revolutionize the banking industry and improve profitability.”The changes we expect are a continuation of many themes we have been writing about since our first 2016 GPS on the future of finance:technology and culture are changing,and with it so is society.This is leading t

41、o a reshaping of the business of money and the role of the players within it.This change is about to re-accelerate.The biggest new thing will be the growth of non-human customers.SHAMEEK KUNDU,CHIEF DATA OFFICER AND AI ENTREPRENEUR AI challenges that old French saying:plus a change,plus cest la mme

42、chose(“the more things change,the more they stay the same”).And the biggest new thing will be the growth of non-human bots and agents.he more things change,the more they stay the same”).This time some things really will change.Bankers may think that they lead the way.But many users are adopting tech

43、nology faster than banks or big business.AI,and GenAI,may be the latest chapter in the digital internet story of the crowds running ahead of the crown.2 Forbes,“The Most Thought-Provoking Generative Artificial Intelligence Quotes Of 2023”,29 November 2023.June 2024 Citi GPS:Citi GPS:Global Perspecti

44、ves&Solutions 2024 Citigroup 9 Lets Talk A-bot AI and Finance Financial institutions have been regularly mentioning AI in regulatory filings since 2018,the year we first discussed AI in Finance in our GPS report(The Bank of the Future The ABCs of Digital Disruption in Finance).Before that,references

45、 to AI in company filings around the world were relatively limited.An even bigger spike in external references is evident in media references to AI by financial institutions.These were exceedingly rare pre-2022.Alongside the growth in public interest in AI,and especially GenAI,financial institutions

46、 began to discuss the topic much more in press and media.Figure 1.Mentions of the Term AI by Financial Institutions in Company Fillings and Press Articles,2014-2024 Figure 2.Google Keyword Search Trends for AI-Related Terms,2022-2024 Note:Latest data as of 23 May 2024.Source:AlphaSense,Citi Global I

47、nsights Note:1 Results for the term“AI”represents average of search results for the terms“artificial intelligence”and“AI”.2 Results for all keywords pulled out individually.Source:Google Trends,Citi Global Insights For many years,AI or,strictly speaking,machine learning(ML)has been used in finance o

48、n structured data and for quantitative tasks.GenAI will now expand these use cases to unstructured data.Most enterprise data(about 80-90%)is unstructured locked away in emails,transcripts,documents,and reports.The industrys reliance on data is massive,driving everything from customer insights to ris

49、k management and fraud detection.With the advent of advanced AI analytics,banks can extract valuable insights to personalize customer products and services,optimize operational processes,and comply with regulatory requirements.Will AI Increase Finance Profits?According to a recent Citi TTS Client Su

50、rvey3,there is a very clear consensus:an overwhelming 93%of financial institution respondents said adoption of AI could improve profitability in the next 5 years,based on productivity gain expectations(Figure 10).What Magnitude of Profits?Applying the survey results to our forecast of the global ban

51、king sector profit pool,suggests AI could boost the total banking industrys 2028E profits by 9%,or$170 billion,from just over$1.8 trillion to close to$2 trillion(Figure 11).This excludes non-bank financial sector profits.Why More Profits?AI could drive productivity gains for banks by automating rout

52、ine tasks,streamlining operations,and freeing up employees to focus on higher value activities.GenAI will likely have a big impact on internal facing tasks such as content and information management,coding,and software.3 N=90;where 71 banks,16 insurance companies,3 asset managers responded.02,0004,0

53、006,0008,00010,00012,0004Q142Q154Q152Q164Q162Q174Q172Q184Q182Q194Q192Q204Q202Q214Q212Q224Q222Q234Q232Q24-tdCompany DocumentsTranscriptsPress Articles020406080100Jan-22Mar-22May-22Jul-22Sep-22Nov-22Jan-23Mar-23May-23Jul-23Sep-23Nov-23Jan-24Mar-24May-24AIGenerative AIChatGPTRelease ofOpenAIs ChatGPT(I

54、ndex:100)Banks have been discussing AI in filings since 2018,but broader public interest took off in 2023.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 10 Innovation versus Industrialization:While finance sector leaders are optimistic about the profit impact of AI,driven b

55、y productivity gain potential,we are cautious about implementation timelines,talent costs,risk of increased competition,rising client expectations,and the costs associated with increased AI-generated activity.Will Resource Use Grow?AI is going to make finance firms and others much better at managing

56、,analyzing,and creating information.But the information in the world is exploding:global data creation+50%CAGR,2023-25E(Figure 12).Jevons paradox:efficiency gains lower costs which creates more demand4.Will AI Reduce or Grow Finance Jobs?Historical technology adoption has not led to reduction of fin

57、ance workforce but has changed the workforce mix over time.New jobs are constantly created.For example,the US economy has 3x more compliance officers from 2000 to 2023(Figure 16).Governance and Talent:The growth of AI may lead to fewer low-skilled roles in operations and technology,but governance an

58、d compliance roles will continue to grow.Our survey responses suggest AI talent availability is a challenge for banks and others(Figure 15).Talent wars are not over.Humans are still in demand.Will AI Drive Better Value for Clients?Integration of AI-powered bots into retail and corporate banking repr

59、esents a significant potential transformation,offering clients benefits such as automated decision making and the search for best offers,and banks enhanced operational efficiency.Challenges from AI-powered Bots:A shift to a bot-powered world also poses concerns on data security,regulatory/compliance

60、,and ethical considerations.Since AI models are known to hallucinate and create information that does not exist,organizations run the risk of AI chatbots going fully autonomous and negatively affecting the business financially or its reputation.AI-powered clients could increase price competition in

61、the finance sector,especially in retail financial services.The balance of power may shift.Profit margins may fall.Can Banks Keep Up?Between the 1950s and 1990s,big business usually had better technology.As the Internet and mobile waves spread across the world over the past two decades,the hierarchy

62、between big business and consumers,when it came to technology,became inverted.GenAI may be the latest such wave.The Race to Adapt:AI is likely to be adopted faster by digitally native,cloud-based firms,such as FinTechs and BigTechs,with agile incumbent banks as fast followers.Many incumbents,weighed

63、 down by tech and culture debt,may lag in AI adoption and lose market share(Figure 3).Trough of Disillusionment Ahead?All technology goes through cycles:hype,disillusionment,and then mass adoption.AI expectations have been high since 2Q23.As financial firms grapple with the transition from Wow to Ho

64、w,the gap between hype and mass production currently remains wide.Digital Leviathans:Many of the largest technology companies are in an arms race to become the AI provider of choice to banks and companies.Similar to the concentration in cloud computing,AI may reinforce single points of control and t

65、hus also single points of failure.4 The Ben&Marc Show,a16z Podcast,9 May 2024 As financial firms grapple with the transition from Wow to How,the gap between the AI hype and development remains wide.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 11 Non-Human Customers Expert

66、 Interview Q:What is the state of adoption for AI applications in financial services?Let us start with traditional/predictive AI.Based on a 2022 Bank of England(BoE)survey and anecdotal evidence from speaking to clients,I would describe traditional AI adoption in financial services as:widespread,sha

67、llow,and inconsequential.Widespread refers to the large number of enterprises experimenting with AI across different use cases.According to the 2022 BoE survey,72%of firms reported they were using or developing ML applications.AI use cases exist across functions like sales,fraud detection,document p

68、rocessing,credit decisioning,pricing etc.Shallow refers to the limited scale of AI adoption across use cases.The BoE survey suggests the median number of ML applications for mainstream UK financial institutions to be just 20-30.For context,most banks would have a few hundred systems,with bigger bank

69、s even having a thousand or more models/sub-models.Inconsequential refers to the limited perceived impact of AI system failures on critical business operations.According to the BoE survey,less than 20%of the already few AI use cases were critical to business.Thus,the net sum of all AI efforts in UK

70、financial institutions averaged 4-6 applications in business-critical areas.There are exceptions,of course we know of FIs in North America,Europe,and Asia that have hundreds of models live in production.At this stage,anywhere from 10-20%of FIs would fall in this category.Q:What is the state of GenAI

71、 adoption in financial services?Current use of GenAI in production is negligible and lesser than predictive AI.We are still very early in the commercialization of GenAI in finance;they are nearly all pilots or Proof of Concepts.But I am quite optimistic about GenAIs impact over the next five years.P

72、roductivity improvements are very likely as GenAI underpins at least partial automation of manual processes e.g.,reading and writing code,document processing,internal helpdesks,sales support.But we are yet to see examples of entirely new banking products getting introduced due to GenAI.It is hard to

73、 invent products in banking but there is white space in distribution and GenAI can help make a difference there.For example,wealth management for the masses and hyper-personalization in banking.The creation of new products for a market of one,not selecting from a menu,has not been done before.The di

74、stinction of predictive and GenAI is also likely to diminish over time.For example,use of AI in finance will not only focus on generating reports,but also understanding trends,plotting future trajectories.Q:What will change due to GenAI?The biggest new thing will be the growth of non-human customers

75、.AI agents will exist in finance by 2030,possibly earlier.Clients may have AI agents and bots before banks do.Bankers often think that they are in control.They may not be.Hyper-personalization may be good in wealth management,but it can be a double-edged sword in insurance.The availability of more d

76、ata,often underpinned by AI,can allow insurers to segment risks more granularly.While this can help make pricing more risk-based,such micro-segmentation could also result in smaller risk Shameek Kundu is an experienced Chief Data Officer helping banks adopt AI responsibly.Most recently,he helped bui

77、ld the Financial Services business at AI testing and monitoring startup TruEra(acquired by Snowflake).Prior to TruEra,Shameek was Group Chief Data Officer at Standard Chartered Bank,where he helped explore/adopt AI across multiple areas.He has served on the Bank of Englands AI Public-Private Forum,a

78、nd the Monetary Authority of Singapore(MAS)initiative on Fairness,Ethics,Accountability and Transparency in AI.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 12 pools and a larger number of them,reflecting more accurately the risks of those within each pool.This could poten

79、tially affect the ability of high-risk consumers to have access to affordable insurance coverage.There is an ongoing debate if an AI-powered world is good for lower quality workers or existing superstars.In an AI-powered world,it is possible that lower quality finance sectors workers could get raise

80、d to median levels and good performers could scale to near superstar levels.5 Overall,employee productivity will improve.5 Harvard Business School,Navigating the Jagged Technological Frontier,2023 June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 13 Identifying Potential Leade

81、rs and Laggards The implementation of AI across finance is poised to significantly transform the financial services sector and economy overall.In financial services,we consider likely leaders and laggards based on two key parameters.1.AI Capability:Measures the technical proficiency,talent,infrastru

82、cture,and overall capability of an enterprise to develop,integrate,and leverage AI tools.This includes data quality,AI expertise,and technological infrastructure.2.Strategy&Execution:Strategic vision of an enterprise,senior leadership focus and execution capabilities,regulatory compliance,as well as

83、 the ability to scale and adopt AI initiatives to a changing market environment.Leaders:Companies with strong technical capabilities,AI talent,and effective leadership and execution are the likely winners.Leading BigTechs,investing heavily in AI,should predominate in this category.Some FinTechs will

84、 also join this group,possibly together with a few product and technology-focused banks.Fast Followers:Companies in this category closely monitor market innovations and adopt proven technologies via strategic investments and partnerships;these companies make the most of their strong execution capabi

85、lities to rapidly adapt and integrate AI solutions.Agile incumbent banks fall in this quadrant.Figure 3.Potential Leaders&Laggards in AI-Led Financial Services Source:Citi Global Insights Laggards:Companies that are disadvantaged with debt tech debt,culture debt,organizational debt are expected to l

86、ag behind in AI terms.This may reflect risk averse corporate leadership and/or a tight regulatory environment.This quadrant could include some traditional banks that struggle with outdated systems and bureaucratic processes,making it difficult for them to respond to market changes and compete effect

87、ively.BIGTECHS&FINTECHSAGILEINCUMBENTBANKSSLOWERINCUMBENTBANKSAI CAPABILITIESSTRATEGY&EXECUTIONHIGHLOWHIGHLOWDARKHORSESLAGGARDSLEADERSFAST FOLLOWERSSTRATEGY&EXECUTIONAI CAPABILITIESDigital natives should lead the way with agile banks likely to be fast followers.Traditional banks with outdated system

88、s risk being laggards.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 14 These views echo the results of the recent Citi TTS survey which underscore the slow adoption of AI by many banks.Nearly a quarter of banks have not started on AI adoption yet.In contrast,this number st

89、ands significantly lower at just about 5%for FinTech and insurance clients.Furthermore,68%of respondent banks are still only in the beginning stages of their AI journey(Figure 4).Figure 4.Banks Are Playing Catch Up in the AI Arms Race Source:Citi Treasury and Trade Solutions Survey 2024 Analysis of

90、developed market banks shows that those with a focus on technology tend to achieve higher profitability(Figure 5).Technology is not the only driver of profitability,of course,but it is an important factor.And in a world of exploding data,and a growing focus on AI,technology will remain an important

91、strategic differentiator.Figure 5.Banking Sector Strategic Focus on Technology vs.Profitability Notes:1 Technology focus measured as tech&communication cost as percent of operating expense over FY2021-23.2 Profitability is measured by the average return on equity over FY2021-23.3 Banks in Australia,

92、Hong Kong,Singapore,the UK,the US,and Western Europe,with market cap$1billion Source:SNL Financial,Citi Global Insights Banks globally rank among the largest tech spenders(outside of the Tech industry),with some large banks allocating close to 20%of their expenses to technology.However,a substantial

93、 portion of banks tech spend is on maintenance costs to support legacy systems,and business as usual.21%68%11%Not startedBeginningMaturingAdvanced/End StateBanksn=734%83%13%Not startedBeginningMaturingAdvanced/End StateFinTechsn=236%88%6%Not startedBeginningMaturingAdvanced/End StateInsurancen=16PRO

94、FITABILITYTECHNOLOGY FOCUSLowHighHighTech&culture debt will hold back some incumbent banks many banks are already falling behind in the AI arms race.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 15 Banks spend a smaller percentage of their tech spend on innovation,new prod

95、ucts,or business model changes.Spending on AI and especially GenAI is relatively small compared to overall tech spend.Much GenAI spend at present is on consultants and external vendors,with work still skewed toward pilot schemes rather than full production.Enterprise investment in AI is estimated to

96、 be only around$70 billion in 2023,with a fraction of that on GenAI.6 But GenAI spending is growing fast.In the first quarter this year,Accenture reported more than$600 million in new GenAI bookings alone7;Tata Consultancy Services(TCS)has reported$900 million worth of GenAI and AI sales combined.8

97、Global spending by banks on AI,including software,hardware,and services for AI-centric systems,was estimated at about$20 billion in 2023.9 With the global spend on AI across industries estimated to grow at a 25%CAGR between 2023-2026 to$300 billion,AI-related tech spend by banks will be at about$40

98、billion by 2026.The banks and financial services that claim to heavily invest in AI need to ensure they have an underlying tech stack that supports building out AI use cases,for example a solid data architecture and infrastructure,investment in cloud,etc.AI spending,be it on capex or consultants,wil

99、l be sub-optimal without getting the foundations correct.It is tough to draw company-level conclusions on current leaders and laggards on AI.A lot of comparative analysis may inevitably be based on proof of talk for example,press releases and strategy statements and not actual proof of work.Evident

100、Insights,a consulting firm,ranks 50 banks globally on parameters such as talent(45%weight),innovation(30%weight),leadership(15%weight),and transparency(10%weight).Their November 2023 Key Findings report ranks North American banks in the lead,occupying 60%of the top ten and top 20 positions.Figure 6.

101、North American Banks Lead in the Evident AI Index Source:Evident AI Index Banks Key Findings Report November 2023,Citi Global Insights 6 Menlo Ventures,2023:The State of Generative AI in the Enterprise 7 Accenture,FY24 Q2 Earnings 8-K Exhibit,21 March 2024 8 CNBC TV18,TCS CEO says AI to be the X-fac

102、tor on determining how fast clients start spending,15 April 2024 9 International Data Corporation(IDC),Worldwide Spending on AI-Centric Systems Forecast to Reach$154 Billion in 2023,07 March 2023.60%60%40%40%20%20%34%40%10%14%10%20%10%11%10%Top 10 BanksTop 20 BanksTop 35 BanksAll 50 BanksAPACUKWeste

103、rn EuropeUS+CanadaBanks spend on AI is small compared to their overall tech spend,but it is growing fast.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 16 The US banks lead is echoed in the US dominance of VC-backed funding for AI-related startups(Figure 7).This dominance c

104、an be attributed to various factors,including established ecosystems such as Silicon Valley.Figure 7.Global VC-Funding to AI-Related Companies by Region,2020-2024 10 Figure 8.Number of AI-Related Patents Granted by Country/Region Note:Calculations based on US dollar VC funding to AI-related firms by

105、 region.Source:CB Insights Research,Citi Global Insights Source:Center for Security and Emerging Technology,Stanford Human-Centered Artificial Intelligence,Artificial Intelligence Index Report 2024,Citi Global Insights The rising interest in AI has spurred a surge in research and development within

106、the field.China leads the way,accounting for more than half of all patents globally in 2022(Figure 8).The distribution of patents granted in China is heavily skewed toward government entities,academic institutions,and universities.By contrast,patents in the US are primarily concentrated among BigTec

107、hs.The EU and UK are relative laggards.Of course,the explosion in the number of patents may be more indicative of increased interest in AI and may not necessarily translate to an increased daily use of AI technology by enterprises and/or financial institutions yet.Much of what we will discuss in the

108、 rest of this report is in the early stages of commercialization.10 CB Insights Research,“State of AI Q124 Report”,08 May 2024( and UKRoW(count in 000)The explosion in the number of AI patents may not necessarily translate to adoption of enterprise AI/GenAI in financial services.June 2024 Citi GPS:C

109、iti GPS:Global Perspectives&Solutions 2024 Citigroup 17 Unlocking the True Promise of AI Expert Interview Q:We have had machine learning and AI in different forms in business and finance for decades.Why has the interest and adoption scaled at this time?I have been in this field for 25 years,but I ha

110、ve not experienced this kind of momentum.AI is not new,so why is this time so different?ChatGPT brought UI to AI and made it real for everybody individuals,institutions,corporates,senior leaders,and even policymakers.Since the launch of ChatGPT,everyone with a consumer device and Internet can experi

111、ence AI technology at scale.This is the reason everyone is more enthused about AI this time around,and the hope that AI will transform businesses is a lot more real and tangible today.Q:What were the building blocks that preceded the ChatGPT moment?Why is this time different than the previous AI wav

112、es?As with any emerging technology,AI went through a hype phase where it was considered a futuristic magic pill,to today becoming a reality.Innovations around vision and speech have progressed and scaled relatively fast.The combination of innovation in software/algorithms and hardware has made the c

113、urrent wave of AI different than previous ones.The surge in compute power has been a cornerstone in propelling the current AI wave to unprecedented heights.According to data from OpenAI,compute power has been doubling roughly every 100 days since 2012.Q:What are some of the immediate and biggest ent

114、erprise use cases of AI?I would say banking,securities business,financial services,and insurance are ahead in terms of thinking about use cases that matter most to them.Within financial services,the immediate applications could be around customer service&engagement and software&code.Next is retail,l

115、argely because there is so much of a consumer touchpoint with retail.There are ways we can make things better,faster,and engage more with customers by using AI in retail.Healthcare is quite cautious because of regulatory compliance,privacy concerns,etc.But healthcare would be the biggest beneficiary

116、 of innovation.Q:What are the potential roadblocks as we put AI in finance into production?Banks and financial services firms need to make AI governance mainstream.AI model explicability is a key factor to prove how banking decisions were made using AI tools.Regulators want to ensure a human is fina

117、lly responsible for any decision that has been made for any banking or financial services business.Banks must also ensure responsible and ethical use of AI.Ive not found many financial services firms where you have a chief ethics officer.How do you build and train models,who governs these model outp

118、uts,are there any bias in the models,is there any cyber-risk involved in using AI,are we helping or harming the consumer using AI?Sandeep Alur is a Director at Microsoft Technology Center,an Experience Center dedicated to empowering customers on their digital transformation journey.Citi GPS:Citi GPS

119、:Global Perspectives&Solutions June 2024 2024 Citigroup 18 Role of AI in Transforming Finance 1 Does AI Create Value for Shareholders?The popularity boom in GenAI has led to a renewed interest by key decision makers in all forms of AI and ML applications.This popularity boom is a double-edged sword,

120、leading to greater focus on the space,likely higher spend,and the risk of a potential overhyping of expectations.Mobile phones and the Internet helped create many novel business opportunities.Similarly,AI could create new business models,transform processes,generate efficiencies,and more.But the jum

121、p from innovation to industrialization,while becoming faster,still takes more time to achieve than creating a PPT presentation.Management of financial services firms are bullish.Senior bankers expect an increase in profitability from the adoption of AI.Findings from the Citi TTS survey suggest an ov

122、erwhelming majority of the respondents(93%)expect an increase in profit in the next 5 years(Figure 10).We discuss AI use cases in finance and the potential resulting productivity boost in greater detail later in the chapter.Figure 10.Majority of the Citi Financial Institution Clients Surveyed Expect

123、 AI to Drive an Increase in Profits Over the Next 5 Years Note:N=90 includes banks,insurance companies,and asset managers.Source:Citi Treasury and Trade Solutions Survey 2024 Based on the findings from Citi TTS survey,we estimate the potential incremental net profit from adoption of AI in global ban

124、king could be around$170 billion in the next 5 years.This represents a 9%increase in profits over baseline 2028E profit estimates(Figure 11).Given the productivity gains touted for AI use cases in finance,especially from the ability of increasing the output of knowledge workers in a white-collar dom

125、inated sector such as banking,a 9%profit boost may seem a reasonable forecast.But we would caution that activity levels and competition may also change dramatically.52%30%11%6%1%0-10%10-20%More Than 20%0-10%More Than 20%Profits IncreaseProfits Decrease93%of surveyed respondents expect adoption of AI

126、 will help increase profits in the next 5Y.7%of surveyed respondents expect adoption of AI will decrease profits in the next 5Y.Figure 9.Impact from Adoption of AI on Profitability Over the Next 5 years Number of respondents=90 including banks,insurers,and asset managers.Source:Citi Treasury&Trade S

127、olutions Survey 2024 Senior bankers expect an increase in profitability from adoption of AI in the next five years.Improve93%Decline7%Impact ofAI on Profits in next 5 yearsn=90June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 19 Figure 11.Adoption of AI Could Add$170 billion o

128、r 9%to the Global Banking Sector Profit Pool by 2028E Notes:1 Global banking sector profit sourced from SNL Financial,represents sum of over 3,700 diversified/regional commercial banks,thrift&savings banks FY2023 data.2 Profit growth 2023-28E estimated in-line with nominal GDP growth(2024:4.5%;2025:

129、4.8%;2026:5.0%;2027:4.9%;and 2028:4.9%(IMF World Economic Outlook).3 Increase in banks profit from AI adoption based on weighted average findings from survey(Figure 10).N=90 includes banks,insurance companies,&asset managers.Source:SNL Financial,IMF World Economic Outlook,Citi Treasury and Trade Sol

130、utions Survey 2024,Citi Global Insights Activity Levels will Explode,Due to GenAI Yes,AI will enable service providers,including banks and finance companies,to become more productive.But remember clients,competitors,and regulators will all also have AI in their toolkit.And many clients may adopt AI

131、faster than the banks themselves,especially if they work in digitally native sectors.AI/GenAI will lead to a new explosion of data and information.An almost incomprehensible amount of data is created every day.In 2023,the amount of data created globally was estimated at 120 zettabytes(ZB)compared to

132、 64 ZB of data in 2020.1 zettabyte(ZB)is equivalent to 1 trillion gigabytes(GB).Estimates suggest global data creation in 2025 could increase to nearly 181 ZB,representing an increase of nearly 51%over 2023.Since 2010,the amount of data created in the world has grown at a CAGR of 35%.With the increa

133、sing use of GenAI,even faster growth rates of data creation are likely.As the amount of data in the world explodes,the GenAI-powered bank or finance firm employee will have to deal with even more information requests from their clients and regulators that will also be GenAI-powered.There will likely

134、 be an arms race between AI generating productivity gains and creating more activity.China(24%)Europe(22%)Asia(x-CN)(18%)US&Canada(15%)UK(7%)RoW(15%)$1,439 billion$384 billion$1,822 billion$170 billion$1,992 billionBankingSector Profit,2023Profit Growthin-line with Nominal GDPBanking SectorBaseline

135、Profit,2028EEstimated Increasein Banks Profitfrom Adoption of AIAI-led BankingSector Profits,2028E+9%increase in2028E baseline profit52%respondents$48 bn30%respondents$82 bn11%respondents$51 bn6%resp.,-$5 bn1%resp.,-$5 bnBREAKDOWN OF BANKS PROFIT INCREASE FROM ADOPTION OF AI0-10%increase10-20%increa

136、seOver 20%increase0-10%decline10-20%declineMore than 20%decline$170 BILLIONAI will likely bring productivity gains,but activity levels will grow exponentially.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 20 Figure 12.Amount of Data Generated Each Year Has Grown 60 x Since

137、 201011 Note:1 Refers to the amount of data that is newly generated,captured,copied,or consumed.2 1 zettabyte(ZB)is equivalent to 1 billion terabyte(TB)or 1 trillion gigabytes(GB).Source:Exploding Topics(Statista,Bernard Marr&Co.)(for data from 2010 to 2025)The importance of regulation in finance wi

138、ll also slow productivity gains from AI.As we discuss later in the report,policy and regulatory approaches vary a lot between countries.But in most countries,banking and finance are highly regulated industries and the adoption of AI by firms in this space will be heavily scrutinized.Furthermore,regu

139、lators are likely to make increased information and analytical demands from financial services companies using AI themselves.Regulators will likely be able to take data from banks and ingest it into their AI tools,scanning for any red flags,inconsistencies,and potential deviations from guidelines.Th

140、is may mean finance firms do not witness material decline in headcount as the volume of activity soars.And any gains made due to a reduction in content and coding-related headcount may be partially or completely offset by an increase in the headcount for AI-related compliance managers and ethics and

141、 governance staff.Clients Powered by AI Could Change Behavior and Profits The stickiness in client relationships helps support revenue and profit margins in consumer banking which,especially for market leaders with scale economies,can be highly attractive.Consumer and SME clients are typically price

142、 takers and can exhibit inertia in their banking behavior,which helps bank profits.The stickiness of consumer banking client relationships remains remarkable.Adults in the US,on average,continue with the same savings and checking accounts for about 17 years.Even members of the younger generation(age

143、d 26-32)have held their checking account(9 years)and savings account(7 years)for a relatively long time.Sticky client relationships have been attributed to reasons such as:1 these were the accounts they always had;2 happy with the customer service;and 3 too much of a hassle to switch.12 Also,crucial

144、ly,most consumers do not want to spend time thinking about money which is usually a means to do something else.The typical client desire not to think about financial services reinforces client inertia.11 Exploding Topics,“Amount of Data Created Daily”,13 December 2023 12 Bankrate,Survey:Consumers st

145、ick with the same checking account for an average of 17 years,04 January 2022.257913 16 9970000222023202420252030Data Generated(zettabytes)35%CAGRAI agents could disrupt the customer relationship stickiness that supports bank revenues

146、 and profit margins.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 21 Could AI change the current equilibrium?In a bot-to-bot banking world,an AI-powered consumer bot would get to work by a simple voice command from the human:Check my finances.The consumer support bot could

147、 quickly provide a detailed report of the actions it has taken or recommendations.The bot informs you that it has refinanced your mortgage rate,paid off the credit card balance,or negotiated the credit card debt and moved the excess funds to a higher interest account after analyzing the users histor

148、ical spending data.Or more likely,the bot informs the clients of actions it recommends and asks for approval.Consumer clients usually do not want to spend time thinking about their finances.And soon there may be an opportunity to delegate all retail banking and consumer finance tasks to an autonomou

149、s AI agent.Of course,humans may still choose not to engage the bot or follow through with its recommendations.It has been argued that that AI-enabled disruption is likely to happen earlier in the areas where:1 there is high customer friction;and 2 the AI is designed to perform a specific task or a l

150、imited range of tasks.For example,refinancing mortgage debt is a tedious process for humans but not as complex as tax or wealth management,hence easier for AI agents to disrupt.Activities with relatively low consumer friction can also suffer from inertia and these tasks may also move.Moving money to

151、 a higher interest savings account or buying single stocks are largely low-friction tasks but still riddled with consumer inertia and could also be disrupted as manual processes become further automated by machines(Figure 13).Figure 13.Potential for AI Agents to Disrupt Consumer Finance13 Source:a16

152、z Consumer,Citi Global Insights Incumbent banks,especially in the retail and SME segments,make profits from customer inefficiency and inertia.Theyre protected by high switching costs and lack of client knowledge or interest.Many borrowers either do not know they can save by refinancing,or do not wan

153、t to go to the effort required to make it happen.AI-powered finance advisors could,in theory,transform the consumer space by 2030 and we discuss in greater detail later in this chapter.Of course,this will be in the context of local market structures and,importantly,local regulations.These idiosyncra

154、tic factors may significantly slow down the pace of adoption of AI agents.13 a16z,Money on Autopilot:The Future of AI x Personal Finance,27 July 2023 HIGHAIDISCRETIONLOWCONSUMERFRICTIONHIGHCONSUMERFRICTIONLOWAIDISCRETIONDEBTREFINANCETAXFILLINGWEALTHMANAGEMENTSAVINGACCOUNTSBUYINGSTOCKSAI could transf

155、orm consumer finance activities,especially where there is currently high friction,and the service has simple rules and guidelines.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 22 2 How Does AI Impact the Finance Workforce?Wider adoption of AI will likely bring productivity

156、 gains to the finance sector by automating and augmenting current tasks and roles.Artificial intelligence is hitting the global labor market like a tsunami.We have very little time to get people and businesses ready for it.KRISTALINA GEORGIEVA,MANAGING DIRECTOR,INTERNATIONAL MONETARY FUND14 Accordin

157、g to Accenture,nearly two-thirds of all work done in banking and insurance has high potential for AI-driven automation or augmentation.15 This is greater than the overall economy average of 40%.According to World Economic Forum,nearly a quarter of all jobs globally will change in the next five years

158、.16 Figure 14.Potential for AI-Led Job Displacement is Highest Amongst Financial Services Note:Details for work-time distribution by industry and potential AI impact sourced from Accenture Research(A new era of generative AI for everyone,2023).Details on top job growth and loss sourced from World Ec

159、onomic Forum(Future of Jobs Report 2023).Source:Accenture Research,World Economic Forum,Citi Global Insights Is AI a level-setter?Results from a 2023 study by Harvard Business School and Boston Consulting Group suggest that while all employees benefited when presented with an AI tool(GPT-4),those sc

160、oring below average in the assessment task exhibited 43%improvement in their performance while those scoring above average in the assessment task exhibited 17%improvement in their performance.This suggests AI tools could likely help reduce performance disparities17.14 Reuters,AI Hitting Labor Forces

161、 Like a Tsunami IMF Chief,14 May 2024 15 Accenture,A New Era of Generative AI for Everyone,2023 16 World Economic Forum,Future of Jobs Report,May 2023 17 Harvard Business School,Navigating the Jagged Technological Frontier,2023 54%48%36%40%43%33%34%31%28%30%26%30%26%38%27%25%26%24%24%20%12%14%21%14%

162、9%13%7%9%11%9%13%6%8%6%6%8%6%6%5%5%24%26%28%29%14%21%12%22%33%35%20%13%16%15%15%17%14%13%14%11%10%12%15%17%34%33%47%38%28%26%41%51%50%41%52%50%54%57%57%64%BankingInsuranceSoftware&PlatformsCapital MarketsEnergyCommunications&MediaRetailIndustry AverageHealthPublic ServiceAerospace&DefenceAutomativeH

163、igh TechTravelUtilitiesLife SciencesIndustrialComputer Goods&ServicesChemicalsNatural ResoucesHigher potential for automationHigher potential for augmentationLower potential for augmentation&automationNon-language tasksWork time distribution by industry and potential AI impactBased on their employme

164、nt levels in the US in 202140%of working hours across industries can be impacted by Large Language ModelsTop Job Growth and LossOne millionLost jobsStable jobsNew jobs23%of todays job will changeJune 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 23 Building Resilience Through R

165、eskilling Technological innovation and resulting disruptions fundamentally alter the workforce mix over time.The further integration of AI into finance in the coming years will speed up this transformation.In the Citi TTS client survey,lack of availability of the right(AI)talent was cited by 55%of t

166、he responding banks as a primary barrier in their AI journey,followed by risk and compliance(42%)and legacy technology(38%).Hence,it is important that we train our existing workforce to be flexible and adaptable.Figure 15.Right Talent,Risk&Compliance and Legacy Tech is Major Barrier for AI Adoption

167、Note:N=71 Banks;57=Others.Others include asset managers,FinTech,insurance companies.Source:Citi Treasury and Trade Solutions Survey 2024 As most organizations are struggling to find the right AI talent,many are looking to balance the demand-supply gap by upskilling,reskilling,and redeploying talent

168、internally or from within the industry.A recent McKinsey&Company survey highlights 56%of its sample are likely to hire GenAI talent internally versus 30%who would hire externally.18 The biggest source of AI talent into banks is still another bank.19 Our recent report,Citi GPS:What Machines Cant Mast

169、er Human Skills to Thrive in the Age of AI,argues in 3-5 years from now,all teams and tasks will likely have embedded AI as colleagues and hence human skills such as critical and analytical thinking,creativity,and ethics will be in greater demand.20 Early users of AI tools have often found the quali

170、ty of output generated by AI tools unreliable and inconsistent.21 Banks would be subject to financial penalty and/or reputational harm if AI bots give out incorrect price quotes or engage in unreliable behavior.A human checker or supervisor will be required to police the AI bots.18 McKinsey&Company,

171、The Human Side of Generative AI:Creating a Path to Productivity,18 March 2024 19 Evident AI,Key Findings Report,November 2023 20 Citi GPS,What Machines Cant Master Humans Skills to Thrive in the Age of AI,31 May 2024 21 The New York Times,Press Pause on the Silicon Valley Hype Machine,15 May 2024 55

172、%42%38%35%27%31%20%11%1%39%40%9%25%2%25%11%32%2%TalentSecurity,Risk&ComplianceLegacyTech/Org.StructureFundingTechMaturityRegulationCultureFindingUse-CasesOtherBankOthersIn 3-5 years,all teams and tasks will likely have embedded AI colleagues.Skills such as critical&analytical thinking,creativity,and

173、 ethics will be in demand.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 24 As AI technology advances,developers are likely to become cheaper and more expendable.This is likely to be true even for bank employees.The nature of jobs will change,with basic roles being replaced

174、 by technology.Those with the best ideas will thrive,earning substantial rewards,whilst others face displacement.ERKIN ADYLOV,CEO&FOUNDER,BEHAVOX In the evolving workplace that is driven by AI,every employee will inevitably take on managerial responsibilities as they engage with AI-driven tools.We w

175、ill see a rise in AI guardrail officers or some form of AI compliance officer,a role that more broadly has already grown 3x in the last two decades(Figure 16).Figure 16.Fastest-Growing Occupations in Last Two Decades(%Change from 2000 to 2023)Source:U.S.Bureau of Labor Statistics,Citi Global Insight

176、s As AI systems generate significant portions of the output driven by client and/or regulator demand,employees must oversee and refine this output,ensuring it aligns with the local/global laws,regulatory standards&guidelines,and organizational goals&objectives.Historical Impact of Technology on Jobs

177、 Employment across industries has seen significant change in the past decades.In 1950,three out of ten US jobs were in manufacturing(30.2%).This fell to one out of every 12 US jobs by 2020(8.4%).By contrast,the share of jobs in finance grew 40%from 1950(4.1%of total jobs)to 2020(5.8%of total jobs).E

178、ven during the 2010s tech revolution,and in the aftermath of the 2008 subprime crisis,US financial jobs as a percentage of total employees in the economy remained surprisingly stable at 6%,while US commercial bank employees as a percentage of total finance jobs declined slightly to 15%(2023)from 18%

179、(2011).22 22 Federal Reserve Bank of St.Louis and the U.S.Bureau of Labor Statistics 413%376%282%276%202%Manicurists&PedicuristsHR ManagerMeeting,Convention&Event PlannersMassageTherapistsComplianceOfficersIn the AI-enabled workplace,every employee will inevitably take on managerial responsibilities

180、.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 25 Figure 17.Change in Jobs in the US from 1950 to 2020,By Sector Figure 18.Bank&Finance Jobs Remain Stable Over the Last Decade Source:U.S.Bureau of Labor Statistics,Citi Global Insights Source:Federal Reserve Bank of St.Loui

181、s,U.S.Bureau of Labor Statistics AI-led disruption of jobs and tasks are likely to change the nature of the workforce.While GenAI will likely change individual finance jobs and tasks,and how they are staffed and managed,it is interesting to note that historically adoption of new technology has not l

182、ed to rapid changes in the finance sector workforce.For example,the introduction of ATMs starting in the late 1960s did not lead to a drop in the number of human tellers employed by banks.On the contrary,between the 1970s and mid-2000s,the number of human tellers employed soared,as the US economy an

183、d the financial sector grew rapidly.Similarly,the advent of spreadsheet tools(e.g.,VisiCalc,SuperCalc)did not lead to a drop in bookkeeping and accounting jobs.Instead,the new tools helped accountants work smarter allowing them to add value,as the software could do tasks in seconds that would takes

184、hours for humans to complete.Accounting and financial management jobs became more valuable for businesses as they could now focus on analysis,forecasting,and strategy and not just numerical compilation tasks.The growth of financial analysis tools created a new growing class of financial managers fro

185、m the 1980s onwards.Over the next decade,look out for the rise of AI managers and similar roles.Figure 19.Introduction of ATMs Did Not Have Immediate and High Impact on Bank Teller Jobs Figure 20.Advent of Spreadsheets Did Not Reduce the Number of Accountant Roles 1 American Banker,Kevin Wack,Alan K

186、line,“The Evolution of the ATM“,May 2017 Source:FDIC,U.S.Bureau of Labor Statistics,Citi Global Insights Note:a.Bookkeeping,accounting,and auditing clerks produce financial records for organizations&check financial records for accuracy.b.Financial managers create financial reports,direct investment

187、activities,and develop plans for the long-term financial goals of their organization.1 Encyclopedia Britannica,“Spreadsheet”,last updated on 23 May 2023 Source:U.S.Bureau of Labor Statistics,Citi Global Insights 30%8%22%18%4%6%7%14%5%16%6%11%14%15%19502020OtherGovernmentLeisure&HospitalityEducation&

188、HealthProfessional&Business ServicesFinancial ServicesInformationTrade,Transport&UtilitiesManufacturingConstructionMining&Logging17.9%15.1%6.0%6.1%20112023Commercial Bank Employees to Total Finance EmployeesTotal Finance Employees to Total Employees000060072001980199

189、0200020102020Number of Branches(in 000)Number of Bank Tellers/ATMs(in 000)Number of Bank Tellers Employed in the USNumber of Commercial Bank Branches in the US(RHS)First ATM in the US installed on 02 Sept.1969 at Chemical Bank1Over 10,000 ATMs installedOver 80,000 ATMs installedNearly 300,000 ATMs i

190、nstalled04008001,2001,6002,0002,4000200020102020Labour Force (in 000)Bookkeeping,Accounting,and Auditing ClerksFinancial ManagersSpreadsheet programVisiCalc,written for Apple II in 1979 1Spreadsheet programSuperCalc,Multiplan,&Lotus 1-2-3 inttroduced in early 1980sHistorical adoption of n

191、ew technology has not led to rapid changes in the finance sector workforce.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 26 3 Does AI Create Value for Banks Clients?AI is likely to transform banking from the inside and the outside.From the inside,it is automating repetitiv

192、e tasks and augmenting information heavy tasks,thus bringing productivity benefits(value to employees,potentially)which may translate into more profits(value to shareholders,potentially).On the outside,it promises to enhance customer experience through AI agents(negotiating price on our behalf and m

193、anaging administrative matters),better customer service(chatbot 2.0),and tailored offers(hyper-personalization).Banking and wealth management clients should be net winners from this.However,AI could also have some negative impacts on some finance clients,especially edge cases in insurance.Access to

194、insurance could be limited,or premiums could rise significantly,for higher-risk individuals or properties where big data insights or AI analysis identify greater risks.Regulation will be needed to manage some of these negative side effects of AI.FinTechs promise over the past decade has been to crea

195、te more options,better prices,greater convenience,and a better experience for consumers.AI could be the latest wave of technology to turbo-charge this promise.It could further empower bank clients,especially consumer and SME clients.The future of retail and corporate banking is likely to be more AI

196、and data driven than before.We believe banks customers(retail,SMEs,and corporate)are also likely be AI-enabled in the future,not just banks and finance firms.This could tilt the playing field,including for prices,a bit more toward consumers.Generative AI,with proper guardrails,can effectively deciph

197、er meaning and provide comprehensive answers.Integrating domain specific LLMs into banking will be transformative.ZOR GORELOV,CO-FOUNDER AND CEO,KASISTO Autonomous Agents The next frontier for AI innovation is likely to be autonomous agents.Instead of prompting the LLMs with simple one-line instruct

198、ions,BigTech companies could embed higher AI capabilities in their existing digital assistant suite such as Apples Siri,Amazons Alexa,and Google Assistant.With many BigTech companies foraying into financial services,consumers are likely to delegate financial decisions to these AI assistants.For exam

199、ple,Googles Bard can process an investors existing investment portfolio and calculate returns relative to the broader market.This functionality is likely to get more sophisticated over time and could involve partnering with different bank and FinTech partners.23 These bots,equipped with sophisticate

200、d algorithms and access to vast amounts of data,will negotiate with financial institutions to secure the best possible deals for their users.This shift will not only streamline financial services but also ensure that decisions are made with a level of precision and foresight that human clients may n

201、ot achieve on their own.So,what will AI-powered retail banking look like?23 Medium,How Google Bard Helps Investors,09 December 2023 AI will create more value for banking and wealth management clients,whilst edge cases in insurance likely face insurability.June 2024 Citi GPS:Citi GPS:Global Perspecti

202、ves&Solutions 2024 Citigroup 27 AI-Enabled Bank Customer:As AI technology advances and wider adoption of autonomous AI agents proliferate,consumers will likely be enabled to make informed financial decisions without them being directly involved,especially in data gathering,comparison shopping,and ca

203、rrying out manual tasks to execute the transaction.Consumers can focus on the yes/no/switch decisions.Enhanced Financial Health:More than three-quarters of Americans are worried about their financial situation,including insufficient funds for retirement(68%),an inability to manage the cost of living

204、(56%),and debt levels(45%).24 The concept of open banking and open finance will allow consumer banking bots to access consumers financial data,making real-time decisions to optimize their financial well-being.Right Bot vs.Right Bank:In this new paradigm,the critical decision for consumers will be se

205、lecting the right bot rather than the right bank.Choosing bot-powered advisors,much like choosing human personal finance advisors,will become a key task.But who will the bots work for?Advisory firms,banks,FinTech,BigTech,or some genre of new service?Will Everyone Have a Bot?Leading banks and financi

206、al firms will most likely opt to provide their own AI-powered personal finance services.But they will also be joined by FinTech and BigTech firms that may have a competitive advantage in terms of being digitally native and have faster go-to-market speed.In some markets,they may also have stronger co

207、nsumer brands.How Will We Choose Our Bot?A universal truth is that most consumers do not want to spend a lot of time on financial services topics.Finance is a means to an end the mortgage is a means to living in your home.Will consumers want to spend too much time thinking about which autonomous age

208、nt to use?Will we default to known and trusted brands?Albeit these may not be bank brands!The Squeezed Middle?Many smaller banks,for example community banks,will continue to base their client proposition on local market knowledge and client proximity.In wealth management and private banking,similar

209、to institutional banking,the human in the loop will remain crucial,albeit they will be increasingly AI-augmented.As always,the players in the middle will get squeezed.Real Life Application of a Customer Bot to Negotiate Debt Collection Credit card debt in the US is at all-time high of$1.13 trillion

210、at the end of 2023 and 25%of those with credit card debt are paying less toward it than they should,leading to a cycle of revolving credit card debt.25 Cambio offers an AI-powered service that negotiates debt collections on behalf of consumers.In the first 60 days of launch,the AI-powered consumer b

211、ot helped around 70%of customers resolve their collections and raise their credit score.24 CNBC Select,77%of Americans are anxious about their financial situation,20 May 2024 25 CNBC,Nearly 1 in 4 Americans with debt are putting less money toward credit card payments,22 February 2024 The next fronti

212、er in innovation will be clients using AI bots to manage their finances.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 28 The Bot Frontier Expert Interview Q:Why do we need bots to do banking?How will it benefit bank customers?David Birch:I see significant scope for AI-led

213、transformation in banking and finance.For example,tasks like opening an individual savings account(ISA)can be a complex and time-consuming task,even for someone with a financial advisor.It involves extensive analysis of different types of ISAs,their cost structures,and offered benefits,etc.Even a ba

214、sic bot could help make better and faster decisions.However,in retail,we are likely to see discontinuity as traditional marketing strategies will not work on bots.For instance,a bot is not influenced by brand loyalty or advertisements;it operates within the model parameters on APIs.The concept of va

215、lue will be critical.Customers are likely to deal with bots that reflect their values and interests.This could lead to the creation of a marketplace for bots where individuals select a bot from a range of options based on desired preferences.Kirsty Rutter:The fundamental premise behind implementatio

216、n of AI-driven client interfaces is that if technology exists,why should humans waste time on petty administrative tasks.Instead,executive bots can perform these tasks on our behalf.In a financial management context,an AI bot can directly engage with the bank on behalf of the customer.For example,if

217、 a mortgage is up for renewal,the bot can notify the customer of the impending end date,conduct research to find best deals,determine the right product,and fulfill formalities for refixing the mortgage rate.Bots can achieve this based on set parameters(retirement age,preference for debt tenure,maxim

218、um acceptable price,etc.),pre-determined by the customer.In the initial phase,the bot could rely on human confirmation before executing tasks;but as trust in the bot grows,several of these tasks can be handled autonomously.Q:How will banking products and services change to serve AI bots?How will ban

219、k customers having autonomous bots affect banks profit pool?Kirsty Rutter:The banking products will have to be designed for AI bots but need to be codified in a manner humans can comprehend.Consumers must be able to understand and interpret the product or service.Products need to be reliably designe

220、d with the appropriate parameters for the intended consumer whether human or bot for this to be true the controls and compliance requirements of any product must be designed in from the beginning.With this outcome in mind,products must be designed primarily for bots,as they are the target users,but

221、they must also be translated in a way that makes them comprehensible to humans.David Birch:There is a significant difference between designing a product/service for people and designing one for bots.For starters,products/services intended for bots can be more complex and sophisticated than those for

222、 humans,as bots can better comprehend and manage these complexities.Further,the impact of using bots for financial products/services could have a substantial impact on banks profitability.For example,a bot managing a savings account could shift funds to optimize interest rates far more frequently th

223、an a human ordinarily would.This rapid movement of money,further catalyzed by open banking initiatives,could dramatically impact banks liquidity management as a large number of small David G.W.Birch is an author,advisor,and commentator on digital financial services.David is Principal at 15Mb Ltd.,hi

224、s advisory practice;and Global Ambassador for Consult Hyperion,the secure electronic transactions consultancy;and holds several other board-level advisory roles.Kirsty Rutter is the FinTech Investment Director at Lloyds Banking Group.In the past,she has held several strategic roles in prominent inst

225、itutions like Barclays Capital,Credit Suisse,Eaton Risk Advisory,and the Ministry of Justice UK.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 29 balances could move from one bank to another and impact profitability.Thus,the strategic implications are substantial and intrig

226、uing,necessitating immediate attention,despite the seemingly distant horizon.Q:How far are we from having bots represent us in choosing financial products and services?Kirsty Rutter:For executing transactions through personal bots,several conditions must be met.This includes cloud-native technologie

227、s with embedded controls and regulations.These are critical to show regulators that decisions/outcomes are fair and responsible.Achieving this level of sophistication requires continuous rethinking and retraining necessitating the combination of product design,data engineering,risk and compliance,an

228、d technological savviness.While the design is not inherently complex,it necessitates a significant shift in competencies and capabilities.David Birch:This change cannot be achieved overnight,rather it needs to be part of the banks strategic planning for the next three to five years.In my view,the ti

229、meline is primarily determined by regulatory developments.As for technology,we already have some of the basic components needed.For example,leveraging ChatGPT plugins and UKs Open Banking regime,it is possible to use ChatGPT to interact with your bank account,even today.However,the regulatory framew

230、ork needs careful consideration.We,as an industry,need to define the appropriate regulatory measures for such advancements.While the technology exists,practical regulations for bots will take time.Q:Since bots lack loyalty/emotions,what will banks compete on to secure clients wallet?What are their c

231、ompetitive advantages in a bot-driven world?David Birch:Banks will likely need to compete on price to retain clients wallet share.Bots can compare multiple prices simultaneously,so being the cheapest will likely always be a significant factor.Banks could adopt an execution strategy focused on operat

232、ional efficiency,enabling them to compete on volumes and win deals purely based on price.Additionally,banks will need to optimize their operations for the bot,which might differ from human preferences.For example,bots might prioritize factors like the speed of API response,API uptime,data accuracy,a

233、nd richness of API data.Optimizing for these factors could provide banks a competitive edge.Kirsty Rutter:Lastly,banks could compete by aligning with the ethical values and personalized frameworks set by the customers bot a reflection of the individuals morals and beliefs.Some customers might priori

234、tize price,whilst others might focus on sustainability aspects.In response to the changing demands of the market,there are likely to be new opportunities for products and services to support the hyper personalized outcome many individuals and industry seek.Citi GPS:Citi GPS:Global Perspectives&Solut

235、ions June 2024 2024 Citigroup 30 Corporate Client Bots We have largely focused on consumer clients so far.Let us take a closer look at scenarios where corporates deploy more sophisticated AI bots for their banking needs.Thus,the bot reviews the companys liquidity status,forecasts future cash flows,i

236、dentifies potential investment opportunities,and flags financial risks all in real-time.The bot could renegotiate loan terms with banks overnight,optimize the companys investment portfolio,close out any unutilized overdraft facility to strengthen financial positions,and ensure more regular complianc

237、e and regulatory reporting tasks,too.This sophisticated AI integration would transform the landscape of corporate banking.Key benefits of corporate bots in banking are:Onboarding:Customer onboarding(KYC and CDD processes),loan origination,and compliance checks.These processes,which are even more com

238、plicated and time-consuming in corporate than retail banking,often involve multiple touchpoints,paperwork,and approval and review cycles.This process requires an ability to efficiently validate,understand,and store content of documents to correctly implement client instructions.Documents often lack

239、standard formats and are not fully digitized.Figure 21 shows two processes.On the left-hand side,a process using established technology is shown,with automation implemented to orchestrate a process.However,once the orchestration completes,a manual process is required.This contrasts with the image on

240、 the right-hand side where AI is deployed to process most cases,leaving a few outliers that require manual intervention.Feedback Loop:The true voice of corporate clients is often obscured by limited feedback channels,rigid surveys,and generic market research.GenAI unlocks novel approaches as it empo

241、wers clients to express feedback in more natural ways and move away from static forms.This enables AI to extract insights that truly illuminate hidden needs and preferences,which fuels a virtuous cycle where client feedback directly shapes transformative,highly tailored products.For example:a client

242、 states the original intent to transfer funds and receives guidance how to best complete,then finishes a wire transfer and the chatbot asks,How could we make this process even easier next time?The clients detailed response is far more valuable than a 5-star rating.Querying Own Payments Data:Clients

243、will be able to effortlessly interact with their data through a GenAI interface without the need for them to be proficient in data manipulation tools or be confined to a set of predefined data analysis features within an application.Instead,they have the flexibility to ask questions in natural langu

244、age,such as,Who are my top suppliers and customers,based on payment value and volume,segmented by geography and payment methods?The responses go beyond mere data;they are intuitive insights in natural language,accompanied by visualizations and the capability to explore what if hypothetical scenarios

245、.The system provides text summaries that dynamically update as the data evolves,offering a seamless and accessible experience.Damian Kowalski Global Receivables Treasury and Trade Solutions Faizan Khwaja Global Receivables Treasury and Trade Solutions Jiten Mehta Global Payments and Receivables Trea

246、sury and Trade Solutions Arash Eskandari Global Payments and Receivables Treasury and Trade Solutions Santiago Dinelli Global Payments Treasury and Trade Solutions Mike Lu Emerging Payments Treasury and Trade Solutions June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 31 Regul

247、atory&Compliance Insights:Graph LLM-powered tools could automate the monitoring and analysis of regulatory and compliance changes with the help of natural language processing algorithms.Relevant updates identified by such tools could then be used not only by the bank to adapt its processes and polic

248、ies accordingly,but also form part of the advisory information that clients can benefit from while incorporating changes into their business.Banks transactional processing flows would also benefit from ongoing compliance checking against such consolidated regulatory positions,flagging non-compliance

249、 early and allowing both the corporate and the bank to address the gaps and avoid penalties.Figure 21.Client Onboarding Process:Existing Process vs AI Enhanced Process Source:Citi Treasury and Trade Solutions There are some drawbacks to letting AI do banking on your behalf.The extensive use of AI an

250、d open banking APIs raises concerns about data security and privacy.Ensuring that sensitive financial data is protected from cyber threats is paramount.Over-reliance on AI bots may lead to vulnerabilities if the technology fails or is compromised.Navigating the complex regulatory landscape can be ch

251、allenging,especially as regulations evolve to keep pace with technological advancements.Ensuring compliance across multiple jurisdictions may require significant effort and resources.Capture Reporter DetailsCapture ID&VInformationCapture Customer Form InformationCustomer MatchingInformationCustomer

252、Account List DetailsOther Supporting DocumentsDocument StoreCase EvaluationList of Simple CasesList of Complex CasesCriteria MetStep 1Step 2Staff Review&Process All CasesCapture Reporter DetailsCapture ID&VInformationCapture Customer Form InformationCustomer MatchingInformationCustomer Account List

253、DetailsOther Supporting DocumentsDocument StoreData Capture&UpdateSuccessfulOutcomeStep 1Staff Review&ProcessIncomplete/Sample CasesRules ProcessingAI LLMNo/Too ComplexOtherSystemsUpdateYesCaptureComponentLogicComponentHumanProcessExternal/LLMComponentLegendsEXISTINGPROCESSENHANCEDPROCESS Citi GPS:C

254、iti GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 32(B.)How Will AI Be Used in Finance?For many years,AI,or strictly speaking ML,has been used in finance on structured data and for quantitative tasks.Today,AI is used in finance primarily for risk and pricing.GenAI will expand use cases

255、to new areas.In the short to medium term,we expect the biggest impact at incumbent financial institutions to be on internal facing tasks and improvements in productivity rather than lots of new products.Incumbents will focus on improvements in areas such as software&coding,transaction monitoring&com

256、pliance,and more.A lot of bank functions such as credit underwriting,algorithmic trading,portfolio construction,and transaction monitoring already utilise AI/deep learning applications.GenAI will create new opportunities beyond productivity improvements but some of the more blue-skies work newer pro

257、ducts&services,bots using tokenized money,and decentralized AI will likely take time to build and be rolled out to market.Autonomous AI agents could turbo charge existing business models and commercial relationships and lead to the creation of new ones.Will AI agents drive adoption of digitally nati

258、ve money tokens and lead to another stage in the disruption of financial services?This is quite possible by 2030,and we discuss this in this chapter.But the exact timing of large-scale usage of AI agents is for now unclear.Business models need to adapt,regulations need to be introduced and/or update

259、d,client behaviour and employee skills need to evolve.The time lag between invention and industrialisation of a technology can be significant.And in the case of financial services,there are the additional challenges of being heavily regulated.In this chapter,we largely focus on the likely use cases

260、of GenAI that will be used at scale in finance in the next couple of years.Changes to how finance does code and software,or search and summarization,are likely in 2024-25(Figure 22).Existing ML/AI use cases in compliance and risk management will be GenAI augmented.Core finance functions such as risk

261、 analysis and underwriting will also be changed,as will investment research and portfolio management,albeit more slowly.Figure 22.Summary of Generative AI Use Cases in Finance,Estimated Wider Adoption&Potential Impact on Tasks Source:Citi Global Insights Impact on Tasks(Low to High)Time to Market(No

262、w to 2030)AI Use-Cases in Financial ServicesCODING&SOFTWARESEARCH&SUMMARIZATIONTRANSACTIONMONITORING,COMPLIANCE&CONDUCTCUSTOMERSERVICE&CHATBOT2.0CREDITRISK&UNDERWRITINGINVESTMENTRESEARCHASSET&PORTFOLIOMANAGEMENTGenAI will create new opportunities beyond productivity improvements,but some of the more

263、 blue-skies work will likely take time to build business models will need to adapt,regulations need updating,and skills must evolve.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 33 An overview of the main AI/GenAI use cases in finance are below:Coding&Software:Large banks

264、have tens of thousands of employees for software and programming tasks,often about 15-25%of the banks total workforce.AI can help streamline coding processes by automating repetitive tasks,optimizing basic coding,and accelerating development cycles.Search&Summarization:The financial services sector

265、is characterized by its data and document-intensive nature.AI tools can sift through vast volumes of data,distill pertinent information,and deliver concise summaries that can be used as input for faster decision making and executing next actions.Transaction Monitoring,Compliance&Conduct:AI-powered s

266、ystems can excel in monitoring external and internal conduct.By continuously analyzing transactions and behavior,and detecting anomalies in real-time,AI mitigates risks,ensures regulatory adherence,and minimizes fraudulent activities.Customer Service&Chatbot 2.0:AI-powered chatbots could enhance cus

267、tomer service by delivering personalized experiences,resolving queries promptly by having access to customer data,while simulating human-like interaction and providing round-the-clock assistance.Credit Risk&Underwriting:AI algorithms could leverage vast datasets to assess credit risk and facilitate

268、underwriting processes with potentially better accuracy and speed compared to humans.AI models could make better credit decisions by analyzing diverse and non-traditional data sources.Investment Research:Fundamental research involves a lot of search and summarization of information,datasets,and gene

269、ration of text and charts.Generative AI could likely bring a lot of time/cost efficiencies as it could automate a lot of information search and retrieval tasks.Asset&Portfolio Management:AI could help identify investment opportunities,optimize asset allocations,and help personalize portfolios in a s

270、calable manner thus enabling advisors and portfolio managers to focus more on high-value activities such as client engagement and alpha generation.Decentralized AI:The extremely centralized nature and economic model of AI stack makes a compelling case for having a co-existing decentralized AI stack

271、which will have a slightly different economic model and would use tokens or tokenized money to compensate cloud providers(for compute)and the model owners.Money&Finance in the Bot World:We believe humans will increasingly adopt AI bots and agents to carry out financial activities and transactions.To

272、kenized moneys features such as programmability and the smart contract functionality makes it ripe for wider adoption by tech-native AI agents who can operate 24x7 and use tokenized money(e.g.,stablecoins,CBDCs)to enable atomic settlement(i.e.,instant and simultaneous settlement)in the bot-based fin

273、ancial ecosystem.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 34 AI Scepticism to FOMO Expert Interview Q:How has application of AI/ML evolved in finance and compliance?In the last five years,AI systems have got sophisticated,led by breakthroughs in AI architecture.Recent

274、 AI models can include several trillions of parameters.These advancements enabled the development of scalable AI systems that can be used in compliance,including detection of money laundering.Compliance practices have also evolved,moving away from a keyword search approach to more intuitive and comp

275、rehensive AI solutions that help reduce the number of false positives.Today,the initial skepticism around AI adoption in compliance has been replaced by the fear of missing out.Q:Under what circumstances is it preferable to build a foundation model as opposed to employing a retrieval augmented gener

276、ation approach for clients?Retrieval Augmented Generation(RAG)is one of the applications of LLMs that leverage custom datasets.They are most relevant in cases where information retrieval from a specific knowledge base is key,as opposed to generating content.For instance,retrieving and analyze inform

277、ation from compliance manuals.In instances requiring high volumes(e.g.,20 million queries a day),classification could be a more efficient alternative.For this,a custom foundation model must be built and then fine-tuned to specific needs.This approach is most relevant for finance as it ensures compli

278、ance with regulatory requirements,including auditability.Q:Will the future of AI be shaped by large general-purpose models or by smaller domain-specific models?Both large general-purpose models and smaller domain-specific models will play a significant role in shaping the future of AI.Domain-specifi

279、c models will deliver superior performance to specific industries they are purpose-built for.Meanwhile,large models such as GPT-5,GPT-6,GPT-7 will excel in general domain understanding.Eventually,as we progress towards artificial general intelligence,there may come a time when a single model dominat

280、es all domains.Q:What are model performance degradations?Why is it relevant in finance?Model performance degradation typically occurs from frequent model changes.Implementing feedback loops or automated training can lead to fluctuations in model performance(for the better or worse).However,it must b

281、e noted that such practices are not permissible for regulated financial institutions.Regulatory requirements mandate locking in the model version to prevent degradation.Q:Will GenAI automate coders or increase the demand for coders?As AI technology advances,developers are likely to become cheaper an

282、d more expendable.This is likely to be true not only for developers,but all professionals involved in content generation such as bank employees,journalists,equity research analysts,etc.The job market is likely to evolve into a celebrity economy,where only those with the best ideas will thrive,earnin

283、g substantial rewards,whilst others face displacement.Erkin Adylov is the founder and CEO of Behavox and leads product innovation at the company.Prior to founding Behavox in 2014,Erkin worked at Goldman Sachs in equity research and later at GLG Partners/Man Group as a portfolio manager in the Financ

284、ials Fund.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 35 Document&Data Management with AI Expert Interview Q:How smart is AI?We often think of AI as a single entity,but it comprises multiple small,specialized models performing different tasks.Think of them like different

285、 layers in a tiramisu,stacked atop another.While impressive,AI in its current form is far away from human-level intelligence.LLMs lack the ability to plan,reason,and understand the physical world.In four years,a child will have seen 50 times more data than the biggest LLMs.And,just like a 4-year-old

286、,AI makes mistakes.It requires supervision.Human involvement in AI is crucial,especially in the context of finance.Q:How can AI tools help manage unstructured data in large enterprises?80-90%of data in any large enterprise tends to be unstructured(i.e.,data found in documents,call transcripts,record

287、ings).This data is not available in spreadsheets or databases and is difficult to quickly summarize by simply glancing through it.Some of the tasks where AI can assist in context of unstructured data include:Metadata Extraction:AI can help identify and extract metadata such as name and addresses fro

288、m documents(e.g.,extract name or date from passport).Text Extraction:AI tools can help locate and extract specific text from complex documents like prospectus,offering circulars,agreements,or other repositories.Analytics and Reporting:AI tools can help analyze extracted data,identify key trends,and

289、generate reports with the aim of enhancing productivity.Obligations Management:AI can help monitor regulatory or client obligations and alert organizations to any upcoming compliance requirements/fillings.Document Generation:AI can help gather information and generate documents that are written repe

290、atedly throughout the year or update them from routinely.Q:What are the best practices to ensure accuracy and regulatory compliance when using AI for generating content in financial services?Data security,privacy,and compliance with regulatory requirements are critical when using AI in financial ser

291、vices.Secondly,select the right tools for the task.For example,LLMs are popular today,but they are not best suited for all tasks due to their tendency to hallucinate.It is preferable to use a more reliable model(or set of models)where accuracy is critical(e.g.,vetting an agreement).Q:What aspect of

292、jobs will be replaced by AI tools?How can professionals approach reskilling?What new opportunities can arise?Looking at history,initial concerns about job displacement have been unfounded.This is true with the printing press,steam engine,and others each time powerful technology becomes available,job

293、 creation dwarfs displacement.The advent of AI is likely to follow a similar pattern.It should be noted that certain tasks,particularly mechanical ones,are becoming obsolete.Q:What should FIs do in next 12-18 months as they look to integrate AI?Business leaders should take a problem-first approach,f

294、ocusing on challenges rather than technique.Stop fixating on implementing specific types of AI(e.g.,deep learning,GenAI)instead,think how the right type of AI can be combined with automation and human input to create a systematic approach to problem solving.Shelby Austin is the founder and CEO of Ar

295、teria AI Inc,an AI company building modern documentation infrastructure for institutional finance.Prior to Arteria AI,Shelby was the Managing Partner of Omnia AI and sat on the executive board of Deloitte Canada.Before joining Deloitte,she founded a company that was named one of Canadas Hot 50 compa

296、nies by Profit Magazine.Shelby was named to the Macleans 2024 list of the 100 Most Powerful Canadians,one of 10 in the AI category.She was also named among Canadas Most Powerful Women:Top 100 from WXN.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 36 1 Coding&Software Techn

297、ology and coding roles in finance will be at the forefront of AI-led productivity gains(Figure 23).This is true for all industries but particularly relevant for finance given the number of developers and coders who work for large banks and the information intensive nature of the sector.Figure 23.Gen

298、erative AI Can Increase Developer Speed,But Less So for Complex Tasks Note:Exhibit from“Unleashing developer productivity with generative AI”,June 2023,McKinsey&Company,.Copyright(c)2024 McKinsey&Company.All rights reserved.Reprinted by permission.Source:McKinsey&Company A study suggests that progra

299、mmers with professional coding experience of average 6 years and about 9 hours of daily coding work see a significant productivity boost of 56%when enabled with AI tools such as GitHub Copilot.26 AI tools can assist in various facets of the software development process,including automating redundant

300、 tasks,creating and optimizing code,detecting errors,and improving the user experience.Automated reviews can help identify issues related to coding standards,best practices,and potential vulnerabilities.For example,GitHub Copilot focuses on generating code based on context and existing code examples

301、.Having been trained on billions of lines of code,it can act as a pair programmer for developers,automatically generating code,providing,and implementing suggestions.A developer productivity boost of 10-50%,weighted for a 15-25%share of staff employed in technology roles,translates to potential cost

302、 savings in the range of$2-16 billion annually,which is equivalent to 1-6%of annual US banking sector profits.This is based on current software programmer salaries in US and India.27 26 Sida Peng et al,“The Impact of AI on Developer Productivity:Evidence from GitHub Copilot”,13 February 2023.27 Mich

303、ael Page,Page Insights Salary Guide 2024.Our analysis assumes,as a simplification,that the US bank sector has half its technology employees in the US and half in India.020406080100CodeDocumentationCodeGenerationCodeRefactoringHigh-ComplexityTasksWithout generative AIWith generative AITask completion

304、 time using generative AI,%45-5035-4520-3010AI can enhance programmer productivity by optimizing code creation and implementation,automating tasks,and improving code quality,yielding substantial cost savings.AI-assisted coding and software jobs can save the US banking sector$2-16 billion annually(1-

305、6%of US bank profits).June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 2024 Citigroup 37 Figure 24.Productivity Boost at Tech Roles in Banks Figure 25.Productivity Boost as Percent of US Banking Sector Profits Source:McKinsey&Company,Federal Reserve Bank of St.Louis,U.S.Bureau of Labor Stat

306、istics,Page Insights Salary Guide,Citi Global Insights Source:McKinsey&Company,Federal Reserve Bank of St.Louis,U.S.Bureau of Labor Statistics,Page Insights Salary Guide,FDIC,Citi Global Insights AI paired with programming and automation introduces multiple benefits throughout the software developme

307、nt life cycle.Automation could reduce the amount of time developers spend on mundane and repetitive tasks,which allows them to focus on more complex and creative aspects of their work.Furthermore,as AI automation is scalable,it makes it suitable for projects of varying sizes and complexities.The ada

308、ptability of AI tools to different programming languages and frameworks enhances their utility across the board,albeit GenAI is not as suited to older coding languages such as COBOL(see expert interview below).While the use of AI in software coding and programming might bring productivity gains,ther

309、e are also potential drawbacks.For example,AI tools might struggle with creative problem-solving or thinking outside predefined patterns.Over-reliance on AI tools could also lead to a decline in developers critical thinking and problem-solving abilities.Lastly,the implementation of AI automation too

310、ls requires initial investment in terms of time,resources,and training.Maintaining and updating these tools to keep them up to date with evolving coding practices can also be resource intensive.2 3 3 4 5 6 6 8 9 8 10 13 9 13 16 15%20%25%of Tech Roles in Major Banks10%20%30%40%50%Productivity Boost%(

311、figures in US$billion)1%1%1%1%2%2%2%3%4%3%4%5%4%5%6%15%20%25%of Tech Roles in Major Banks10%20%30%40%50%Productivity Boost%Productivity Benefits as%of US Bank Sector ProfitsAI boosts productivity but struggles with thinking beyond the frontier,leading to drop in output quality.Citi GPS:Citi GPS:Glob

312、al Perspectives&Solutions June 2024 2024 Citigroup 38 Measuring Developer Productivity Expert Interview Q:What will be the scope of impact of AI on developer productivity?Andrej Karpathy,former director of AI at Tesla and founding member of OpenAI,tweeted on 31 December 2022,“Copilot has dramaticall

313、y accelerated my coding.Its hard to imagine going back to manual coding.Still learning to use it,but it already writes 80%of my code,80%accuracy.I dont even really code,I prompt&edit.”At a broad level,when our(Microsoft)customers have measured the value of GitHub Copilot,which writes code,they have

314、witnessed productivity improvements in the range of 35%to 55%,compared to somebody writing code without the help of any AI tool.However,there can be different levels of productivity gains and it is likely a function of the programming languages being used by the coder.It is also possible that one mi

315、ght experience more accurate responses for a particular programming language relative to another as the AI model is trained on massive composite data that already exists.Programming languages such as Java,C-Sharp,HTML,and Python are some of the most popular languages,and much of the content has been

316、 built using these open-source lines of code which are available in abundance.As a result,AI models trained on these codes become proficient in generating code in these languages.On the other hand,if one wants AI models to write a code in ABAP program(ABAP is a programming language developed by SAP

317、for the development of business applications in the SAP environment)or COBOL program(Common Business-Oriented Language“COBOL”is a level programming language for business applications and is still in use in many financial and business applications today),it will not perform very well as you might not

318、 find an extensive open-source ecosystem it is a legacy language and a legacy code.If a COBOL programmer wants to write code using GitHub Copilot,it is possible to do so and the model will generate an output to help the programmer,as it has been trained with some minimal data.However,the accuracy co

319、uld be relatively lower(roughly 10-20%lower)compared to C-Sharp,Java,or Python,which are likely to generate up to 80%accurate code.In my view,the way forward is to have domain-specific models.For example,if a financial institution has 10 million COBOL lines of code,I suggest fine tuning a LLM to cre

320、ate a bespoke solution for an industry that has legacy tech and legacy lines of codes.Sandeep Alur is a Director at Microsoft Technology Center,an Experience Center dedicated to empowering customers on their digital transformation journey.June 2024 Citi GPS:Citi GPS:Global Perspectives&Solutions 202

321、4 Citigroup 39 2 Transaction Monitoring and Compliance Financial crimes have risen with the advent of newer technologies.Advancements in AI have ushered in a new era for frauds and scams from chatbots that mimic human interactions to voice synthesis and deepfakes.Global phishing rose 341%in the peri

322、od October 2023 to March 2024,and was up 856%in the 12 months from April 2023 to March 2024.28 There were on average 31,000 phishing attacks daily in 2023,and it rose 1,265%from October 2022 to September 2023.29 A continued rise in financial crime rates imposes substantial economic costs and jeopard

323、izes regulatory compliance.In 1Q 2024,the global average weekly cyber-attacks per organization increased 28%from 4Q 2023 and 5%from 1Q 2023 to 1,308 the highest recorded in the past 13 quarters.Average weekly cyber-attacks in finance and banking reached 1,172 in 1Q 2024(-2%yoy).30 We believe AI-powe

324、red scams,frauds,and threats are likely to rise in volume,sophistication,and potential losses as cybercriminals discover novel ways to access customers personal and financial data using AI.Deepfake attacks were up 31x from 2022 to 2023.31 Equally,this bad AI can be combatted by good AI.Harnessing AI

325、 in transaction monitoring offers the potential to significantly reduce financial crime rates.By leveraging machine learning algorithms,AI systems can analyze vast volumes of transactional data in real-time,enabling the detection of suspicious activities with greater accuracy and efficiency.How Can

326、AI Improve Transaction Monitoring AI could offer significant enhancements to transaction monitoring compared to traditional methods.Firstly,AI algorithms can process vast amounts of data swiftly,enabling real-time monitoring of transactions,which is crucial in detecting fraudulent activities promptl

327、y.Criminals are incredibly agile and were using tech thats 20 years old or more to combat them.In anti-money laundering,its important for us to keep in mind that there are humans on the other side of the equation that are intelligent and adopting new technologies.One of the biggest reasons why we ha

328、ve false positives too,is we had a rules-based system.We write a rule to catch a particular pattern of activity,the bad guys notice that theyre being interdicted,they tweak some aspect of their approach,so that rule no longer triggers.MATTHEW(MATT)VAN BUSKIRK,CO-FOUNDER AND CEO,HUMMINGBIRD REGTECH T

329、hrough machine learning techniques,AI can continuously adapt to new patterns of fraudulent behavior,becoming more effective over time.Furthermore,AI can analyze diverse data sources beyond structured transaction data,including social media,emails,and other sources,providing a more comprehensive unde

330、rstanding 28 SlashNext,The State of Phishing 2024 Mid-Year Assessment 29 SlashNext,The State of Phishing 2023 30 Check Point Research,“A Shifting Attack Landscapes and Sectors in Q1 2024 with a 28%increase in cyber-attacks globally”10 April 2024.31 Onfido,Identity Fraud Report 2024 Phishing rose 341

331、%in the period October 2023 to March 2024.We believe AI-powered scams,frauds,and threats are likely to rise in volume,sophistication,and potential losses.Citi GPS:Citi GPS:Global Perspectives&Solutions June 2024 2024 Citigroup 40 of customer behavior and risk factors.This holistic approach enhances

332、the accuracy of identifying suspicious transactions while reducing false positives.Moreover,AI-powered transaction monitoring systems can incorporate advanced anomaly detection algorithms,identifying irregularities that may go unnoticed by traditional rule-based systems.This proactive approach impro

333、ves fraud detection rates and reduces financial losses.Additionally,AI can automate routine tasks,freeing up human analysts to focus on complex cases,thereby improving efficiency and reducing operational costs.Overall,AI offers a transformative solution to transaction monitoring,enhancing accuracy,efficiency,and agility in detecting and preventing fraudulent activities.Anomaly Detection:AI-powered