《IBM:利用人工智能产生ROI-六大驱动世界级成果的能力(英文版)(38页).pdf》由会员分享,可在线阅读,更多相关《IBM:利用人工智能产生ROI-六大驱动世界级成果的能力(英文版)(38页).pdf(38页珍藏版)》请在三个皮匠报告上搜索。
1、IBM Institute for Business Value|Research InsightsGenerating ROI with AISix capabilities that drive world-class results2Clients can realize the potential of AI,analytics,and data using IBMs deep industry,functional,and technical expertise;enterprise-grade technology solutions;and science-based resea
2、rch innovations.For more information about AI services from IBM Consulting,visit more information about AI solutions from IBM Software,visit more information about AI innovations from IBM Research,visit IBM can help1Most AI projects arent profitable enoughyet.As AI matures,ROI on enterprise-wide ini
3、tiatives averaged only 5.9%,well below the typical 10%cost of capital.Data is crucialbut only one piece of the puzzle.Building on trusted,high-quality data at scale boosted returns on AI investments to as much as 9%.Companies can strike gold with the right approach.Best-in-class companies that have
4、built six mature capabilities reported average ROI of 13%on AI projects.Mature capabilities differentiate the AI projects that deliver the highest ROI.Key takeaways23All eyes are on AI Artificial intelligence is suddenly sexyagain.Generative AI has taken the business world by storm,with large langua
5、ge models(LLMs)such as OpenAIs ChatGPT,Baidus ERNIE,Googles LaMDA and Facebooks LLaMA splashed across the news.And executives arent immune to the hype.References to AI on earnings calls were up 77%year-over-year in early 2023.1 And the money is following.AI is becoming an ever-larger component of IT
6、 budgets,with worldwide spending on AI-centric systems expected to hit$154 billion this yearup 27%over 2022.2 But will enterprises spend these resources wisely?As AI models become faster,smarter,and more reliable,organizations are racing to capitalize.Can the return on investment(ROI)match expectati
7、ons?The answer:Yesbut only if organizations take a disciplined approach.To come to this conclusion,we surveyed 2,500 global executives in 34 business and technology roles across 16 countries from all major regions.We asked them to deconstruct how their companies are investing in AI today,what real-w
8、orld ROI is being produced,and which elements are required to boost effectiveness.In partnership with Oxford Economics,we then analyzed what key business and technology capabilities are connected to the most successful AI initiatives.(See“Study approach and methodology”on page 31.)Can the ROI of AI
9、match the hype?4Average ROI of all organizationsOur findings reveal yawning outcome gaps across AI projects.Few deliver the financial value shareholders expect.In fact,average ROI on enterprise-wide initiatives is just 5.9%,well below the typical 10%cost of capital.Yet,there are distinct improvement
10、s as you move along the AI maturity continuumwith best-in-class companies reaping an enviable 13%ROI(see Figure 1).So what sets these world-class performers apart?And how can leaders across sectors learn from their success?This report will outline:Why ad hoc AI projects deliver less value than strat
11、egic programs The impact of trusted data and the virtuous cycle of AI-data symbiosis Six key capabilities that define and enable top-tier organizations.Building a mature AI organizationon a solid foundation of trustis required to unlock AIs full potential.Companies that get it right are generating s
12、ignificant business valuenot just media buzz.FIGURE 1 A cut above the restBest-in-class companies deliver AI ROI that exceeds their cost of capital.14%13%12%10%9%8%7%6%5%4%3%2%1%0%Average cost of capital for companies10%AI RoIAverage ROI for top 10%of organizations30%moreCompanies that develop matur
13、e AI capabilities will generate profitsnot just media buzz.5.9%13%5Beyond opportunistic AIOrganizations have been betting on AI for the better part of a decadeand the learning curve has been steep.3 Some got caught up in the“wow factor”of the technology,forgetting to align projects to strategy.Other
14、s saw AI as a hammer,and every business problem a nail.Almost all struggled to scale their implementations beyond experiments,proofs of concepts,and pilots.The good news from our research:Many organizations have turned a corner.AI rollouts are more successful than ever,with more than twice as many e
15、xecutives saying their organization used AI effectively in 2021(54%)than in 2020(25%).They also expect AI investment to grow to 6.5%of IT spend by 2024.Overall,return on investment has been rising steadily since 2020.For enter-prise-wide AI initiatives,average ROI has grown from just over 1%in early
16、 2020 to nearly 6%by the end of 2021.4 This could be a result of the pandemic pushing organizations to invest in AI-driven solutions that would expedite remote working,enhance the user experience,and decrease costs.To gauge whether AI ROI has kept pace with this trend,we surveyed more than 350 execu
17、tives again in April and May 2023.We found that AI ROI has continued on the expected growth trajectory,reaching an estimated 8.3%in 2022(see Figure 2).5 As organizations figure out where and how to deploy AI,bold bets translate to bigger gains.6Still,these returns are lower than the cost of capital,
18、which is typically 10%across most industries.Overall,fewer than one in four organizations in our survey say theyve achieved AI ROI higher than 10%.FIGURE 2 Returns are on the riseROI from AI projects grew more than 6 times between 2020 and 2022.10%9%8%7%6%5%4%3%2%1%0%AI RoI5.9%1.3%2.1%2020Early 2021
19、Late 2021Source:2020:Deloitte/ESL AI survey;Early 2021:IBV AI ethics survey;Late 2021:IBV AI capability survey;2022:IBV generative AI pulse survey,April-May 2023.2022(estimated)8.3%In essence,AI is following the“J-curve”pattern typical for transformative technologies.6 Adopting emerging tech at scal
20、e requires reimagining business models,workflows,skills,and many other facets of business.Returns often stagnate while teams work out the kinks.Yet,as capabilities mature,performance can rise quickly.In this environment,enterprises need to have a strategic plan for scaling the impact of AI over time
21、.7Our analysis reveals how ROI improves across a continuum of AI maturity(see Figure 3).The average enterprise,which pursues ad hoc and/or opportu-nistic AI initiatives,is the laggard.In 2021,those that intentionally embedded AI in products,services,business units,and functions saw ROI climb above 7
22、%.As maturity extended a step further,with AI deployed as part of a strategic business transfor-mation,returns improved again,to 8%.FIGURE 3 AI-centric strategies boost ROI Aligning AI with business priorities yields much higher ROI than ad hoc projects.AI strategic continuumOpportunisticHorizontall
23、y deployedVertically integratedProduct embeddedNew business model enablerStrategic importance of AIAd hocCustomer/operational effectivenessCustomer/operational effectivenessOrganic growth,innovation,and differentiationTransformation and platform strategyCompany typesVariedDeploying AI into functions
24、Integrating AI within business unit(s)Embedding AI into core productsUsing AI to help achieve platform economicsROI of organizations 4.7%7.2%7.4%8.0%As organizations figure out where and how to deploy AI,bold bets translate into bigger and bigger gains.At the top of the curve is our best-in-class ca
25、tegory,reporting an average ROI of 13%.By taking a steady,balanced approach to adopting AIwhich includes building data and analytics skills,developing a multidisciplinary approach,creating diverse teams,and training teams through AI Centers of Excellence(CoEs)theyve developed comprehensive capabilit
26、ies throughout the enterprise.8PerspectiveFrom Alphabet to Walmart:Becoming an AI-first business Legacy enterprises are like industrial-era cities.Theyre connected by winding corridors that grew organically rather than straight,efficient avenues.They face persistent challenges with modernization,as
27、progress is often hindered by century-old infrastructure.AI-first companies start with a clean sheet of paper.Leaders can be more creative,more flexibleand make emerging technologies central to the business model.“I havent encountered digital natives who say,Im going to innovate.I only hear those st
28、atements from legacy organizations,because theyre trying to get out of that rut,”said Ravi Simhambhatla,Chief Digital Officer at Avis Budget Group,in a recent IBV report.“For digital natives,its all about disrupting themselves.”7Digital natives,such as Alphabet,Netflix,Amazon,and Metahigh-growth bus
29、inesses with AI at the corehave seen outsized returns on AI investments.8 Yet some legacy brands have also thrived with AI.Walmart,for instance,uses AI to match its inventory to shifting customer needs.The brand taps customer and shopping trend data to anticipate where and when people will want spec
30、ific products.This lets Walmart stock each warehouse with the right items,stream-lining logistics and enabling speedy delivery,even during peak shopping seasons.9 This capability didnt appear overnight.It was built on responsible data collection and curation,the creation of flexible algorithms,and a
31、 holistic approach to technology.Taken together,these initiatives produce AI-driven insights Walmart can trust.Walmart,which has been a leader in data and analytics for decades,understands that AI requires measurement and optimization to achieve its full potential.Its ability to track desired outcom
32、es and diagnose challenges enables the company to drive more value from AIand build a scalable capability that can be leveraged across many current and future business applications.Walmarts success demonstrates that,while a traditional enterprise cant become a digital native,it can transform to emul
33、ate inborn agility.Companies that focus their efforts in key areasembedding AI into core operationswill see better results than those that spread themselves too thin.89To build a world-class AI organization,one factor comes first:How an organization chooses,collects,governs,and uses its data that ab
34、undant but elusive resourceeither enables or constrains what AI can achieve.Data has sometimes been compared to oil:a valuable resource thats expensive to extract and difficult to process.If dirty,it can pollute an entire ecosystem.But when tapped responsibly,its worth billions.Thats because reliabl
35、e,representational,consensual data is foundational to trustworthy AI.People wont use AI solutions they dont trustand organizations that place greater importance on AI ethics report a greater degree of trust from their customers and employees.10 It also helps close the ROI gap.Companies with high“dat
36、a wealth”arent yet at the world-class level,but they have large stores of high-quality data,monetize data effectively,and say their data is trusted by internal and external stakeholders.Our analysis reveals that these attributes drive higher-than-average ROI and enable more effective AI projects(see
37、 Figure 4).Data and AI:Feeding a virtuous cycleData can close half the gap between average and world-class AI ROI.FIGURE 4 Data differentiatorsCompanies with more holistic data practices see better business outcomes.ROI realized from overall enterprise AI capabilitiesEffectiveness of AI projectsAll
38、others4.8%9.0%47%77%Data wealth outperformers10In effect,high-quality,high-value,trusted data unlocks half of the ROI improvement we see in best-in-class organizations.That said,data alone is not enough to fully realize AIs potential.While data quality,quantity,robustness,value,and trust are all imp
39、ortant,how businesses harness data has a bigger cumulative impact on ROI than what data they have.Todays top-performing Chief Data Officers(CDOs)specialize in getting value from their organizations data.An elite groupjust 8%of CDOs in IBVs most recent survey of data leadersreap more value than peers
40、 while spending less.14 Whats key is how they use AI to improve their data:Three out of four say that applying AI to their data helps them make faster and better business decisions.So,its not just about using data to improve AIAI can also help companies make better use of data.Its a virtuous cycle.A
41、s Mirco Bharpalania,Senior Director of Cross Domain Solutions for the Lufthansa Group said,“AI is so critical because it actually opens up the world of the data that were sitting on.”15PerspectiveDisrupt like a digital native Lyft has disrupted the transportation industry with data-driven processes
42、that optimize business decisions and redefine customer experiences.By leveraging technology to tap unmet market demand,it broke$1 billion in revenue in its first year of operation.By the end of 2022,that figure had topped$4 billion.11 Lyft focuses on meeting customer needs in real time,using machine
43、 learning models to make hundreds of millions of decisions each day,including optimizing ride prices,matching riders with drivers,and predicting arrival times.12Making real-time inferences with machine learning at scale requires access to a vast amount of data and computation resources,optimized pro
44、cesses,and a talented team of data scientists,engineers,and AI experts.And Lyft has tons of information:20.3 million active riders in Q4 2022,hundreds of millions of trips per year.13 This massive store of data powers real-time business decisions that reduce costs,optimize resources,and streamline t
45、he customer journey.1011What enables some organizations to achieve world-class ROI from their AI investments?How do they amplify high-quality,trusted data to unlock financial and business value?To answer these questions,we carefully analyzed our study results,looking for patterns,insights,and applic
46、able real-world lessons.We learned that best-in-class AI performers build capabilities across six key areas,in a holistic,integrated waywith trust at the core(see Figure 5):Vision and strategy AI operating model AI engineering and operations Data and technology Talent and skills Culture and adoption
47、 Six key capabilities that enable world-class results12Talent and skillsDeploy an enterprise-wide approach to develop AI ethics,skills,and talentCulture and adoptionDevelop a human-centered approach to AIwith dynamic,open feedback loops across the ecosystem AI engineering and operationsDeploy AI sol
48、utions that are flexible,user-friendly,and scalableVision and strategyIdentify where AI can boost competitiveness,innovation,and performanceand prioritize accordinglyAI operating modelEmbed an AI operating model into the fabric and culture of the organization Data and technologyBuild core capabiliti
49、es that simplify,automate,control,and secure access to dataTrustFIGURE 5 Becoming best-in-classCompanies that see the highest ROI from AI have matured 6 key capabilitieswith trust at the core.6 key capabilities for high AI ROI13#1 Vision and strategyDont throw AI at everythingApplying AI,automation,
50、or any other technology to poorly designed processes still delivers subpar outcomes.By assessing where strategic investment is planned for core and non-core functions(for example,customer service,marketing,supply chain,finance,and so forth),as well as business units,leaders can uncover strategic opp
51、ortunities to embed AI.A well-thought-out AI strategy can catalyze transfor-mation efforts and increase the ROI of individual AI projects(see“Boston Scientific spends$50,000 to save$5 million,”page 15).According to our research,organizations that view AI as important to their business strategy are 1
52、.8 times more likely to be effective with their AI initiatives and achieve nearly twice the ROI(see Figure 6).Leaders also balance competitive differentiation with cost optimization.Some are even leveraging publicly available and open-source AI resources to deliver faster,cheaper,and more scalable s
53、olutions to market(see“Foundation models lay the groundwork for AIs future,”page 14).Ethical questions about how these tools have been trained will also play a part in AIs futureso companies need to define where they stand before they push too far forward.AI is NOT important in the area of enterpris
54、e/business strategy4.7%ROI realized from overall enterprise AI capabilitiesFIGURE 6 Strategy comes firstCompanies that use AI to advance strategy see nearly twice the ROI.8%7%6%5%4%3%2%1%0%AI IS important in the area of enterprise/business strategy8.0%1.7x more14PerspectiveFoundation models lay the
55、groundwork for AIs generative future AI is getting smarter and faster every day.But most solutions are still bespoke.Theyre trained using a specific data set to complete a pre-defined taska process that is both energy-intensive and time-consuming.16 To make AI investments more cost-effective,compani
56、es need flexible,reusable models that can be applied in a variety of waysincluding generating new content.Todays foundation models are paving a path toward this future.They offer an opportunity to accelerate and scale AI adoption,as foundation models can,in theory,be applied to many domains.For exam
57、ple,LLMs can transform how information is generated and shared across an organization.It just needs to be adapted for semantic search,classification,prediction,summarization,and translation.The adoption of foundation models is also supported by a set of emerging AI engineering best practices that ha
58、ve gone mainstream.From model development to prompt engineering,these common practices and approaches streamline collaboration across the enterpriseand the ecosystem.A set of layered stacks and AI architectures with strong open source,ecosystem,and research contributions are also giving rise to comm
59、on,re-usable development and deployment approaches.While foundation models offer real promise and potential,they also come with new challenges.For one,they require significant compute,storage,and network resources,which makes them energy intensive.Training one large natural language processing model
60、 has roughly the same carbon footprint as running five cars over their lifetime.17 It is also important to consider how the usage scale of a foundation model influences ROI.A LLM that is trained to serve hundreds of millions of users may deliver more value faster than a model that is used by only th
61、ousands.In a smaller scale deployment,optimization,fine-tuning,specialization,and portability may not translate to near-term returnseven though the LLM can be used for many downstream tasks.Other challenges that come with larger models include trustworthiness,explainability,and transparency.Addressi
62、ng these issues requires additional effort,investment and,in some cases,new inventions and solutions.Teams must understand what large models can do,how they should be deployed,and what type of data curation is required.Broader data engineering skills will be criticalas will a serious focus on ethica
63、l issues.Like any other disruptive technology,there are trade-offs that come with adopting generative AI and foundation models.Success will only come from experimentation and iteration.Especially for enterprises,this journey will involve balancing the scales between the value generative AI can creat
64、e and the investment it demands.The future of AI will be defined by those who hit the right mark.1415Case studyBoston Scientific spends$50,000 to save$5 million Boston Scientific wanted to automate its stent inspection process to improve accuracy when searching for defects,such as broken links or su
65、rface imperfec-tions.Accurate inspections are critical for successful clinical outcomes.18The company has approximately 3,000 experts doing inspections,costing several million dollars each year.Boston Scientific considered a neural network model to help cut back on manual labor,but those models requ
66、ire much more data than it had on hand.And collecting or generating this data would be impractical and cost prohibitive.The solution?First,teams scaled down the problem by focusing on smaller and narrower tasks.Then,they reduced data requirements to align with the new focus.Lastly,they leveraged“off
67、-the-shelf”open-source AI models to streamline the inspection process.The result?$5 million in direct savingsdelivered on a modest budget of roughly$50,000as well as increased accuracy.Now the companys employees can identify issues faster and focus on the important work that only humans can do.1516#
68、2 Operating model Ditch the science fair mentalityGroundbreaking AI is built on a foundation of open innovation.However,leading companies learn to mitigate against the myth that anything goes in innovation.19 To keep experiments and implementa-tions in line with strategy,organizations must treat AI
69、as a discipline.They must outline ethical principles,develop rigorous governance,and emphasize pragmatism over theory.This starts with understanding which AI operating model best aligns with the business need(for example,centralized versus hub-and-spoke versus decentralized structures).Our research
70、finds that organizations with high data wealth that have also embedded an AI operating model into the fabric and culture of the organization are able to generate up to 2.6 times more ROI than their peers.20 What does this look like?One example revolves around how companies create minimum viable prod
71、ucts(MVPs).Leaders should outline a clear process for applying AIstarting with identifying the business problem the solution hopes to solve.By setting clear goals for even experimental rollouts,companies can choose to advance only the most effective AI projects.#3 Engineering and operations Agile De
72、vOps+automated ITOps+MLOps=AIOps AI engineering and operations(AIOps)brings big ideas to life,serving as a flywheel for the operating model.It integrates people,processes,and platforms to apply AI at speed and scale(see“Bestseller unlocks AI value in fast fashion,”page 17).And organizations that suc
73、cessfully design processes that help teams build to scalewhile also monitoring the performance of AI applicationssee up to 2.6 times higher ROI.Engineering discipline can accelerate this AI flywheel and make it work effectively.Just as many companies use DevOps and other software engineering approac
74、hes to speed up projects without sacrificing quality,AIOps helps shorten development cycles,improve collaboration,increase operational efficiency,and deploy solutions more successfully.21 Standardization and structured focus are essential to keep up with the pace of innovationwithout sacrificing the
75、 principles of ethical AI.17Case studyBestseller unlocks AI value in fast fashionIn the fashion industry,around 80%of merchandise is sold across two seasons each year.Everything else is highly discountedor gets donated or dumped.This over-production translates to suboptimal profits and presents an e
76、normous sustainability issue for clothing designers and retailers.22 To help teams more accurately predict demand,clothing and accessory company Bestseller took 10,000 images(one seasons catalog)and developed an AI model for each of its four brands.In just three weeks,the company was able to develop
77、 and train a convolutional neural network to classify an image according to various features.Deep-learning details were then fed into traditional analysis models(for example,regression or principal component analysis)to help the company better understand the factors that drive sales.Incorporating th
78、is information into Bestsellers core forecasting engine increased the companys selling efficiency from 78%to 82%and reduced the number of design samples it needed to create for each brand by 15%.#4 Data and technology Support industrial-strength scalingAnyone can create a proof of concept.But for AI
79、 models to be effective,useful,and trustworthy,they must be properly integrated into operational systems.What a company can do with AI is defined,in large part,by how it selects,governs,analyzes,and applies data across the enterprise.Because humans are fallible,teams need skills and processes that h
80、elp ensure the right data is chosen to power AI models.This also has a major impact on AIs ROI.At world-class organizations,data teams review governance,management,ethics,literacy,and other frameworks needed for people to access,understandand trustenterprise data.IT teams assess infrastructure and p
81、rocesses to balance AI experimentation with industrial-strength scaling(see“How IBMs Chief Analytics Officer helps boost AI ROI,”page 18).18PerspectiveHow IBMs Chief Analytics Office helps boost AI ROI Everyone has big ideas for AI.The office of IBMs Chief Analytics Officer (CAO)helps those visions
82、become reality.The CAO office partners with business units within IBM to identify opportunities to improve revenue,save time,and add a layer of intelligence to day-to-day business workflows with machine learning.It taps enterprise-trusted data and delivers business insights into the strategic platfo
83、rms where work gets done.AI can improve the business in many ways,so the CAO must be strategic about project selection.The office uses a prioritization framework based on five criteria for enterprise-level initiatives:Drives measurable business value Aligns to IBMs strategy Leverages the CAOs data s
84、cience and AI capabilities Drives at-scale solutions Includes transformation partners.When a projects data has already been standardizedwoven into the organiza-tions data fabricthat also helps push it to the top of the list.While these initiatives often see faster speed-to-market,the CAO doesnt disc
85、ount projects that take a heavier data lift.The upfront work they require also lays the groundwork for future AI projects.Like most enterprises that werent born on the cloud,IBM has a complex data environment to rationalize.So,the CAO is focused on building a data foundation that makes it easier to
86、develop insightsnot project-by-project,but for everything the company does.Discipline also helps the CAO office speed development.Using standards and best practices in experimentation,development,and deployment protects quality as the pace picks up.Reusable assets,templates,connectors,APIs,and cookb
87、ooks help projects start quickly and helps team leverage lessons learned.19This rigor also applies to the preservation of AI ethics.Each project must be built atop the five pillars that support trustworthy AI:231.Explainability.A layperson should be able to understand how an AI system worksand how i
88、t arrived at a specific conclusionwhen given a simple explanation.2.Fairness.Bias can be present in both AI algorithms and the data used to train and test AI.Creating fair systems requires building a diverse development team and seeking input from impacted communities.3.Robustness.As cybersecurity t
89、hreats become more advanced,AI must be able to withstand interference and defend vulnerabilities.For example,organizations need a way to guard against the use of poisoned training data in AI-powered systems.4.Transparency.Transparency reinforces trust,and the best way to promote trans-parency is thr
90、ough disclosure.Users must be able to see how the service works,evaluate its functionality,and comprehend its strengths and limitations.5.Privacy.Trustworthy AI systems prioritize and safeguard consumers privacy and data rights.They provide full disclosure to users about what data is being collected
91、,how it will be used and stored,and who has access to it.On a quarterly basis,all CAO projects are reviewed.Goals may be adjustedor projects terminated.Tough decisions and trade-offs must be made,but this focus and rigor drives measurable business value:The CAO office helped generate$357 million in
92、net benefits for IBM in 2022.$278 million in benefits were associated with developing insights that directly led to revenue growth,in part by transforming client-facing capabilities.Another$79 million was tied to automating and augmenting enterprise workflows with AI,driving faster and smarter decis
93、ions.24 The CAO can now deploy projects up to nine weeks fasterwhich helps IBM bring strokes of brilliance to life with the speed and quality its enterprise clients demand.20FIGURE 7 AI requires reskillingMost CHROs have active plans to retrain workers Retrain/reskill workers impacted by AIRetrain/r
94、eskill workers that interact with AI2018202137%55%2018202141%70%#5 Talent and skillsFill the jobs of tomorrow todayA lack of skilled talent and technical expertise has been a top barrier to implementing AI since its inception.To stay competitive in a tight labor market,companies must train their tea
95、ms to use AI effectivelyand responsibly.Our research demon-strates that when organizations help teams strengthen their AI chops,AI projects are more successful.Organizations that actively encourage AI knowl-edge-sharing across the enterpriseand offer business and technical training to attract new ta
96、lentachieve ROI up to 2.6x times greater than others.HR and talent leaderswith sponsorship from the businessare driving this work.Between 2018 and 2021,the portion of CHROs with active plans to retain and reskill workers jumped notably(see Figure 7).25#6 Culture and adoption Cultivate change and co-
97、creationChange management is often the first line item cut when companies are facing cost pressures.But this fiscal austerity can be dangerously short-sighted:penny-wise and pound-foolish.The right culture one that puts a premium on trusthelps anchor AI capability and maturity.If people dont trust t
98、he work AI doesor the data its built onadoption will lag and returns will slump.On the flip side,our research reveals that maturity of culture is one of the greatest contributors to best-in-class ROI.When AI is part of the companys DNA and change management is a widespread skill,organiza-tions perfo
99、rm better.In fact,project teams that used standardized and documented methodologiesincluding value realization or benefits trackingsee up to 2.5 times the ROI.21Trust at the coreAct on principleAs government leaders debate how they should regulate the development of AI,one thing is certain:How data
100、is used will be highly scrutinized going forward.As a result,executives are investing in AI capabilities that deliver trustworthy outcomes.This requires doubling down on data technology and platforms,in addition to building core AI capabilities.As such,executives estimate that their spend on trainin
101、g,teams,processes,tools,and other operational capabilities to institutionalize AI ethics doubled between 2018 and 2021.And they expect investment to increase over the next three years.26 Yet,few organizations have put intent into practice.While more than half of organizations have endorsed principle
102、s of AI ethics,less than a quarter have operationalized them.Fewer than 20%strongly agree that their organiza-tions practices and actions on AI ethics match(or exceed)their stated principles and values.27To create trustworthy AI,organizations need to build their capabilities on a strong foundation o
103、f trusta foundation that can only exist when people across the organization agree on a set of principles,processes,and practices that will drive responsible development and innovation.22Everyone wants to be first-to-market.But is the glory worth the risk that comes with cutting corners?While new LLM
104、s promise to streamline research and content creation,they also tend to hallucinate,or confidently present inaccurate or misleading information.As a result,people arent sure if they can rely on their results.28 As some tech leaders race to advance these bleeding-edge models,others are calling for ca
105、ution and a greater focus on responsible practices.29This echoes our argument for building mature capabilities on the principle of trust.While the frenzy around the latest generative AI innovations may be warranted,enterprises need to make investments for the long haul.Building business and technica
106、l capabilities doesnt happen overnight.And quick wins wont deliver the stable returns that leaders(and shareholders)expect.Strengthening capabilities across the enterprise,however,puts organizations on the path toward transformation and growthand puts the downslope of the J-curve behind them.The hig
107、h road is longer but it may be faster23Action guideVision and strategy What value should AI bring to my company?How is AI being used to differentiate my company from its competitors?How are AI projects reviewed for strategic alignment?Do we have an enterprise-wide approach to AIand a roadmap to deli
108、ver results?Operating model Is AI embedded in our internal systems and operational processes?What is our process for developing MVPs to meet a specific business need?How are AI insights being generated and delivered to the business to create value?What checks and balances do we have in place to ensu
109、re we are using AI ethically?AI engineering and operations How are machine learning and data models being tested,deployed,and maintained?How are we tracking changes made to AI solutions?Do we have systems and processes in place to identify and solve problems as they appear?Can we measure and tune AI
110、 models once theyre deployed?Data and technology Do we have high-volume,high-quality,trusted data?Do our governance processes prioritize data security and enable trustworthy AI?What level of data advocacy and literacy exists across the organization?Do we have the information architecture in place to
111、 scale AI solutions?Talent and skills How is the organization attracting talent with data and AI skills?How are we developing AI skills and expertise across the enterprise?How are teams sharing knowledge to increase everyones comfort level with AI applications?Culture and adoption Is the organizatio
112、n change-ready?What level of change management support is available to bolster AI adoption?Do all AI projects have a named executive sponsor?Are KPIs baked into use case adoption?The secret to successfully scaling AIHow can an organization become best-in-class?That depends on where its starting from
113、.The first step is to identify your companys strengthsand areas where it could improve.Assess your companys own maturity across AI capabilities by answering these key questions:24Do our governance processes prioritize data security and enable trustworthy AI?FIGURE 8 Assessing AI maturityQuestions to
114、 ask on the road to world-class ROITalent and skillsCulture and adoptionAI engineering and operationsVision and strategyAI operating modelData and technologyTrustHow is AI being used to differentiate my company from its competitors?Is AI embedded in our internal systems and operational processes?How
115、 are machine learning and data models being tested,deployed,and maintained?How are we attracting talent with data and AI skills and developing expertise across the enterprise?What level of change management support is available to bolster AI adoption?Do our governance processes prioritize data secur
116、ity and enable trustworthy AI?25With these insights in hand,heres how you can evolve your organizations capabilities across six key capabilitieskeeping trust at the core:#1 Vision and strategyIdentify where AI can boost competitiveness,innovation,and performanceand prioritize accordingly.Make strate
117、gy AIs North Star.Re-examine how projects are run,recalibrate the backlog,and reduce complexity to align AI projects with business goals.Define a vision for strategic,responsible AI that can guide teams across the enterprise.Be innovative,but dont get distracted by the latest innovation.Identify val
118、ue-rich,industry-leading use cases.Examine which AI domains(for example,vision,language,prediction)and disciplines(for example,deep learning,generative AI,other machine learning techniques)would be best suited to help the organization solve its business problems.30 Explore where foundation models co
119、uld accelerate progress and add value,keeping ethical principles top of mind.Demystify AI maturity.Chart a course for building the foundational AI capabilities that underpin success across multiple projects,initiatives,and programsand integrate this roadmap into annual operating plans and long-term
120、strategy.31 Set the bar for success.Develop a performance management framework that helps teams baseline,measure,and improve performance across a range of metrics,such as volume,value,speed,and efficiency.Create checks and balances.Create a governance model that lets sponsors easily monitor how AI i
121、s being used across the enterpriseand the supply chain.Be prepared to terminate projects that arent delivering the intended value,supporting strategic goals,or following ethical guidelines.#2 Operating model Embed an AI operating model into the fabric and culture of the organization.Consistently pri
122、oritize AI projects that align to strategyand principles.Introduce a formal scoring framework that considers responsible use,business value,speed to value,risk,urgency,and resource availability.Elevate projects and workstreams that are expected to deliver the highest ROI when viewed holistically.Exp
123、eriment oftenwithin guardrails.Define a clear process for developing MVPs quickly to achieve business-defined objectives.Include ethics and bias reviews of AI models in operating procedures to check that AI is being used properly.Outline policies and explicit operational guidelines for where and how
124、 generative AI can be used.Feed the virtuous cycle.Establish a mechanism for delivering insights to the businessand manage data responsibly to build trust in the insights that emerge.Set and evaluate performance benchmarks and diagnostic metrics on a regular basis.Enable independence and innovation.
125、Co-create a governance model for AI-enabled transformation that lets business units and functions drive their own agendas while maintaining an enterprise-wide approach.Avoid business interruption.Introduce a dedicated support team for AI models with monitoring,configuring,and upgrading responsibilit
126、ies.Craft a clear escalation plan to call in executive support when blockers impede engagement across business units.26#3 Engineering and operations Deploy AI solutions that are flexible,user-friendly,and scalable.Dont reinvent the wheel.Consider where“off-the-shelf”models can provide a more cost-ef
127、fective starting point for AI projects.Document changes for faster troubleshooting.Implement robust versioning in the current development lifecycle for AI solutions,tracking source code,configuration,and input data appropriately.Make adoption foolproof.Deploy a framework that automates implementatio
128、n,focusing on efficiency,robustness,transparency,and scalability.Group meaningful components together and deploy them as microservices.Stay vigilant.Testing should cover functional,performance,and load tests,as well as checking incoming data for changes,such as distribution shifts.AI outputs should
129、also be assessed for quality and appropriateness.Solve problems as they appear.Put tools and metrics in place for monitoring and bias mitigation.Make sure teams have the technology needed to do an efficient root cause analysis when AI applications experience a performance drop or security event.#4 D
130、ata and technology Build core capabilities that simplify,automate,control,and secure access to data.Know AIs role.Define AI use cases within a companys principles,broader technical guidelines,and architecture.Be upfront about when it is not appropriate to use AI.Support AI with strong IA.Build matur
131、e data management processes that can be commonly shared across the information ecosystem and are understood by both technical and business unit experts.Put data in context.Make sure representative data sets are curated and categorized responsibly(and consensually),so that teams can derive meaningful
132、,trustworthy insights.Include data integration capabilities within architectural guidelines.Cultivate trustworthy data.Capture data accurately,cleanse it throughout its lifecycle,and make it available to AI teams and the business in a timely manner.Connect data scientists and domain experts to help
133、teams select the right data set for each use case.Increase data literacy.Provide training that helps employees think critically about data and deliver better insights to the business.Make more room at the table for those with complementary skill sets to co-develop AI.This will also help teams determ
134、ine when data can be trusted to train AIand when better inputs are needed.27#5 Talent and skills Deploy an enterprise-wide approach to develop AI ethics,skills,and talent.Build AI muscle.Define clear roles,responsibilities,and expectations,and develop growth plans that help employees strengthen key
135、skills.Create opportunities for continuous learning.Encourage greater familiarity with AI and deliver targeted training to help employees find valuable ways to use it.Highlight the limitations and opportunities associated with new tools,such as generative AI.Collect lessons learned at the end of eac
136、h project and catalog them in a central repository to encourage knowledge sharing.Let technology do the heavy lifting.Empower teams with“no-/low-code”modeling toolspaired with robust governance that prevents improper data usage.”32 Keep ethics top of mind.Educate employees on the principles of AI et
137、hics and be transparent about when and how they should escalate concerns.Create clear organizational structures that empower people to challenge AI use cases and raise ethical questionsregardless of whos running the project.Foster a data-focused mindset.Keep data management at the forefront.Make sur
138、e all teams understand how critical responsible data is to any AI implementationand require them to consistently measure and report on the success of their AI projects.#6 Culture and adoptionDevelop a human-centered approach to AIwith dynamic,open feedback loops across the ecosystem.Dont let AI be a
139、n afterthought.Infuse AI into the DNA of the enterprise.Incentivize teams to look for ways to create business value with AI.Bring designers into the process early on to deliver human-centric solutions.Enlist champions of change.Engage employees by publicly recognizing individual contributions and te
140、am successes.Publicly reward the AI vanguard in your organization to inspire others to follow.Empower teams to reimagine workflows.Ignite change at the grassroots level by developing a culture of continuous collaboration and a“use-case first”mindset.Develop a communication strategy to keep people up
141、-to-speed on whats happening with AI across the company.Double down on change management.Use standardized and documented methodologiesincluding value realization or benefits trackingto successfully transform the enterprise with AI.Work from the same playbook.Maintain a centralized dashboard with the
142、 KPIs for multiple AI use cases to inform comparisons and benchmarking.Showcase teams that embody the principles of trustworthy AI.28Trust at the coreEmbed ethics throughout the AI lifecyclefrom design to deployment to dynamic feedback.Listen up.Establish a governance approach for ethical AI impleme
143、ntation that incorporates feedback from executives,employees,customers,and regulators.Take a stand.Establish organizational structures,policies,processes,and monitoring that address privacy,robustness,fairness,explainability,transparency,and other relevant principles.Dont work in a vacuum.Consider t
144、he enterprise and broader ecosystemespecially as organizations adopt generative AI.Know when youre in the danger zone.Set an AI and data risk profile and threshold leveland clearly outline what to do when risks become reality.Help teams understand the limitations and risks of foundational models.Dis
145、cuss how best to use themand when not to.Operationalize integrity.Integrate ethics into the AI lifecycle and create accountability every step of the way.29Maryam AshooriHead of Product,watsonx.ai,IBM SoftwareMaryam.A Brian GoehringGlobal Research Lead,AI,IBM Institute for Business VMaryam Ashoori is
146、 a technology leader with more than 15 years of experience developing data-driven technologies that drive demand and delight customers.Dr.Ashoori is Director of Product Management at IBM,where she manages the portfolio of emerging technologies in data and AI.She has built high-performing research,de
147、sign,product,and development teams that have delivered enterprise and consumer products in data and AI,automation,lifecycle management,IoT,cloud services,and mobile.Timothy HumphreyChief Analytics Officer,IBMTimothy Humphrey is IBMs Chief Analytics Officer and the companys Senior State Executive in
148、North Carolina.Tim has more than 25 years of global experience with IBM and Lenovo.He has held various roles spanning hardware,software,battery technology,supply chain,acquisitions,data,and AI.He has earned numerous patents as well as management,innovation,and excellence awards for his contributions
149、 to the computing industry.Tim engages in several nonprofit fundraising activities,special events,and volunteer efforts,and serves as a board member for many local nonprofit organizations.Brian Goehring is an Associate Partner in the IBM Institute for Business Value,where he leads the AI business re
150、search agenda,collaborating with academics,clients,and other experts to develop data-driven thought leadership.He brings over 20 years of experience in strategy consulting with senior-level clients across most industries and business functions.He received an AB in Philosophy from Princeton Universit
151、y with certificates in Cognitive Studies and German.About the authors2930Mahmoud NaghshinehVice President of AI Impact and Partnerships,IBM R Dr.Mahmoud Naghshineh is responsible for accelerating AI technologies and scaling AI business impact with IBM clients and partners.His past work has spanned b
152、oth technology and business.Previously,he was the CTO of IBM Global Industries,and prior to that,the General Manager of IBM Strategic Partnerships with major technology companies.Naghshineh has extensive experience in information technologies with deep expertise in networking,mobile,and edge computi
153、ng,as well as analytics and AI.He has played a key role in commercializing several technologies and offerings in these areas and has held senior executive responsibilities in R&D as well as in consulting.Naghshineh is also a Fellow of Institute of Electrical and Electronics Engineers(IEEE)and a memb
154、er of Columbia Universitys Data&Society Council.Cathy Rodenbeck ReeseAmericas Data&Technology Transformation Leader and Senior Partner,IBM CCathy Reese leads the Americas team for the Data&Technology Transformation(D&TT)service line of IBM Consulting.The D&TT service line is focused on helping clien
155、ts gain greater precision and predictability out of business decisions by applying cognition to fundamentally transform their business.Cathy specializes in customer transformations that leverage enterprise and external data,agile practices,AI,and advanced analytics.She builds upon deep data and tech
156、nology expertise,paired with a user experience and design background,to achieve adoption and drive business outcomes with clients.About the authorsContributorsKristin Biron,Phaedra Boinidiris,Beatriz Etchegaray Garcia,Rachna Handa,and Tegan Jones3031Study approach and methodology In partnership with
157、 Oxford Economics,the IBM Institute for Business Value surveyed 2,500 executives in 22 countries across North America,Latin America,Europe,Middle East and Africa,and Asia(including China and India)from May through July 2021.The survey scope included 34 business and technology rolesprimarily executiv
158、es but also IT and AI professionalsfrom 17 industries.We identified a group of“data wealth outperformers”who were in the top quartile of respondents based upon their data wealth score.The data wealth score was calculated as a cumulative self-assessment score across the data dimensions.These could ha
159、ve taken a value of 1 to 5,wherein 1 is equivalent to significantly below peers and 5 is significantly above peers.The data dimensions were:managing volumes of data;quantity of data sources;data value/quality of data and data sources;monetization of data and data sources;trust in AI,data,and data so
160、urces by internal stakeholders;trust in AI,data,and data sources by external stakeholders;and agility in working across business units/functions and IT.We also categorized respondents based upon their response to taking a strategic view of AI in their organization,as well as their self-assessment of
161、 ROI from enterprise-wide AI initiatives,and analyzed their responses to the survey,particularly around the six capabilities.Related ReportsProven concepts for scaling AI“Proven concepts for scaling AI:From experimen-tation to engineering discipline.”IBM Institute for Business Value.September 2020.h
162、ttps:/ibm.co/scaling-ai Dealing with the AI data dilemma“Dealing with the AI data dilemma:The right approach to integration,governance,and tools.”IBM Institute for Business Value.June 2021.https:/ibm.co/ai-data-integration Rethinking your approach to AI“Rethinking your approach to AI:How to ground a
163、rtificial intelligence in business strategy.”IBM Institute for Business Value.September 2021.https:/ibm.co/ai-business-strategyAI ethics in action“AI ethics in action:An enterprise guide to progressing trustworthy AI.”IBM Institute for Business Value.May 2022.https:/ibm.co/ai-ethics-actionHow to cre
164、ate business value with AI“How to create business value with AI:12 stories from the field.”IBM Institute for Business Value.August 2022.https:/ibm.co/ai-examples 32IBM Institute for Business ValueFor two decades,the IBM Institute for Business Value has served as the thought leadership think tank for
165、 IBM.What inspires us is producing research-backed,technology-informed strategic insights that help leaders make smarter business decisions.From our unique position at the intersection of business,technology,and society,we survey,interview,and engage with thousands of executives,consumers,and expert
166、s each year,synthesizing their perspectives into credible,inspiring,and actionable insights.To stay connected and informed,sign up to receive IBVs email newsletter at can also follow IBMIBV on Twitter or find us on LinkedIn at https:/ibm.co/ibv-linkedin.The right partner for a changing worldAt IBM,w
167、e collaborate with our clients,bringing together business insight,advanced research,and technology to give them a distinct advantage in todays rapidly changing environment.About Research InsightsResearch Insights are fact-based strategic insights for business executives on critical public-and privat
168、e-sector issues.They are based on findings from analysis of our own primary research studies.For more information,contact the IBM Institute for Business Value at .33Notes and sources1 Tobin,Michael,Redd Brown,Subrat Patanik,and Bloomberg.“A.I.is the star of earnings calls as mentions skyrocket 77%wi
169、th companies saying theyll use for everything from medicine to cyberse-curity.”Fortune.March 1,2023.https:/ “Worldwide Spending on AI-Centric Systems Forecast to Reach$154 Billion in 2023,According to IDC.”IDC.March 7,2023.https:/ 3 “The business value of AI:Peak performance during the pandemic.”IBM
170、 Institute for Business Value.November 2020.https:/ibm.co/ai-value-pandemic 4 “Driving ROI through AI:AI best practices,investment plans,and performance metrics of 1,200 firms.”ESI Thoughtlab.September 2020.https:/ Rossi,and Beth Rudden.“AI ethics in action:An enterprise guide to progressing trustwo
171、rthy AI.”IBM Institute for Business Value.May 2022.Unpublished data.https:/ibm.co/ai-ethics-action5 IBV generative AI pulse survey of 369 C-suite executives in Australia,Germany,India,Singapore,United Kingdom,and United States,conducted April-May 2023.6 Dickson,Ben.“AIs J-curve and upcoming produc-t
172、ivity boom.”TechTalks.January 31,2022.https:/ “Pursuing transformation like digital natives:Lessons for enterprises from tech leaders who have lived it.”IBM Institute for Business Value.January 2023.https:/ibm.co/digital-native-transformation 8 “Spring 2022 Snapshot Of The S&P 500s Market Cap.”Seeki
173、ng Alpha.April 12,2022.https:/ 9 Torres,Roberto.“How Walmart enhances its inventory,supply chain through AI.”CIO Dive.December 13,2022.https:/ Goehring,Brian,Francesca Rossi,and Beth Rudden.“AI ethics in action:An enterprise guide to progressing trustworthy AI.”IBM Institute for Business Value.May 2
174、022.https:/ibm.co/ai-ethics-action11 Iqbal,Mansoor.“Lyft Revenue and Usage Statistics(2023).”Business of Apps.February 20,2023.https:/ Mathur,Mihir and Hakan Baba.“Powering Millions of Real-Time Decisions with LyftLearn Serving.”Medium.January 30,2023.https:/ 13 Iqbal,Mansoor.“Lyft Revenue and Usage
175、 Statistics(2023).”Business of Apps.February 20,2023.https:/ “2023 Chief Data Officer Study:Turning data into value.”IBM Institute for Business Value.April 2023.https:/ibm.co/c-suite-study-cdo 15 Ibid.16 “What are foundation models?”IBM Research.May 9,2022.https:/ 17 Ibid.3418 “How to create busines
176、s value with AI:12 stories from the field.”IBM Institute for Business Value.August 2022.https:/ibm.co/ai-examples 19 Lipp,Anthony,Anthony Marshall,and Jacob Dencik.“Open the door to open innovation:Realizing the value of ecosystem collaboration.”IBM Institute for Business Value.December 2021.https:/
177、ibm.co/open-innovation 20 In this context,“peers”refers to organizations that scored low on both data wealth and the degree to which it has an embedded AI operating model.21 Rudden,Beth,Wouter Oosterbosch,and Eva-Marie Muller-Stuler.“Proven concepts for scaling AI:From experimentation to engineering
178、 discipline.”IBM Institute for Business Value.September 2020.https:/ibm.co/scaling-ai 22 “How to create business value with AI:12 stories from the field.”IBM Institute for Business Value.August 2022.https:/ibm.co/ai-examples23 “Artificial Intelligence:Our fundamental properties for trustworthy AI.”I
179、BM.Accessed April 11,2023.https:/ 24 Internal IBM analysis.25 Goehring,Brian,Francesca Rossi,and Beth Rudden.“AI ethics in action:An enterprise guide to progressing trustworthy AI.”IBM Institute for Business Value.May 2022.https:/ibm.co/ai-ethics-action26 Ibid.27 Ibid.28 Toews,Rob.“The Next Generati
180、on of Large Language Models.”Forbes.February 7,2023.https:/ Westfall,Chris.“Elon Musk And Other Tech Leaders Call For Slowdown On AI Development.”Forbes.March 29,2023.https:/ 30 “How to create business value with AI:12 stories from the field.”IBM Institute for Business Value.August 2022.https:/ibm.c
181、o/ai-examples31 Prabhakar,Aparna,Veena Mosur,and David Cox.“Rethinking your approach to AI:How to ground artificial intelligence in business strategy.”IBM Institute for Business Value.September 2021.https:/ibm.co/ai-business-strategy 32 Haydock,Michael,Steven Eliuk,and Susara van den Heever.“Dealing
182、 with the AI data dilemma:The right approach to integration,governance,and tools.”IBM Institute for Business Value.June 2021.https:/ibm.co/ai-data-integration 35 Copyright IBM Corporation 2023IBM Corporation New Orchard Road Armonk,NY 10504Produced in the United States of America|May 2023IBM,the IBM
183、 logo, and Watson are trademarks of International Business Machines Corp.,registered in many jurisdictions worldwide.Other product and service names might be trademarks of IBM or other companies.A current list of IBM trademarks is available on the web at“Copyright and trademark information”at: docum
184、ent is current as of the initial date of publication and may be changed by IBM at any time.Not all offerings are available in every country in which IBM operates.THE INFORMATION IN THIS DOCUMENT IS PROVIDED“AS IS”WITHOUT ANY WARRANTY,EXPRESS OR IMPLIED,INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABI
185、LITY,FITNESS FOR A PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NON-INFRINGEMENT.IBM products are warranted according to the terms and conditions of the agreements under which they are provided.This report is intended for general guidance only.It is not intended to be a substitute for detaile
186、d research or the exercise of professional judgment.IBM shall not be responsible for any loss whatsoever sustained by any organization or person who relies on this publication.The data used in this report may be derived from third-party sources and IBM does not independently verify,validate or audit such data.The results from the use of such data are provided on an“as is”basis and IBM makes no representations or warranties,express or implied.DD0RN0B2-USEN-00