1、InMoment White PaperInMoment White PaperA Better Approach to AI for Regulatory ComplianceCopyright 2022 InMoment,Inc.All rights reserved.Eric Smuda,Principal,CX Strategy&EnablementPAGE 02 A Better Approach to AI for Regulatory ComplianceCompanies with regulatory compliance burdens are flocking to ar

2、tificial intelligence(AI)for time-savings and cost reductions.The money involved is staggering:In 2021,Healthcare AI funding reached$8.5 billion across 366 deals at an average of$30 million each.1 The reasons for these investments?AI can deliver the global healthcare systems and services up to$906 b

3、illion of value annually,2 and AI could deliver the global banking industry up to$1 trillion additional value annually.3 And PwC predicts that AI could contribute$15.7 trillion4 to the global economy.But these big numbers dont eliminate the fact that every month brings new articles and think-pieces

4、about bursting the AI hype bubble.So,before you pay too much attention to those types of articles,let us tell you how our thoughts and experiences warrant a much better approach to AI for regulatory compliance.PAGE 03 AI for Compliance Requires More Than AINATURAL LANGUAGE PROCESSINGSEMI-STRUCTURED

5、DATA PARSINGMACHINE LEARNING AND AIThe truth is that“artificial intelligence”is just a tool.It doesnt really“do”anything on its own.What matters is how you combine AI with other technologies and the experts that know your business.In a field like regulatory compliance,for example,the challenges invo

6、lved can vary wildly by industry and country.So,in order to address these problems,you need a combination of three technologies and human expertise:The complex structure of regulatory and legal documents means that each of these technologies will fall short if its not supported by the other two and

7、many times with a helpful human eye.But most off-the-shelf analytics tools and vendors use only one or two of the three,leaving valuable data behind or overlooking critical context.InMoment,on the other hand,has developed all three tools into a comprehensive platform and has a team of highly trained

8、 experts.We build semi-custom applications that solve specific compliance challenges for our clients.The rest of this paper will explain:1.Why Many“AI for Compliance”Solutions Fall Short2.What You Need to Know to Build“AI for Compliance”3.Three Real-world Examples of Effective AI for ComplianceBACKG

9、ROUNDInMoment acquired Lexalytics,the company that was first-to-market with a commercial sentiment analysis engine.Since then,weve been pushing the limits of natural language processing(NLP),machine learning(ML)and artificial intelligencePAGE 04 Why AI Often Falls Short in Regulatory ComplianceEvery

10、 organization is subject to some regulation.Healthcare providers,pharmaceutical companies,and financial services firms face particularly heavy burdens.But despite massive investment and clear market opportunities,technology solutions for regulatory compliance are proving difficult to develop.Why?In

11、short:Because the complex structure of the documents involved means that traditional data analytics techniques arent able to extract and understand all of the data that compliance professionals rely on.Legal documents are full of both structured elements(such as tables and subheaders)and unstructure

12、d elements(free-standing text sections,descriptive lists,or commentary on information within tables).Two documents may share a format,such as a contract template.But the unstructured text sections are not standardized,and data may differ wildly even within the structured elements.This distribution o

13、f data makes it very difficult to create a single piece of software that can automatically identify and extract everything of value.Most existing systems use one or two of the three technologies:semi-structured data parsing,natural language processing,and machine learning and AI.But the truth is tha

14、t none of these can do the entire job on its own:A commercial semi-structured data parser wont understand the meaning of the text it extracts An open-source NLP engine wont understand how structured elements fit in An AI trained on regulatory docu-ments can choke when presented with new content to a

15、nalyzeTo clarify this last point:AI relies on learning from examples.But regulatory compliance often deals with documents where the format changes but the information within remains consistent.In these cases,an AI system may stumble because the information its asked to process looks very different f

16、rom the data it was trained on.A key factor to the success of AI for compliance is having the right technologies in place to extract data from legal documents that are complex and hard to analyzePAGE 05 The Key to Building an Effective AI for Regulatory ComplianceAs we discussed earlier,large-scale

17、AI systems that promise to replace humans are turning out to be unreliable or downright dangerous.The real opportunity here is to create technology solutions that help humans do their jobs better and faster by alleviating tedious,repetitive tasks.To build an AI solution that actually helps humans wi

18、th regulatory compliance tasks,you need to understand three points:1.Regulatory compliance is very complex,but generally means confirming something:whether a particular document is compliant or that your organization is properly tracking and adhering to shifting regulatory updates2.Legal,medical,and

19、 financial documents have important information contained within both structured elements(tables and lists)and unstructured text3.Traditional data analytics techniques fall short in regulatory compliance because identifying and extracting all of the data from a legal document requires more than AIPA

20、GE 06 InMoment is one of a handful of compa-nies who has both the technology stack and experience required here.Much like tailoring an off-the-rack suit,we use AI to customize our proven NLP platform to meet our clients unique require-ments.We dont believe in replacing humans or“solving”an entire in

21、dustry.Instead,we help our clients do their work better and faster.Weve built AI applications to:Help medical coding teams keep up with monthly Medicaid updates by extracting,structuring,and inserting diagnostic and procedure codes into an EHR Accelerate medical information offices by guiding suppor

22、t agents to the right answers Reduce non-compliance risk by ensuring that financial advisors make required disclosures and appropriate recommendationsThese solutions are interesting as unique use cases.But they also serve as good general demonstrations of how to approach the use of data analytics an

23、d AI in regulatory compliance.Its not AI or humans,its AI for humans to help businesses work better and fasterHealthcare organizations employ teams of professional medical coders(Glassdoor average salary of$37,000)5 to read monthly Medicaid updates and manually enter the new information into EHRs an

24、d other databases.These documents may be dozens of pages long and each states Medicaid system distributes its own unique guidelines.The task is monumental and human coders struggle to keep up.NBC News found medical claim error rates ranging from 7%to 75%.6 Medicaid has improved its 2010 error rate o

25、f 42%,7 but they still lost$31.6 billion to coding errors in 2018,and more than$230 billion over a 6 year period.8 At one point,Medicaid sued UnitedHealth Group,9 alleging hundreds of millions of dollars in over-billing.Fully-automated medical coding solutions face major hurdles and may never be via

26、ble.Instead,this is a perfect example of where technology can help humans,not replace them,by easing workloads and reducing error rates.PAGE 07 THE COMPLEXITIES OF MEDICAL CODING CAN CREATE HURDLES FOR AI SYSTEMS The processes and documents involved in medical coding create big hurdles for any AI sy

27、stem.As demonstrated in Figure 1,relevant diagnostic and procedural codes and updates are contained within a mix of structured elements(tables)and unstructured elements(plain text).Helping Medical Coding Teams Keep Up With Medicaid Updates2.2.25 TOTAL PARENTERAL NUTRITION(TPN)SOLUTIONSJust like a hu

28、man coder,any AI system must identify and extract all of this information and then relate each data point to the others(which treatment code goes with which diagnosis codes).In addition,any automated system must be able to identify footnotes and addendums that clarify special cases and exceptions,an

29、d then connect them to the related diagnostic or procedural codes.EXAMPLE 1PAGE 08 HOW AI HELPS HEALTHCARE ORGANIZATIONS OVERCOME MEDICAL BILLING HURDLES Rather than trying to train one giant AI model to solve all of these problems at once,InMoment selectively utilizes each of our core technologies

30、to build a system thats more reliable and customizable.We never tried to build an AI system to“solve”medical coding.Instead,we brought together the right technology and tools to help human medical coders operate faster and more reliably,saving healthcare organizations time,money,and headaches.Stakeh

31、olders use our solution to keep up with monthly code changes from Medicaid,extracting data and relevant special cases,and then updating corresponding codes in their EHR systems.We use our semi-structured data parser to identify and extract data from structured elements,natural language processing to

32、 analyze the unstructured text,and machine learning to understand EXAMPLE 1where special cases and exemptions relate to specific codes.Our solution for“AI for medical coding”can:Extract procedural and diagnostic codes Relate them to the notes and addendums connected to each Insert this structured da

33、ta into an Electronic Health Records system or other databasePAGE 09 HOW AI HELPS:Building an artificial intelligence system to directly answer patient questions would be a herculean task,even without the threat of lawsuits when it got things wrong.Instead,AI can support medical information offices

34、by helping human agents answer questions faster and more reliably.Medical Information Offices(MIO)are responsible for setting up a system to field and respond to patient questions about drugs,both during testing and after release.Given that peoples lives may be at stake,its imperative that every que

35、stion be answered in a consistent and correct manner each time its asked.To ensure compliance,MIOs draw up detailed Frequently Asked Questions(FAQ)files.Patient questions are routed to a call center,sometimes outsourced,where agents reference the FAQ to answer each question.But the complexity of the

36、se FAQs means that each agent can take many minutes to answer a single question,even when the same question gets asked many times a day.In one of our data sets,80%of drug-related medical information questions relate to just 20%of the FAQ.Building an artificial intelligence to directly answer patient

37、 questions would be a herculean task,even without the threat of lawsuits when it got things wrong.Instead,AI can support medical information offices by helping human agents answer questions faster and more reliably.Our semi-custom application is built on our platform and leverages AI trained with ou

38、r Pharmaceutical Industry Pack.Our system can immediately understand what conditions,drugs,ailments,or issues a customer is calling about and deliver a correct response for the operator,all during the call.For example,a patient might call in asking,“How much of this painkiller should I take?Im a 180

39、-pound guy.”Our system can tear this down to understand that the question relates to dosage of a painkiller for an adult male.Based on this analysis,we can immediately direct the human agent to the proper answer.(Note:this is a made-up example to illustrate how and where the system works.)Helping Me

40、dical Information Offices Answer Patient Questions FasterSaving Thousands of People HoursOur AI helps medical information offices save thousands of people-hours and dollars,all while reducing non-compliance risk.And while this system was developed for a call-center environment,it also shows great pr

41、omise for use at all patient touchpoints,including emails,social media comments,chatbots,or even a physicians desk notes.EXAMPLE 2PAGE 10 Financial services firms must demonstrate that their employees are working in their clients best interests.This imperative can include a number of disclosure requ

42、irements:Conflict of interest Commission structure Cost of credit Own-product recommendations Reducing Risk of Disclosure Non-Compliance for Financial Services FirmsEach disclosure,in turn,may contain a dozen or more sub-components.This adds up to a major burden for the service provider.On average,f

43、inancial firms dedicate 10-15%of their workforce and spend a combined$270 billion on regulatory compliance annually.EXAMPLE 3$270 billion industry spending on compliance 94%increase in compliance costs 10-15%of workforce dedicated to governance,risk management,and compliance Average 10%of revenue sp

44、ent on compliance by 2022 45x increase in regulatory fines and settlements from 2009-14 Up to$19.1 million lost revenue$1.1 million to$20.4 million business disruption Average$3.7 million in productivity lossCOMPLIANCE SPENDING IN FINANCIAL SERVICESTHE COSTS OF NON-COMPLIANCEPAGE 11 EXAMPLE 3To impr

45、ove compliance,many firms build a library of template documents with all of the proper disclosures and legal language.Each advisor or broker then modifies the appropriate template on a client-by-client basis.But an average-size financial services firm may produce thousands of pages of client-facing

46、documents every week.In the process,important disclosures may be accidentally modified or removed entirely.Many firms rely on spot-checks and keyword searches to confirm compliance.But looking for individual Review all documents Clearly see whats inside Fast and efficient Reduced risk of non-complia

47、nce+keywords may return hundreds of irrelevant matches littered through a document.On the other hand,searching for whole phrases may miss where a disclosure has been truncated or deleted.The cost of failure is high.GlobalScape estimates that non-compliance can cost a firm up to$39.22 million in lost

48、 revenue,business disruption,productivity loss and penalties.McKinsey found a 45x increase in regulatory fines and settlements over 5 years.Spot-check a few documents Keyword and phrase search Slow and cumbersome High risk of non-complianceWITHOUT INMOMENT-PAGE 012 Using AI to Improve Existing Proce

49、ssesRather than trying to build an end-to-end,failure-prone“AI for disclosure compliance,”InMoment focuses on improving existing processes.Our system empowers financial auditors to review all of their documents almost simultaneously,instead of spot-checking a handful.This substantially reduces non-c

50、ompliance risk for financial services firms and banks.We combine our semi-structured data parser with text analytics to quickly analyze long financial documents and extract all of the data,wherever its located:legal disclosures,asset allocation tables,statements of advice,client roles,and more.Then,

51、our natural language processing algorithms parse the underlying structure and meaning of the information.This enables us to make complex connections between data points wherever they appear in the document.Finally,we use artificial intelligence to structure this data and prepare it for further analy

52、sis.The nature of this analysis can vary based on your goals and existing technology stack.For example,we can export this data into an Excel spreadsheet.Or,as in Figure 3 below,we can just give you a simple text file.In this example(mocked up for demonstration purposes),the output of our application

53、 shows at a glance whether proper disclosures were made,highlights own-product recommendations,and flags suspected cases of predatory lending techniques for further review.BALANCED INVESTMENT OBJECTIVE:You wish to adopt a limited portfolio to protect you from inflation and tax You have a limited und

54、erstanding of the investment markets and their operation When you make a financial decision,you are more focused on minimizing possible losses You can accept a low level of volatility in the value of your investments You are a low risk taker and prefer low levels of investment riskFig.3|Using InMome

55、nt to flag where a recommendation goes against the clients organizationRECOMMENDED STRATEGY(To address your stated needs,goals,and objectives,we recommend the following strategy to you):Smith,you have informed us that you will receive an inheritance of approximately$550,000.Due to this,your Age Pens

56、ion will cease and therefore you need to replace the loss of pension income with a drawdown from the inheritance funds.Smith,we recommend you invest all($550,000)of available funds from your inheritance through our High-Yield Self-Managed Tech Fund.This allows you to maximize the return on your inve

57、stment by actively investing in self-selected technology companies that are leaders in their fields.PAGE 13 Regulatory compliance is a complicated area that varies wildly from industry to industry and country to country.The nature of the documents involved means that to build an“AI for regulatory co

58、mpliance,”you need more than just AI.In fact,you need a combination of semi-structured data parsing,natural language processing,and machine learning and AI.Despite massive investment,many“AI for compliance”tools fall short because 1.https:/ 2.https:/ useful regulatory compliance solution requires al

59、l three of these technologies,not just one or two and additional human expertise and intelligence.InMoment combines the necessary technology stack with an experienced professional services staff to build semi-custom applications that solve specific regulatory compliance problems in healthcare,biotec

60、hnology,financial services,pharmaceuticals,and more.SummaryNATURAL LANGUAGE PROCESSINGSEMI-STRUCTURED DATA PARSINGMACHINE LEARNING AND AIAbout InMomentImproving experiences is why InMoment exists.Our mission is to help our clients improve experiences at the intersection of valuewhere customer,employ

61、ee,and business needs come together.The heart of what we do is connect our clients with what matters most through a unique combination of data,technology,and human expertise.With our hyper-modern technology platform,decades of domain authority,and global teams of experts,we uniquely deliver a focus

62、on Experience ImprovementTM(XI)to help our clients own the moments that matter.Take a moment and learn more at To demo a product or to contact us call:T L-01859-03-N A|2022 INM O M EN T,I N CNORTH AMERICA+1 385 695 2800 APAC+61(2)8397 8131UK&IRELAND+44 121 296 5245GERMANY+49(0)40 369 833 0Or email us at