“Nobody phrases it this way, but I think that artificial intelligence is almost a humanities discipline. It's really an attempt to understand human intelligence and human cognition.” — Sebastian Thrun (innovator, computer scientist, and chairman & co-founder of Udacity)  

In an earlier article, where I had shared insights on artificial intelligence and machine learning (AI/ML) use cases for the regulators, I had stated that while AI/ML solutions are being implemented by financial institutions (FIs) for a variety of use cases across their business functions, compliance management is an area where its adoption is still in a relatively nascent stage.

There is, however, huge potential for AI/ML adoption in the compliance management domain – both by the regulators and the FIs. Through this article, I would like to share some key AI/ML use cases for FIs in this domain.    

Effective compliance management: An ongoing challenge for FIs…

As per a report from Boston Consulting Group, since the 2007-2008 financial crisis and until 2016 end, banks across the globe have paid USD 321 billion in regulatory fines. Given such huge cost of non-compliance, it is understandable that implementing effective compliance management solution has remained a key focus area for most FIs.

Unfortunately, however, many FIs’ compliance management functions continue to lack potency, and truth be told, remain an Achilles heel. FIs’ compliance management challenges have, in fact, only further compounded in recent years – what with the massive rise in volume, frequency and complexity of new and revised regulatory mandates. Today, Basel III, MiFID II, EU GDPR, FRTB, CRR/CRD IV, FATCA, CRS, PSD II, IRRBB, AMLD IV, PRIIPs and BCBS 239 are just few of the many regulatory mandates that a global FI must comply with.

For FIs, wading through thousands of pages of these new/revised regulations and ensuring effective ongoing compliance is a herculean task. Managing this enormous effort is a costly affair - FIs have had to increase their compliance staff size and/or leverage support of third-party firms. As per Thomson Reuters’ “Cost of Compliance 2018” report, 61 percent of financial services firms expected an increase in total compliance budget in 2018. 

Can AI/ML solution help FIs overcome some of their compliance management challenges and substantially reduce the associated costs? I believe, the answer is a resounding yes! Refer below some of the key AI/ML use cases for FIs in this regard.  

AI/ML use cases in compliance management for FIs…

1) Mapping of regulatory changes: AI/ML solution can help FIs automatically identify, analyze, interpret and even implement to an extent the new/revised regulatory mandates. For this, solution would leverage its natural language processing (NLP) and cognitive computing capabilities to proactively, and on an ongoing basis, scan through, evaluate and interpret huge volumes of unstructured regulatory content - that are scattered over hundreds of regulators’ websites and databases. Further, the solution would automatically shortlist applicable regulatory requirements for the FI. [Solution can also share valuable insights such as how the new AML regulatory mandate in Europe differs from that in U.S.]

As the next step, solution would extract metadata and automatically map the new/changed requirements to the FI’s products, services, contracts, processes and functions. Additionally, system can translate these requirement into common machine-executable form and link these to the concerned policies, procedures and systems of the FI’s affected business/compliance functions. The concerned functions would receive automatic XML feeds comprising the new/revised requirements and associated interpretation. Solution’s robotic process automation (RPA) and ML capabilities could be further leveraged to codify the requirements and generate implementation workflows.

Commonwealth Bank of Australia has, in a pilot, leveraged NLP based AI solution to convert 1.5 million paragraphs of regulatory content into actionable compliance obligations. Results showed that the solution could achieve desired goal with up to 95 percent accuracy. The pilot also significantly reduced the time required - using the AI solution, typical six months of manual work could be completed within two weeks.

2) Regulatory compliance assurance: Using AI/ML solution, FIs can ensure effective and ongoing compliance with multitude of relevant regulatory mandates. Solution, for example, would be able to provide FIs deep insights into high-risk regulations, and enable digitized risk-based compliance assurance. Consequently, cognitive biases that are ever present in FIs’ traditional human-judgment based compliance assurance approaches can be effectively eliminated.

The real-time solution would scan through and analyze the FI’s various structured/unstructured data sources (such as transactional databases, business lines data stores, documents, e-mails, chats, telephonic conversations etc.) to evidence and verify compliance. Using ML and deep learning techniques, solution can, for example:

-          Identify and flag compliance data inconsistencies

-          Match and interpolate missing data

-          Monitor and ascertain compliance gaps

-          Generate comprehensive gaps analysis report

-          Provide rectification recommendation

Further, solution can support the FI’s stress testing activities such as modelling, forecasting and scenario analysis. Regulatory reporting aspects such as automated data collection, data quality improvements and workflow digitization would also be effectively supported. For enabling a clear view of cost of non-compliance, solution would automatically link historic data related to regulatory enforcement actions and sanctions to the specific regulatory requirements and the FI’s compliance control parameters.

Citi had selected Ayasdi (an ML solutions firm) to enable robust models for the bank’s revenue & capital reserve forecasting and to help it clear the US Federal Reserve’s CCAR stress tests. As per Ayasdi, the solution shortened Citi’s process duration from nine months to 3 months and significantly reduced the number of staff required for supporting this process.   

3) KYC management. FIs can benefit from AI/ML solution in numerous KYC aspects, such as - identity & background pre-checks for remote KYC, customer onboarding, real-time transaction-based KYC anomaly detection, and KYC workflow automation.

For remote KYC, solution for example, can leverage its image recognition feature to ascertain if the images in identifying documents match one another. Also, solution would automatically scan various sources (such as police registers, social media etc.) to calculate the customer’s KYC risk score. Solution’s sophisticated clustering techniques (such as K-Means, Mean-Shift, Agglomerative Hierarchical, DBSCAN etc.) would help group customers according to their profession, jurisdiction, expected net-worth and other parameters. This would enable high quality KYC risk assessment. For customer onboarding, the self-learning solution would enable dynamic and customized questionnaire that can adapt as per the customer’s responses and according to prior intelligence already gathered from various additional sources. 

For real-time transaction-monitoring based KYC anomaly detection, solution would conduct holistic transactions analysis by automatically deep-diving into the customer’s transaction, transaction history, behavioral profile and other unstructured data sources (such as emails, chats, negative news, social media etc.). Solution would leverage its NLP, ML, generative modeling and sophisticated analysis (e.g. Triple Exponential Smoothing) capabilities to unearth inconsistency. It would also automatically evolve the customers’ transaction profile and behavior archetypes as needed.

Through its RPA, NLP and cognitive computing capability, solution can facilitate robust KYC workflow automation related to:

-          Customer data extraction from documents for KYC verification

-          Transfer of relevant data from documents into customer onboarding systems

-          Capturing of KYC screening results

-          Flagging of customer profiles for missing/mismatched/outdated KYC information

-          KYC profiles updates and data remediation

-          Chatbots for seeking additional KYC documents from customer

-          KYC reports generation 

Standard Bank - Africa’s largest bank - has successfully leveraged WorkFusion’s RPA and AI solution for significantly reducing the time taken for client onboarding and KYC from 20 days to just five minutes.

4) AML/fraud management: For AML, the context-sensitive AI/ML solution would support advanced and adaptive real-time monitoring for high-risk entities – including against the SDN, OFAC and other sanctions lists, and/or related to unstable geographies. Using its ML, NLP, linguistic search, exploratory data analysis (EDA) and probabilistic/deterministic matching capabilities, solution would intelligently analyze, in real-time, customers’ transactions and other contextual data from various sources (such as web login activity, account profile data from CRM, public and proprietary AML databases, news sites, social media, monitoring reports, regulatory alerts, consortium data, financial records etc.). It would then leverage this information to screen for numerous money laundering risk signals, enable true identity-matching, identify complex money laundering typologies and patterns, and detect early warning signals.

Solution would empower sophisticated link analysis and help reveal highly opaque, complex, remote and multilayer relationship linkages between entities (individuals, firms, business partners, suppliers etc.) and transactions. It would also help effectively identify the ultimate beneficial owners (UBOs) and politically exposed persons (PEPs). Further, by leveraging its powerful cross-border transaction analysis capabilities, solution can help uncover cryptic SWIFT messages, erroneous invoice numbers, duplicate/linked addresses, originator to beneficiary information (OBI) messages, and many other forms of concealed money laundering clues. Additionally, solution would enable sophisticated risk scoring to help prioritize the investigation queues for suspicious activity reporting (SARs).

For fraud management, solution would automatically analyze, in real-time, customers’ transactions against their transactions history, the established spending profile, and myriad other parameters. Solution would then generate a sophisticated fraud score that depicts the probability of the transaction being fraudulent ­­– this fraud score would be generated using an ensemble classifier comprising hundreds of self-calibrating models. Further, if the fraud score is above a certain threshold, the transaction would get automatically rejected. The self-learning solution would quickly learn and adapt to the new fraud methods to generate sophisticated insights. For example, solution can intelligently search the public digital footprint and databases to locate the whereabouts of an absconding fraudster and the FIs that may be assisting him to move the funds internationally.

Danske Bank has engaged Think Big Analytics (a Teradata firm) to implement AI-driven fraud detection platform. The solution’s ML capability analyzes tens-of-thousands of potential features, and scores millions of online transactions in real-time, to enable actionable insights on fraudulent transactions. The solution has helped to significantly reduce the number of false-positives. HSBC is leveraging AI solution to combat money laundering, fraud and terrorist financing. The bank is planning to integrate AI solution of Quantexa (a UK-based start-up) to screen vast amounts of customer and transactions data against the publicly available data.  

5) Trade & market surveillance: AI/ML solutions can help FIs to effectively combat rogue trading, insider trading, benchmark rigging and other forms of trade and market manipulation. For this, solution would leverage its ML, deep learning and NLP capabilities to analyze, in real-time, the entire trading portfolios. It would automatically compare each trading transaction against the historic transactions and pattern, and analyze information from other external/internal sources (such as trade floor communications, public news sites, stock prices movement, trader behavioral profile, trader’s calendar items/e-mails/ chat logs/phone transcripts/social media activity etc.). Consequently, solution would be able to reveal complex trade/market manipulation, hidden links and patterns, and even an intent of a trader to commit market abuse in the future.

Hong Kong Exchanges and Clearing Limited (HKEX) has successfully deployed, across its equity market, Nasdaq SMARTS Market Surveillance’s latest ML solution & participant relationship discovery technology. Japan Exchange Regulation (JPX-R) and Tokyo Stock Exchange, Inc. (TSE) are leveraging AI solution for market surveillance operations. The solution is helping surveillance personnel accomplish preliminary investigations more quickly. It has acquired knowledge on earlier operations conducted by surveillance personnel vis-a-vis trading irregularities evaluation for preliminary investigations. AI startup Neurensic has launched a tool which creates “integrity score” for traders depending upon how their trading patterns match with the patterns that are deemed suspect by regulators.

6) Rogue employee detection: AI/ML solution can continuously monitor in real-time, FI’s entire employee communications (emails, chat logs, phone recordings etc.), their transactions, and behavioral profile patterns, and conduct deep contextual analysis. By doing this, solution can proactively unearth unusual activities of rogue employees and their involvement in perpetrating fraud, money laundering or other illegal activities. For example, solution can reveal the fake accounts opened by an employee by linking many of the recently opened accounts to the same IP address. Solution can also uncover entire network of staff collusion.  

7) Internal audit compliance: By leveraging AI/ML solution, FIs would be able to achieve significant improvements in the quality and effectiveness of their internal audits processes. For example, solution can help FIs move away from their traditional random and backward-sampling audit approaches to an automated, real-time and comprehensive audit framework. For instance, unlike their existing approaches where, say, only 40 samples get manually audited, the AI/ML solution would be able to examine, in real-time, millions of transactions and activities to timely reveal the huge number of audit exceptions. 

Using its sophisticated ML, NLP and semantic analysis techniques, solution would automatically review FI’s internal control logs/documents, financial statements, contracts & agreements, and communications (emails, chats) to reveal complex and hidden audit issues - related to insider threat, AML, fraud, improper payments, cybersecurity, unfavorable contractual agreement terms etc. Further, solution’s RPA capability would help to optimally automate the audit team’s data extraction and interrogation related activities.

In a World Economic Forum survey, 75 percent of the surveyed executives had opined that by 2025, AI solution would be able to perform 30 percent of corporate audits. Deutsche Bank has leveraged AI solution for its audit function. The solution can sift through huge volume of video and voice recordings, and automatically search for terms monitored by the auditors. KPMG has partnered with IBM’s Watson - its auditors have been training Watson to grade bank loans. The solution, which can scan through 800 million pages per second, has helped digitize the credit evaluation activity. Solution can read through thousands of loan documents, extract relevant information, evaluate the grading of each loan, and inform auditors of any discrepancy or potential risks.  

8) Tax & accounting compliance: AI/ML solution can, for example, help a global FI to automatically categorize the taxable income into appropriate country-specific tax buckets, detect tax computation errors, and propose beneficial tax strategies. Solution can enable dynamic dashboards for sophisticated scenario analysis, tax forecasting and reporting. Moving beyond the simplistic modeling techniques (such as linear regression or linear interpolations), that is currently adopted by many FIs, solution would facilitate sophisticated modeling techniques such as polynomial regression or lasso regression.  

Solution would also help automate high number of manual accounting tasks such as data aggregation, reconciliations and initial analysis, and enhance the FI’s accounting transparency and accountability. Resultantly, FIs can more effectively comply with the new accounting standards such as IFRS 16, US-GAAP ASC 842 etc.

9) Compliance staff tools. AI/ML solution can enable additional support tools for the FIs’ compliance management teams - such as: a) virtual assistants and chatbots for queries resolution, b) dynamic dashboards and data visualization tools (that use graph analytics, dynamic Bayesian network etc.), c) false-positives suppression tools (that use clustering, dimensionality reduction, topological data analysis etc.), d) case management tools (that automatically capture relevant case information and populate the results in templated investigation report), e) alerting and alert investigation tools (that make annotated data automatically available to the investigators; self-learns past investigation patterns to support, where feasible, automated future investigations; leverages deep learning to provide intelligent resolution recommendation for the new cases).

Conclusion

Owing to its huge potential to provide intuitive, intelligent and contextual support, AI/ML solution can help FIs to significantly overcome their compliance management challenges, enhance their compliance process effectiveness, and substantially reduce the overall compliance costs. Further, predictably, as this technology gains optimal maturity in the future, the AI/ML solutions would be able to provide even more comprehensive support to FIs – across their compliance management processes and lifecycle stages.