“The real question is, when will we draft an artificial intelligence bill of rights? What will that consist of? And who will get to decide that?” — Gray Scott (futurist, techno-philosopher and amongst the world's leading experts in emerging technologies)

In recent times, much has been written about and debated upon by industry and technology experts vis-à-vis artificial intelligence (AI). Most experts agree that the potential benefits for firms to be had from AI is immense. These benefits span productivity gains, improved efficiency and effectiveness, enhanced customer experience and service, reduced cost, increased revenue, and more. Unsurprisingly then, AI solutions have been gaining traction across the globe and the industry sectors. As per a report by orbisresearch.com, the global AI market is expected to grow at a CAGR of over 35% between 2015 and 2022. 

Artificial Intelligence: A brief introduction…

AI enabled systems emulate human cognitive ability – such as reasoning, problem solving, learning and planning. These “intelligent” systems can sift through mountains of data; to aggregate, blend, and unearth hidden relationships and patterns. AI comprises numerous branches - such as machine learning, deep learning, natural language processing (NLP), visual recognition, artificial neural networks, affective computing, concept mining, fuzzy logic, and semantic technology. Machine learning (ML) makes use of sophisticated algorithms that enable systems to “think” - the algorithms don’t require to be expressly programmed for their evolution and learning. ML can be categorized into supervised & unsupervised learning.

Today’s powerful AI/ML enabled self-learning systems are contextually aware, possess robust predictive capabilities, can make “independent” judgment, and can even enable intelligent process and cognitive automation. These systems – backed by their high-speed data processing and analysis power – possess probabilistic data matching, linguistic analysis, and intelligent ontology based information extraction, classification & scoring capabilities.

AI/ML adoption in financial services has seen significant uptake… 

In recent years, like other industry verticals, AI has been gaining significant traction within the financial services industry. Today, AI/ML based solutions are being leveraged by financial institutions (FIs) for a variety of use cases – such as customer segmentation for improved marketing, cross- and up-selling, campaign management, client-facing chatbots, creditworthiness evaluation & credit score prediction, augmented products recommendations & personalized financial advice, investment & portfolio management, algorithmic trading, trade strategizing & execution, dynamic portfolio rebalancing, capital optimization, and more.

Many FIs – such as Goldman Sachs, JPMorgan Chase, Bank of America, Wells Fargo, HSBC and OCBC to name a few – have been investing in and are successfully implementing AI/ML based solutions in their business functions.

There is, however, scope for more AI/ML adoption in the compliance management space...

While FIs has been adopting AI/ML solutions in many aspects of their businesses, compliance management is an area where its adoption is still in nascent stages. However, there is significant potential for AI/ML adoption in this space – both by the regulators and the FIs.

Inarguably, compliance management related costs for institutions have been rising significantly. As per the Thomson Reuters 9th annual Cost of Compliance survey (conducted in Q1 2018) of compliance & risk practitioners from ~800 FIs across the globe, 66% of FIs expect the cost of senior compliance staff to increase. According to LexisNexis Risk Solutions’ “Future Financial Crime Risks 2017” survey of over 170 senior financial crime compliance management pros from banks in UK, 63% of the survey respondents saw a rise in financial crime compliance management costs over the past two years.

In this article, I would like to share some of the key AI/ML use cases for regulators in the compliance management space. In another article in future, I will share compliance management related AI/ML use cases for FIs.

AI/ML use cases for the regulators in the compliance management space…

Since the 2008 global financial crisis, the responsibilities of regulators and supervisors have increased substantially. Post the crisis, in addition to putting in place more robust regulations and safeguards, regulators and supervisors have also been working overtime to ensure transparency and proper conduct by FIs, enforce market discipline, protect the interests of customers, maintain financial stability, and minimize macro-level prudential and systemic risks.

However, unfortunately owing to their prevalent budgetary, human resource and other constraints, doing all these effectively and efficiently has proved to be quite a challenge for most regulators. To alleviate the challenges, regulators therefore can consider leveraging AI/ML based solutions, where appropriate. Refer below few AI/ML use cases for the regulators in the compliance management space. 

1) Regulatory compliance assurance: Regulators can, where appropriate, consider converting certain aspects of their rules – for example, regulatory reporting rules - into unambiguous machine-executable format, and which would be interpreted by the regulated FIs’ systems directly. Along with this, regulators could leverage AI/ML based solutions for monitoring the concerned FIs’ compliance; FIs in violation of particular regulatory mandates would be duly flagged. Solution would also automatically analyze the regulatory filings of FIs to ascertain incorrect/inconsistent submissions. 

As another example, consider the EU General Data Protection Regulation (GDPR) that became effective on 25th May 2018. EU GDPR require firms to provide comprehensive and clear explanation to users on the data that is collected from them, how the firm uses this data, and who all it shares this data with. Firms are also required to gain explicit consent from users for retaining and processing this data.  

In the above case, AI/ML based solution (having NLP capabilities) could automatically examine the concerned FIs’ privacy policies, and compare their policy documents with the EU GDPR policy clauses – to automatically ascertain, for example, whether in the policy document:

-        Firm has identified 3rd parties with which it shares the users’ personal data

-        Language is confusing or vague

-        There is a clause of implicit users’ consent (as opposed to EU GDPR’s mandate of explicit consent).

The solution could thus make it much easier for the regulator to monitor data privacy related compliance of large number of firms in scope.  

UK’s Financial Conduct Authority (FCA), with help from Corlytics (a global leader in regulatory risk intelligence) has developed “intelligent” regulatory handbook. Through this, FCA has ensured that its handbook becomes accessible to all concerned parties. The FCA handbook – which is used daily by thousands of regulated FIs and their advisers – consists of regulatory obligations and provisions’ manuals. Earlier, it was in the form of online legal text book. The new solution has leveraged AI and intelligence search technology to create a fully searchable and accessible handbook. With the new solution in place, existing content in the handbook can be tagged and machine read. Users can – using even non-exact search criteria – get accurate results and quickly find the required regulatory information.

Division of Economic and Risk Analysis (DERA) in US’ Securities and Exchange Commission (SEC) has leveraged ML to unearth patterns in the SEC regulatory filings of investment advisers. For this, vast amount of unstructured and structured data from the filings are ingested into a Hadoop computational cluster. DERA staff then utilize unsupervised ML algorithms to identify outlier/unique reporting behaviors – using both tonality analysis and topic modeling. Further, the output from this stage is combined with the outcomes of past examination and fed into another ML algorithm that predicts the presence of idiosyncratic risks (if any) at the level of each investment advisor.

2) Prudential supervision & monetary policy definition: AI/ML based solutions can assist regulators with their prudential supervision responsibilities – such as those related to banks’ capital and liquidity risk requirements. For example, solution can speedily analyze numerous indicators scattered across banks’ several activities and identify risky banks needing regulator’s attention. The solution could also help in identifying specific key risk indicators for a large and complex bank.

Central banks too can leverage AI/ML solutions to support in swift monetary policy decision making and execution – such as those related to interest rate setting. For instance, the AI/ML solution can leverage traditional data collected at the central bank, real-time feed of transaction level data from all banks governed by the central bank, and the rich data from government databases on various economic aspects (such as production levels, unemployment rates, consumer behavioral data etc.) In addition, solution would crawl the web for sentiment analysis, and utilize web scraping to gather relevant data. For example, through web scraping of real estate buy/sell websites, data of real-estate prices can be gathered and fed to the AI/ML algorithm.

By intelligently parsing through the above data points, solution would be able to simulate multiple scenarios, run numerous impulse responses, and identify in real-time the optimal interest rate. Using such an approach, solution could also assist central banks in:

-        Creation of econometric models

-        Economic forecasting; predicting economic indicators and trends for swift and timely action

-        Identification and analysis of myriad complex macro-financial linkages

-        Systemic risks identification

Central Bank of Russia has embraced ML solution for its new economic indicator. The new indicator assesses the national economic activity using the big data that is derived from large number of news stories available on online sources, and the Purchasing Managers Index. ML capabilities are then leveraged to derive meaningful insights from this big data. The People’s Bank of China has also said that it would increase the usage of AI, big data, and cloud computing capabilities to enhance its ability to identify, prevent and decrease cross-market and cross-sector  financial risks.

3) Market surveillance: Regulators can leverage AI/ML based solutions for market surveillance and monitoring, and for identification of market misconducts such as insider trading, investment adviser misconduct, and equity market manipulation. Solution can monitor daily and examine holistically millions of transactions and communications - across trading venues, trading firms, products, asset classes, communication channels etc. 

Solution could, for example, scrub messages from chat-rooms to detect suspicious communication at the time of a large trade. Additionally, by analyzing real-time feeds from trading venues, solution can quickly unearth complex malpractices (such as spoofing and layering), and new types of market manipulations and dubious trading patterns. Further, the solution would assist investigators – who otherwise have to spend huge efforts on this task – to quickly cross-reference the trading data with past electronic communications and build strong cases against the players involved in market manipulation (e.g. rigging of foreign exchange benchmarks.) 

The Australian Securities and Investments Commission (ASIC), had launched a pilot program in partnership with a regtech firm to leverage cognitive learning tools on webpages of accountants. The goal was to unearth potential misleading or unlicensed conduct in self-managed superannuation fund (SMSF) activities. In the US, Financial Industry Regulatory Authority (FINRA) has leveraged AI to identify potential instances of cross-asset manipulation that uses options and equities. In this manipulation, a trader holds an options position and tries to move the underlying equity to change its price. 

4) AML & fraud detection: As per United Nations Office on Drugs and Crime,  worldwide, money amounting to 2 - 5% of the global GDP is laundered annually. In absolute terms, this figure range between USD 800 billion and USD 2 trillion per year. As per Annual Fraud Indicator 2017, fraud annually cost UK £190 billion.   Considering their magnitude, it is unsurprising that combatting money laundering and fraud remains a key priority for regulators across the globe.

Alas, for the regulators, investigating suspicious money laundering and fraudulent transactions is extremely time consuming. A key reason for this is the significant number of defensive suspicious activity report (SAR) filed by regulated FIs. In order to avoid regulatory criticism, many FIs file SARs with regulators even for the cases where they know or are fairly confident that the transaction is clean. For regulators, this results in huge number of false positives.  In such scenarios, AI/ML based solution can help regulators sift through voluminous SAR filings and shortlist the truly suspicious transactions that warrant closer scrutiny; thereby significantly reducing the number of false positives.

Also, many times, money laundering does not become apparent when only the concerned individuals are focused upon.  Instead, it becomes evident only after the entire transaction histories across the groups are observed. AI/ML solution can help regulators in unearthing such money laundering cases as well – which otherwise are not directly observable from individual FI’s SAR filings. Towards this, solution would monitor, analyze and correlate the banks’ detailed transactions, client profiles and networks, regulatory lists (e.g. AUSTRAC, OFAC, OSFI), 3rd-party PEP & sanctions lists (e.g. WorldCompliance, WorldCheck, Accuity) and huge amount of other unstructured data from the web (news articles, social media etc.)  It would then uncover associated non-linear relationships amongst myriad attributes and entities, and detect the complicated money laundering patterns and cases.

In Australia, researchers at RMIT University in Melbourne have been helping Australian Transaction Reports and Analysis Centre (AUSTRAC) – Australia’s financial-intelligence agency -  by enabling AI/ML tools for detecting and deterring suspicious activity. The solution can detect unknown money laundering networks more speedily and accurately than ever before; and can efficiently and precisely flag transactions that needs to be further investigated.

Conclusion 

It is nobody’s case that AI/ML is panacea for all of the compliance management challenges of regulators. However, AI/ML solutions can certainly help regulators in many pertinent aspects of their compliance management undertaking. It is therefore my view, that with the passage of time, as the AI/ML technology gain even higher levels of maturity, many more regulators across the globe would become amenable to leveraging this technology.