Leading Point Financial Markets recently hosted a roundtable event to discuss the feasibility of adopting Artificial Intelligence (AI) for Anti-Financial Crime (AFC) and Customer Lifecycle Management (CLM).
A panel of SMEs and an audience of senior execs and practitioners from 20+ Financial Institutions and FinTechs discussed the opportunities and practicalities of adopting data-driven AI approaches to improve AFC processes including KYC, AML, Payment Screening, Transaction Monitoring, Fraud & Client Risk Management.
“There is no question that AI shows great promise in the long term – it could transform our industry…” Rob Gruppetta, Head of the Financial Crime Department, FCA, Nov 2018
EXECUTIVE SUMMARY
AFC involves processing and analysing vast volume and variety of data; it’s a challenge to make accurate & timely decisions from it.
Industry fines, increasing regulatory requirements, a steep rise in criminal activities, cost pressures and legacy infrastructures is putting firms under intense pressure to up their game in AFC.
90% expressed the volume and quality of data as a top AFC/CLM challenge for 2019.
Applying standards to internal data and client documents were deemed as quick wins to improving process
80% agreed that client risk profiling and the analysis across multiple data sources can be most improved – AI can improve KPI’s on False Positives, Client Risk, Automation & False Negatives.
While the appetite for AI & Machine Learning is increasing but firms need to develop effective risk controls pre-implementation
Often the end to end process is not questioned; firms need to look beyond the point tech, and define the use case for value
Illuminating anecdotes shared on how to make the business case for AI/ Tech. Business, AFC Analysts and Ops have different needs
Firms face a real skills gap in order to move from a traditional AFC approach to an intelligent-data led one. Where are the teachers?
60% of respondents had gone live with AI in at least one business use-case or were looking to transition to an AI-led operating model
AI & Anti-Financial Crime
Whether it is a judgement on the accuracy of a Client’s ID, an assessment of the level of money laundering risk they pose, or a decision on client documentation, AI has the potential to improve accuracy and speed in a variety of areas of the AFC and CLM process.
AI can help improve speed and accuracy of AFC client verification, risk profiling, screening and monitoring with a variety approaches. The two key ways AI can benefit AFC are:
- Process automation – AI can help firms in taking the minimum number of steps and the data required to assemble a complete KYC file, complete due diligence, and to assign a risk rating for a client
- Risk management – AI can help firms better understand and profile clients into micro-segments, enabling more accurate risk assessment, reducing the amount of false positives that firms have to process
Holistic examination of the underlying metadata assembled and challenging AI decisions will be necessary to prevent build up of risk and biases
Mass retraining will be necessary when AI becomes more integral to businesses
KYC / Customer Due Diligence (CDD)
Key challenge: How can anti-money laundering (AML) operations be improved through machine learning?
Firms’ KYC / CDD processes are hindered by high volumes of client documentation, the difficulty in validating clients’ identity and the significant level of compliance requirements
AI can link, enrich and enhance transactions, risk and customer data sets to create risk-intelligence allowing firms to better assess and predict clients’ risk rating dynamically and in real-time based on expected and ongoing behaviour – this improves both the risk assessment and also the speed of onboarding
AI can profile clients through the use of entity resolution which establishes confidence in the truth of the clients identity by matching them against their potential network generated by analysis of the initial data set provided by client
Better matches can be predicted by deriving additional data from existing and external data sources to further enhance scope & accuracy of client’s network
The result is a clear view of the client’s identity and relationships within the context of their environment underpinned by the transparent and traceable layers of probability generated by the underlying data set
To improve data quality, firms need to be able to set standards for their internal data and their client’s documentation
82% of respondents cited ‘Risk Analysis & Profiling’ as having the most opportunity for improvement through AI
If documentation is in a poor state, you’ve got to find something else to measure for risk – technology that provides additional context is valuable
Transaction Screening
Key pains faced by firms are the number of false positives (transactions flagged as risky that are subsequently found to be safe), the resulting workload in investigating them, as well as the volume of ‘false negatives’ (transactions that are flagged as risky, but released incorrectly)
AI can help improve the accuracy and efficiency of transaction and payment screening at a tactical and strategic level
Tactically, AI can reduce workload by carrying out the necessary checks and transactions analysis. AI can automate processes such as structuring of the transaction, verification of the transaction profile and discrepancy checks
Strategically, AI can reduce the volume of checks necessary in the first place by better assessing the client’s risk (i.e., reducing the number of high risk clients by 10% through better risk assessment reduces the volume of investigatory checks).
AI can assist in automating the corresponding investigative processes, which are currently often highly manual, email intensive with lots of to-and-fro.
A ‘White List’ of transactions allows much smoother processing of transactions compared to due diligence whenever a transaction is flagged
82% of respondents cited ‘Risk Analysis & Profiling’ as a key area that could be most improved by AI applications
Transaction Monitoring
Firms suffer from a high number of false positives and investigative overhead due to rules-based monitoring and coarse client segmentation
AI can help reduce the number of false positives and increase the efficiency of investigative work by allowing monitoring rules to target more granular types of clients (segments), updating the rules according to client’s behaviour, and intelligently informing investigators when alerts can be dispositioned.
AI can expand the list of features that you can segment clients on (e.g. does a retailer have an ATM on site?) and identify the hidden patterns that associate specific groups of clients (e.g., Client A, an exporter, is transacting with an entity type that other exporters do not). It can use a firm’s internal data sources and a variety of external data sources to create enriched data intelligence.
Reinforcement learning allows firms to adjust their own algorithms and rules for specific segments of clients and redefine those rules and thresholds to identify correlations and deviations, so different types of clients get treated differently according to their behaviour and investigative results
Survey Results
90% of respondents to Leading Point FM’s survey on AI and Anti-Financial Crime cited ‘Volume & Quality of Data’ as being one of the top 3 biggest challenges for CLM and AFC functions in 2019
82% of respondents to cited ‘Risk Analysis & Profiling’ as having the most opportunity for improvement through AI
60% of respondents had gone live with Artificial Intelligence in at least one business use case or were looking to transition to an AI-led operating model.
However, 40% were unclear on what solutions were available 60% of respondents cited ‘Immaturity of Technology’ or ‘Lack of Business Case’ as the biggest obstacle to adopting AI applications
Conclusion
To apply AI practically requires an understanding of the sweet spot between automation and assisting, leveraging human users’ knowledge and expertise
AI needs a well-defined use case to be successful as it can’t solve for all KYC problems at the same time. In order to deliver value, clarity on KPI’s that matter and reviewing AI considering the end-to-end business process is important.
Defining the core, minimal data set needed to support a business outcome, meet compliance requirements, and enable risk assessment will help firms make decisions on what existing data collection processes/ sources are needed, and where AI tech can support enrichment. It is possible to reduce data collection by 60-70% and significantly improve client digital journeys.
There are significant skills gaps in order to move from a traditional AFC op model to more intelligent-data AI led one. When AI becomes more integral to business, mass re-training will be necessary. So, where are the teachers?
The move from repetitive low value-added tasks to more intelligent-data based operating models. Industry collaborations & standards will help, but future competitive advantage will be a function of what are you doing with data that no one else is.
70% of respondents cited ‘Effort. Fatigue & False Positives’ as one of the top 3 biggest challenges for CLM and AFC functions in 2019?
More data isn’t always better. There is often a lot of redundant data that is gathered unnecessarily from the client.
Spotting suspicious activity via network analysis can be difficult if you only have visibility to one side of the transactions
If there’s a problem worth solving, any large organisation will have at least six teams working on it – it comes down to the execution