Lessons the AML world can learn from the ‘A Level Algorithm’ Fiasco.
By Andy Lee, – Founder Sphonic
Understanding the Problemspace
Don’t get me wrong, I’m not trying to be a Smart Alec here having been a practitioner in the AML and Financial Crime industry for many years, I certainly recognise the value of AI (Artificial Intelligence) and its eradication of illicit activity. AI has been around for a long time as a must-have tool for financial services firms to counter and reduce financial crime. As the world got all Fintech in recent years the use of AI has continued to be a common tool used by firms large and small, with many new and exciting AI technology firms entering the market and disrupting incumbent players.
What we do know is that the use of AI over the last 20-30 years hasn’t reduced:
· the prevalence and typologies of fraud and AML
· the level of fines and regulatory scrutiny to counter the illicit transfer of funds
· remediation projects to retrospectively review cases flagged as a potential AML risk, often through costly and time consuming manual processes
· the dogged problem of false positives and declining genuine customers access to accounts and funds
Similarly in other environments for AI such as its use for predicting and awarding A-Level results for students impacted by Covid-19 in the Summer of 2020, it has had a devastating effect for many kids who have seen their predicted results slashed by 3-4 grades in many cases. As such, the resultant impact on their future careers and prospects by not being able to get onto the courses or institutions they had hoped for, which has the potential for long-term adverse effects.
So whilst we believe AI still has a significant role to play in the reduction of financial crime it’s important we look at the way we profile customers and transactions in a more holistic and linear way to avoid some of the fallouts described previously. We use a technical term at Sphonic “RIRO” – Rubbish In, Rubbish Out – meaning that if your AI system doesn’t have the appropriate level of data to make that customer decision the potential for customer fallout is increased, as is the chance you might let many cases slip through the net. The role of data is covered in more detail later on, but given the Sphonic platform is underpinned by a marketplace of over 85 of the worlds leading risk management technologies, the ability to unlock innovation to improve AML processes is endless.
Sphonic’s Focus on Solving AML challenges in the new Digital Economy
In addition to the considerable amount of Customer On-Boarding (both KYC and KYB) we are best known for, the foundations of the Sphonic tech stack by design was to enable the activation of further use cases :
· Account changes – to identify account takeover
· AML Transaction monitoring Screening Capabilities
· Log-on & Customer Authentication
· Bonus abuse & supporting responsible gambling
Our platform has always been able to “join data up” across multiple use cases and break down silos which is a common challenge with most firms using multiple disparate systems profiling the same customer or business.
With our recently launched Transaction Monitoring capability, we act on the premise that Knowing Your Customers is not purely about sign-up and KYC requirements at the point of origination, but rather “knowing them” through their entire tenure with a Client.
Any transaction monitoring system should know each and every customer and reach a real time decision through utilisation of such knowledge. As such, knowledge grows where a customer is acting within “normal” bounds – they should score lower in such a system and only be flagged for review where any “anomalies” are identified in real time in the data acquisition and processing phase at the point of the transaction.
Such an approach ultimately means
· Our clients don’t waste unnecessary funds in acquiring new data
· Our clients’ customers are not unnecessarily impacted
· Our clients can utilise efficient resource levels to manage reviews
· Our clients can fully deliver their AML transaction monitoring policies
Add in the fact that as with all Sphonic technology we save our clients valuable time, design and development resource and costs through our ease of implementation and on-going management.
Key to our new improved system
· Accurate data (both present & historical) has to be readily available to the decisioning process
· Data has to be available both through real time APIs and available for display to a reviewing agent in a contextualised and logical way
· Agents must be able to prioritise what to review
· Agents must be able to efficiently and logically identify the data of significance. This last part is especially crucial in the world of “big data” with hundreds and even thousands of data points available.
· Audit trail – delivering the level of data in the API to the client to maintain their own records for regulatory scrutiny and/or the capabilities in the Sphonic Case Management solution.
Unlocking Innovation in Data to tackle growing AML challenges
The world of big data in itself has thrown up huge challenges to Banks, Fintechs & Gaming Operators. As a result (and a big plus to the industries) more data is available to aid decisioning, be it in real-time or post review. Over the last 8 years of supporting some of the leading digital brands in the world, we have developed unprecedented insights and knowledge of data points from the 85+ vendor APIs we have access to. This is a strong asset in collaboratively supporting our clients with the correct data acquisition and utilisation strategy.
As such when considering a data-driven strategy the following factors are key:
· What data is of importance?
· Is such data available?
· What data should be captured, processed and paid for?
· How should data be both organised and managed?
· What value & context is attributed to each data point?
Key to Sphonic’s new AML capability is the ability for our flagship Workflow Manager (WFM) product to “dynamically” create scores. The team at Sphonic has combined dozens of years of developing scorecards for digital clients.
· Many businesses set up scorecards and rarely change them
· Many businesses set up scorecards and rarely analyse them
Even at set-up clients spend ages deciding what scores to attribute (having spent ages identifying what rules are required). This last part is challenging as it requires the data to be captured and processed to create the rule in the first place. As more vendors have entered the market with new innovations in fraud and financial crime data, the problem is further exacerbated by on-boarding new APIs and bringing data from these APIs into existing scorecards.
What value to new data points is available through innovation?
Besides initial design and on-going management, the utilisation of dynamic scores through Sphonic’s platform enables “bespoke” scores to be calculated on a customer-by-customer basis comparing them to themselves. Imagine one of our Client’s having 1M customers and each and everyone of the 1M having their own bespoke scorecard? This is what Sphonic has delivered.
Sphonic’s technology has also been able to compute such values in super-fast time and record for future retrieval (in-line with various DPA regulations). After years of analysing the innovation in data, Sphonic has designed a proprietary algorithm to dynamically assign scores to the various variances computed ensuring each customer receives unique scores per each applicable rule. Of additional importance it only processes and computes data of interest required to reach a result.
In summary, the role AI has played on existing AML scoring methodologies has huge room for improvement as does AI as a technology itself. As with the A-Level fiasco, if the data entered is limiting, irrelevant or has incorrect bias, it has the ability to create more havoc than the problem it is ultimately trying to solve. We created our AML transactional monitoring capabilities to focus on identifying the ‘needles in the haystack’ who are going to be the bigger problem cases and assert tolerance levels to the broader customer base through the use of additional innovative data and clever algorithms.