In generic terms, a false positive is a result which indicates that a certain condition might be present when it is not. In screening solutions for financial crime compliance, false positive hits of transactions or customers come with many associated outcomes in addition to being costly to process: blocking legitimate transactions because they’re flagged as suspicious by the system which negatively impacts the customer experience. Additionally, they divert compliance analysts' focus and time away from investigating higher-risk cases, among other issues.
What causes false positives, and how can you reduce them?
The key root causes of false positives are inefficient matching strategies, inflexible screening processes, and legacy screening systems which use inadequate algorithms that lack the ability to enhance matching results by leveraging additional data.
A robust screening strategy clearly defining who to screen and when with set criteria for auto discounting and defining decision alerts can be achieved through:
1 ) An optimal screening policy
‘One size fits all’ approach to who to screen and for what often leads to over screening increasing the potential for false positives. It’s important to have a risk-based approach to screening to ensure you are targeting the exact sanctions lists, types/status of PEP’s and type of adverse media that are appropriate for the risk and profile of customer you are screening. This is why the ability to easily build multiple screening configurations that enable targeted risk based screening for different customer types is an important first step in reducing false positives.
2) Data quality and coverage
With so much data to be processed and analysed, it’s essential that systems capture data in a clear, structured way. This means, for example, gathering title, first, second, and last names in different fields. Not only does this avoid any potential ambiguity, but it becomes easier to match against data in external sanctions, watchlist or politically exposed persons (PEPs) lists.
The more relevant data you have for verification, the lower the risk of false positives and false negatives occurring.
3) Improved matching algorithms
A company’s risk appetite must determine just how stringent the matching terms should be. A low risk appetite may see a company implementing a policy whereby a similarity threshold of just 50% would warrant a check. The opposite of blacklists, whitelists can be used to support sanctions compliance efforts by identifying and saving repetitive matches that materialised as false positives. Fuzzy name matching algorithms that consider nicknames, phonetic similarities, transliteration etc can also improve the quality of alerts, by identifying that a name could be all of William, Bill, Liam or Will.
Secondary scoring and filtering
Advanced filtering and secondary scoring can be used alongside sophisticated name matching algorithms to enable auto-discounting further reducing false positives. Even when there is a high confidence of a name match other secondary data such as date of birth/ registration, country of citizenship/registration/ residence etc may not match. These secondary data attributes can be automatically matched alongside the name to enable auto discounting when certain combinations/thresholds of both name matching confidence and secondary data matching are not met.
4) Sandbox Testing
A sandbox is an environment that provides compliance officers a controlled and isolated space for testing and developing customised rules in their anti-money laundering (AML) systems. By testing, tuning and running ‘what-if scenarios’ on a snapshot of historical data in an isolated environment, organisations can optimise rules without committing to changes in the live environment, thus reducing false positives to enhance operational efficiency.
This enables compliance professionals to compare and contrast results with current live screening configurations before making changes to the live environment via user access based controls. A sandbox also acts as a centralised repository with a full audit trail of screening configurations, with version control.
5) Alert decision optimisation
Once screening alerts/hits have been generated by the screening solution, AI machine learning can be employed to further optimise how decisions/investigations are being made by people responsible for reviewing the alerts. Through dynamic learning of screening not only can decision recommendations/auto decisioning of alerts be enabled but also these insights can support the ongoing refinement/tuning of screening configurations to further reduce false positives.
Screening configurations need to be continually modified to allow compliance professionals to be proactive in their response to emerging risks. AI tuned to the bespoke risk appetite and decision making criteria of your institution should be trained to recognise what a good match looks like and determine whether an alert needs to be auto discounted/ decisioned.
Improve your compliance processes with an award-winning solution