制裁措置におけるAIと自動化に関するCapgeminiのラウンドテーブルからのインサイト
Observations from Capgemini's roundtable on AI and Automation in Sanctions
I recently attended an executive roundtable of the heads of AML, Sanctions and Screening programs for some prominent global banks and fintechs. The roundtable was hosted by Supriyo Guha, Global Practice Lead for Capgemini's Financial Crime Compliance (FCC) practice. Capgemini did a great job bringing together the industry and experience of 12 bankers with the technical savvy of their own team plus two technology firms with solutions for automating and applying AI to sanctions processes - Workfusion and Hummingbird.
The conversation started with FCC executives comparing notes on the challenges they face - particularly the difficulties they face in digesting the influx of names added to SDN lists and implementing controls on payments linked to the 1,300+ new commercial items, equipment and materials added to export control lists in response to the Russia-Ukraine conflict. When discussing how to tackle them, very few mentioned technical solutions to the problem. Instead, bank compliance with the astounding increase in the number and type of sanctions is generating large amounts of incremental manual work.
“Even if an AI-enabled process will make fewer mistakes than a manual one, it easier to blame a mistake on human error than explain to a regulator why the AI made a mistake.”
This increase in manual work creates a strong use case for Banks to use automation and AI technology to modernize sanctions screening. However, while sanctions may be the area of financial crime compliance most in need of new technology, it is the area where AI has been least adopted. Only one of the roundtable participants had plans to look at applying AI to sanctions screening.
This jibes with Celent's own research. We asked 200 Anti-Financial Crime executives where they see AI having the most impact within their organizations. Looking across KYC, fraud, AML transaction monitoring and sanctions, sanctions was by far the laggard with only 10% of AFC executives citing it as where AI was having the most impact.
AI Effectiveness by FCC Function
Source: Celent Financial Crime Survey, 2023
We see AI adoption lagging in sanctions because of:
Bandwidth
Sanctions organizations are scrambling to comply with the volume and complexity of new sanctions. The resulting strain on their resources have left these organizations so overwhelmed, they lack the time and resources needed to explore potential AI or automation technologies. Technologies that could streamline workflows and provide much needed slack in the system. The pressure to stay compliant creates little time to select, plan and implement these technologies. This situation creates a vicious cycle where the lack of time and resources prevents the adoption of solutions that could ultimately alleviate the very issues they are facing.
Regulatory acceptance
Concern about regulatory reception of new AI solutions is somewhat slowing down implementation of AI not only in sanctions but across all FCC processes. Banks are hesitant to be the first to implement and review new solutions with their examiners. The burden to prove efficacy of an AI-enabled process is higher because of the need to educate regulators on how the AI works in that process, the methods of testing and validating AI systems are new to regulators, and the lack of consistency in examination teams creates additional uncertainty on how AI projects will be received.
As one participant put it, "better to stick with the devil you know. Even if an AI-enabled process will make fewer mistakes than a manual one, it easier to blame a mistake on human error than explain to a regulator why the AI made a mistake."
Risk of organizational disruption
Every technology project carries implementation risk but the stakes are higher when the sanctions organization is already running at full capacity and a hiccup in the cutover to a new system can lead to regulatory fines and censures. For this reason, banks considering using AI in their sanctions process have no interest in replacing their core screening technologies. Instead they are looking at solutions that piggyback on the current processes and technologies.
Despite these hurdles, AI is starting to make inroads into sanctions. AI leaders are, most commonly, using machine learning to reduce false positives, and some are exploring the use of natural language processing (NLP) and now GenAI to support supplementary risk assessments to improve the confidence of matches performed by the core screening engine. The pain of compliance with the current sanctions environment is driving more and more FCC executives to consider how AI and automation technologies might be the best way to contend with the ever-increasing complexity and number of new sanctions.