人工智能如何改变银行打击金融犯罪的方式（英）

Risk & Resilience PracticeHow agentic AI can change the way banksfight financial crime Financial institutions are allocating significant resources to fightingfinancial crime, but they are generally making little progress. AI-basedsolutions may be an accelerator. This article is a collaborative effort by Alexander Verhagen, Angela Luget, Olivia Conjeaud, andVasiliki Stergiou, with Debanjan Banerjee, representing views from QuantumBlack, AI by McKinsey,and McKinsey’s Financial Services and Risk & Resilience Practices. Banks are spendingever-larger sums of moneyon know-your-customer and anti-money-laundering(KYC/AML) activities. But there is little evidencethey are getting a good return on their investments.In fact, according to Interpol, the financial industrydetects only about 2 percent of global financialcrime flows, despite increasing spending by upto 10 percent a year in some advanced marketsbetween 2015 and 2022.1A potential solutionlies in agentic AI2—an evolution of analytical AItechnology that offers automation and productivitythroughout the client life cycle (Exhibit 1). up to 10 to 15 percent of their full-time equivalentsto KYC/AML alone.3In parallel, automation ratesare generally low amid fragmented data resourcesand unstandardized data sets. The result is thatteams waste a lot of time on manual tasks whileclients complain of tiresome interactions and lumpyprocesses. AI, specifically agentic AI, could be the antidote toKYC/AML headwinds. In this article, we map the AIlandscape and examine options for implementation,highlighting how some leading institutions havedeployed the technology to their advantage. Our keyconclusion is that AI offers transformative potential,but only if institutions put in place the foundationsand capabilities that will support an at-scale rollout. Much of the cost of combating financial crimerelates to inefficiencies in operating models andways of working. Indeed, banks commonly assign Exhibit 1Exhibit <1> of <3> Financial crime is a high-potential area for AI. Operational cost and gen AI potential, by banking risk sector Financial crime (FC) challenges Large cost base:Up to ~20% of banks’ full-timeemployees are typically dedicated to FC activities Low automation rates:Case-handling processeslack automation and optimization, resulting in manymanual reviews performed across segments Data fragmentation:Analyses depend on a mix ofinternal and external data, both structured andunstructured, making it difficult to deployautomated data extraction and analysis tools Multitude of reports created:FC officers spendmost of their time creating detailed, case-specificreports such as know-your-client memos andnegative news reports Suboptimal client journeys:Existing clientprocesses, such as onboarding, are inefficient andfail to meet growing expectations for speed andconvenience Analytical AI, generative AI, andagentic AI: A short tutorial onfinancial crime use cases anomaly detection. And it can apply decision-tree-based models, a type of machine learning algorithm,to improve underperforming rules. AI is not, in reality, a single technology but ratheran umbrella term for a range of technologies thatcan understand and generate language, recognizeimages or speech, make decisions or predictions,and learn from data over time. In the KYC/AMLcontext, these capabilities are broadly expressed inthree forms (Exhibit 2). Generative AI Generative AI (gen AI) learns from patterns indata sets and uses those learnings to generateoriginal output. In KYC/AML, it can support humaninvestigators across a number of use cases,including onboarding and in-life client reviews,based on analysis of structured and unstructureddata. The technology can save human time incollecting and extracting data from documents,summarizing large sets of information (for example,on adverse media) about individuals and entities,and accelerating investigations, including analyzingpurpose and nature statements, source of fundsor wealth drafts, and corporate business activitydescriptions. In transaction monitoring, genAI is useful in producing alert conclusions andtransaction analysis insights, supporting draftingof suspicious activity reports, and contributing toquality control and quality assurance (QA). Analytical AI Analytical AI can complete analytical tasks fasterand more efficiently than humans can. Prominentuse cases include false positive detection incontrols, including transaction monitoring,sanctions detection, name screening, and frauddetection. The technology can also produce moredynamic and integrated customer risk ratingmodels, for example, by incorporating a highernumber of behavioral (including transaction-based)factors. In transaction monitoring, it can sharpenaccuracy and facilitate peer group comparisons and Exhibit 2Exhibit <2> of <3> Three successive generations of AI development show a clear evolution intask handling. Three examples of agentic AI use in investig