The increasing role of artificial intelligence (AI) in financial services has sparked debates on whether human oversight remains necessary. While AI brings efficiency, its implementation must align with regulatory requirements and business objectives.
In this blog, we answer some of the common questions people have when implementing AI for financial services, especially in anti-money laundering (AML) compliance.
AI isn’t human and therefore doesn’t have bias. Is that true?
AI systems can inherit biases from the data they are trained on, requiring diligent oversight and tuning by diverse human teams. AI is all about pattern detection, so when a model gets a hit for money laundering it has a risk of ‘overlearning’ everything about that transaction. As a result, biased and discriminatory outcomes are possible because they’re based on flawed data, which is inadequate and unrepresentative of the populations from which they are drawing inferences.
Read how using synthetic data to train AI models can mitigate biases in financial crime compliance
A team with varied backgrounds can provide unique perspectives on data analysis, helping to spot potential issues before model construction even begins. Diverse viewpoints can help identify biases, imbalances, and gaps in the data that might otherwise go unnoticed.
Cross-functional teams such as KYC, data, processes, regulations, and systems should work harmoniously to minimise such vulnerabilities. Effective AI governance frameworks must integrate regular reviews of models for fairness and compliance.
Is human oversight still needed?
AI systems must operate within ethical and regulatory boundaries, ensuring transparency and accountability. Regulations, such as the EU AI Act, emphasise human oversight in AI-driven processes to maintain trust in financial services.
At Napier AI, we advocate for compliance-first AI. Rather than deploying closed AI systems that generate alerts without context, AI should be tailored to the business’s risk appetite and regulatory environment. The Napier AI / AML Index found that the UK alone could recover £90 billion annually from financial criminals, but achieving this requires human oversight in AI decision-making.
Soft skills remain indispensable in financial services. While AI enhances fraud detection and compliance, human expertise is necessary for interpreting results and making impactful decisions. AI can process vast amounts of data, but it cannot replace human judgement, particularly when ethical considerations and nuanced customer interactions are involved.
My financial institution is too late to adopt AI, how do we start and keep up?
When implementing AI into your organisation’s AML function, it’s important not to rush the process or sacrifice crucial steps in the pursuit of a speedy transition. Some financial institutions fear they are too late to adopt AI or that their staff cannot keep up. Successful AI adoption requires careful planning, including:
- Readiness & maturity assessment
- Regulatory environment assessment
- Risk assessment
- Data evaluation
- Business operating model definition
For financial crime compliance teams to make informed decisions based on AI-generated recommendations, they need to properly understand the outputs and the methodology followed to reach them. Ensuring the team can explain the AI and its outputs also encourages wider confidence within a company through knowledge diffusion
Focused training programmes can upskill staff, ensuring they understand and effectively work with AI-driven compliance solutions. AI solutions tailored to compliance priorities can be implemented without overhauling entire systems, ensuring smoother transitions for staff and infrastructure.
Regardless of whether you are developing an AI system in-house or using a third party vendor, you must carefully consider how you will migrate from your incumbent systems to the new software. The goal is to achieve a smooth system transition, to make the new technology – the AI – available to the relevant teams as quickly as possible. You should aim to do this with the least operational disruption, at the lowest cost, and with the least amount of internal and regulatory risk incurred.
Learn more about Napier AI’s compliance-first approach to artificial intelligence:
Read the 12 step guide to implementing AI for financial crime compliance