Can an AI agent review a new product for compliance risk?

It depends on how complete the product proposal is. With a clear proposal, an AI agent can map the product to every obligation it touches, check fees against value, flag data and financial crime risks, and draft the assessment. The compliance opinion, the conditions attached and the decision to launch stay with the compliance officer and the product approval committee.

Why is new product review a good fit for an agent?

New product approval is where firms prevent problems instead of fixing them. It is also where compliance is most often rushed: the product team has a launch date, the proposal arrives late, and the reviewer has a week to consider conduct, financial crime, data protection, prudential and operational questions.

Much of that review is systematic. For each feature, which rules apply? What does the target market look like? Are fees consistent with the value delivered? Is the onboarding flow collecting the right data for due diligence? Does the product create new payment flows that monitoring does not cover? An agent can work through these questions against the firm's obligations register and produce a first assessment that the reviewer then challenges.

What may the agent prepare, and what stays with people?

The agent may:

• Break the proposal into features, fees, channels, customer types and processes.
• Map each to the obligations it engages, from the obligations register.
• Assess the target market definition and whether distribution matches it.
• Compare fees and charges with the benefits, and with similar products the firm offers, as input to the fair value assessment; see how to prepare product fair value assessments with an AI agent.
• Review draft customer documents and promotions for clarity and consistency with the terms.
• Flag financial crime risks: new payment routes, cash, cross-border flows, anonymous use.
• Flag data protection triggers, such as new data types or processors that need assessment.
• Check supplier contracts against required clauses; see can an AI agent review contracts for compliance?
• Draft the assessment with risks rated against your methodology.

People decide:

• The compliance opinion: support, support with conditions, or oppose.
• Whether the target market and value are acceptable.
• What conditions must be met before launch.
• Whether the product fits the firm's risk appetite.
• Launch approval, by the product approval committee.

What does an assessment look like?

Take a consumer lender proposing a new "flexible credit line" sold through a partner's checkout. Customers draw small amounts and repay over a few months.

The agent maps the product to consumer credit, affordability, promotion, complaints and data obligations. It flags five issues. The target market says "creditworthy adults", which is too broad to test value against. The late payment fee is high relative to the typical draw. Customers will see the partner's branding first, so the lender's identity may not be clear at the point of sale. The partner will collect application data, making it a data processor or joint controller, which the contract does not address. And the repayment flow includes card repayments from third parties, a financial crime risk the monitoring rules do not cover.

The compliance officer gives a conditional opinion: a narrower target market, a lower late fee, clear lender identification at checkout, a data processing agreement and a new monitoring rule before launch. The committee approves on those conditions. Whether the selling itself creates conduct risk is a separate check; see how to review sales calls for conduct risk with an AI agent and what conduct risk is.

Which controls keep the review sound?

• The register drives the mapping. The agent maps to obligations your firm has recorded, and flags features it cannot map.
• Gaps in the proposal are findings. A missing target market or fee rationale is reported, not assumed.
• Conditions are tracked to launch. The agent checks each condition is met before the launch date and reports any that are not.
• Post-launch review is scheduled. The agent compares actual outcomes with the assessment after launch.
• The assessment feeds the risk assessment. New product risks enter the next compliance risk assessment.

The general principle is in our guide to AI compliance agents.