How to prepare product fair value assessments with an AI agent

A fair value assessment asks whether what customers pay is reasonable for what they get, for each group of customers who hold the product. The agent gathers price, charges, costs, usage and outcome data, compares them across customer groups and similar products, and drafts the assessment. The product owner proposes a conclusion, and the product governance committee decides.

What is a fair value assessment, and who owns it?

Under the FCA Consumer Duty, which sets expectations for good outcomes for retail customers, firms that make or distribute retail financial products are expected to assess whether those products provide fair value and to act where they do not. Similar expectations appear in other markets under product governance and conduct rules. The assessment looks at the total price customers pay, including charges and fees, against the benefits, quality of service and features, and at whether some customer groups get much worse value than others.

The product owner prepares the assessment. Product governance, usually a committee, approves it. Compliance challenges it. The work is data-heavy and repeated on a cycle for every product, which makes it suitable for an agent, but the conclusion is a judgement about fairness that a person must own. It sits alongside other product reviews such as reviewing a new product for compliance risk.

What does the agent do at each step, and at what depth?

1. Person, product owner: defines the scope. The product, its versions, the distribution channels and the customer groups to assess.
2. Agent, depth 1: gathers the data. Prices, interest rates, fees and charges actually paid; costs to serve where finance provides them; feature usage; complaints; claims or rejection rates; customer outcomes such as arrears or early closure; and any vulnerable customer indicators.
3. Agent, depth 1: segments. Value for each customer group: new versus long-standing customers, those who use the product as intended and those who do not, customers with vulnerability markers.
4. Agent, depth 1: compares. Against the firm's similar products and, where available, published market information. External data is marked as external.
5. Agent, depth 2: drafts the assessment. Findings per group, where value looks poor and why, and options to address it, such as changing a fee, contacting customers or withdrawing a version.
6. Person, product owner: proposes a conclusion and actions.
7. Person, compliance: challenges.
8. Person, product governance committee: decides and records the outcome.
9. Agent, depth 1: tracks actions and sets up the data for the next review.

Which controls apply, and what evidence does each leave?

Actual charges, not list prices. Value depends on what customers really paid. Evidence: the assessment uses transaction-level charge data, with its extract recorded.

Group-level analysis required. A product can offer fair value on average while one group pays far more for less. Evidence: findings for every defined group, not just the whole book.

Benchmark sources marked. Internal data is from systems of record; competitor pricing from web pages is external and may be out of date. Evidence: each comparison names its source and date of capture.

Conclusion owned by a person. The agent's draft presents findings and options. Evidence: the product owner's signed conclusion and the committee's decision.

Compliance challenge recorded. Evidence: challenge notes and responses. A supervisor reading a fair value assessment looks for evidence that someone pushed back; preparing for a supervisory visit covers what they typically ask for.

What should you measure?

• Assessments completed on schedule across the product range.
• Groups identified as receiving poor value, and actions taken.
• Actions completed and the change in outcomes afterwards.
• Committee challenges that changed a conclusion.
• Data gaps, such as missing cost-to-serve data, that weaken assessments.
• Customers contacted or refunded as a result of assessments, which shows whether findings turn into outcomes for the people affected.

Where does it usually go wrong?

Assessments that always pass. If every product passes every time, the method is probably too lenient. Look for the groups that would complain if they knew what others paid.

Loyalty penalties missed. Long-standing customers on older terms often pay more. Segmenting by tenure makes this visible. The agent should compare what new and long-standing customers pay for the same product today, not just at the point of sale, and show the gap for each version still held.

Costs without benefits. Comparing price to cost tells you about margin, not value. The benefit side, features used, claims paid, service received, must be assessed too.

Bundled products assessed in pieces. Where a product comes bundled with insurance or add-ons, assessing each part alone can miss that customers pay for features they cannot use. The agent should assess the bundle as customers buy it, as well as each component.

Distributor data not obtained. Where a third party distributes the product, its charges and customer outcomes are part of the value picture. Request them.

Assessment separate from risk work. Fair value findings should feed the wider compliance risk view; see running compliance risk assessments.

The overall stance on compliance agents is in AI compliance agents.