Can an AI agent monitor customer outcomes for Consumer Duty?

Yes, for the measurement. An AI agent can join product, pricing, communications, complaints and servicing data, track outcome measures over time and show where some groups of customers fare worse than others. Judging whether an outcome is good, what counts as foreseeable harm and what the firm changes stays with the people accountable under the FCA Consumer Duty.

What does outcome monitoring require?

The FCA Consumer Duty requires firms to act to deliver good outcomes for retail customers and to be able to show that they do. That means evidence: whether products meet the needs of their target market, whether price is reasonable for the value received, whether customers understand what they are told, and whether support works when they need it. Firms are also expected to look for differences in outcomes, especially for customers with characteristics of vulnerability.

The data exists, but in pieces. Product holdings sit in one system, fees in another, call transcripts in a third, complaints in a fourth. Joining them by customer and by period, then cutting by group, is the heavy part. Agents do that joining well.

What evidence does the agent assemble?

The agent also monitors outcomes produced by other agents. If a support agent handles a share of customer contact, its outcomes need the same scrutiny. How to evidence this for agents is set out in FCA Consumer Duty evidence: monitoring outcomes of AI agents.

What decision does a person keep?

Suppose the agent shows that customers over 75 contact the firm twice as often after an annual statement. It cannot say whether that is a bad outcome. Perhaps the statement confuses them. Perhaps they prefer to confirm by phone and are well served. Telling the difference needs someone to read calls, talk to the front line and make a judgement.

So people decide:

• Which measures matter for each product and what good looks like.
• Whether a difference in outcomes is a harm, and whether it was foreseeable.
• What to change: the product, the price, the communication or the support.
• Whether customers already affected need redress.
• What goes to the board in the firm's assessment of whether it is meeting the Duty.

What does a finding look like?

Take a credit card provider. The agent's quarterly outcome pack shows that customers who move to a payment arrangement are three times more likely to receive a late fee in the following two months than other customers in arrears.

It digs further. The payment arrangement process sets a new due date, but the fee engine still uses the old one for the first cycle. Around 600 customers were charged a fee in the last quarter for paying on the date they had agreed. Many of them were flagged as vulnerable.

The agent drafts the finding with the evidence, the likely cause and the affected population. The product owner confirms the cause, agrees it is a poor outcome that was foreseeable, refunds the fees and orders a system fix. The agent tracks the refunds and re-runs the measure after the fix. Testing whether agents themselves would avoid this kind of harm is covered in FCA Consumer Duty good outcomes: testing AI agents against them.

Which controls make outcome monitoring credible?

• Measures defined by people. The agent tracks what the firm has decided matters, and proposes additions.
• Group cuts are routine. Every measure is cut by vulnerability and other relevant groups by default.
• Findings cite data. Each finding links to the queries and records behind it.
• Agent outcomes included. Agents that serve customers are monitored like any channel; see Consumer Duty and AI support agents.
• Decisions recorded. What the firm decided about each finding, and why, is kept for the board and the regulator.

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