For financial services
The Customer Comprehension Test
Consumer Duty asks you to show that your customers understood what you sent them. We test your communications against a synthetic customer population calibrated on your own customer base, at a scale face-to-face panels cannot reach, and we report the customer who understood least, not the average.
The problem
You are expected to evidence consumer understanding. Panel research gives you thirty people, once a year, six weeks after you needed the answer, at a price that means you test the letters you have to and not the ones you should. So most communications go out untested, and you find out they were unclear when the complaints arrive.
What we do
We build a population of AI agents modelled on your actual customers, using your own evidence: complaints data, call transcripts, existing research, segmentation, vulnerability flags, product holdings. No personal data, aggregate and anonymised only. Then we put your communication to that population and ask each of them what it means.
Not whether they liked it. What it is asking them to do. What happens if they do nothing. What it will cost. What the deadline is. What they would do next. And which words stopped them.
The floor, not the average
Ninety-two per cent comprehension is not an answer if the other eight per cent are the customers who most needed to understand. Every test reports the worst-served decile alongside the headline. Every test also runs twice: once against your evidenced population mix, and once as a vulnerability stress test where vulnerable personas are deliberately oversampled.
You choose how many agents, how many versions to compare, and whether the human validation panel is included. You do not choose the vulnerability weighting. That comes out of discovery, evidenced, because a weighting you set yourself is a number you will have to defend.
Vulnerability as it looks in wealth
Vulnerability in asset management is not the retail banking definition. It is cognitive decline. It is customers in their eighties holding complex products. It is the newly widowed managing money their partner always handled. It is attorneys acting for a relative, people mid-treatment for a serious illness, inherited wealth with no experience behind it, and customers whose eyesight has made your layout unreadable. The persona population is built around those drivers.
Validated against real people
Synthetic testing is fast and wide. It is not evidence on its own. A small human panel drawn to match your profile answers the same questions, and we report the agreement rate between the two openly, including when it is poor. That number is what makes the synthetic result defensible, and hiding it would make the whole thing worthless.
What you get
- A calibrated population profile. Built once at discovery, reused on every test after it.
- Comprehension results overall, by segment and by vulnerability driver, with the floor reported alongside the mean.
- The specific language that failed. The phrases, sentences and figures that confused people, and how many people each one stopped.
- Version comparison, where you are testing more than one draft.
- Recommended rewrites, and a retest of them.
What this is not
We are the instrument rather than the auditor. This is testing and analysis. Regulatory advice, compliance sign off and any verdict on whether a communication passes all sit outside what we do. We give you the reading. What you conclude from it, and what you do about it, is yours.
Who it is for
Asset managers, wealth managers, platforms, life and pensions providers, and the compliance, product and customer teams inside them who have to evidence consumer understanding and would rather do it before the letter goes out than after.