Help
What Dutified does
You give us a customer communication. We put it in front of AI customers built to match your customer mix, then run it again with the same number weighted so every vulnerable circumstance has its own sample. Each AI customer reads only what a person like them would take in, then answers in their own words. We mark every answer against the key points you say the communication must get across.
Every report says how many AI customers read it. During the pilot a run uses up to 100 AI customers, so a test has up to 200 readers; full runs, not yet switched on, use 1,000 per run. Groups under 100 readers are labelled "small sample" on every report, and scores show their margin of error.
AI customers are a model of how people read, not real customers. Dutified helps you judge your communications. It is feedback, not legal or compliance advice, and it is not approval by the FCA.
You get the eight lenses for every type of customer, who struggled and why, the exact words and sentences people trip over, a short fix pointer for each finding and a governance trail you can export. The eight lenses are Dutified's way of reading a communication against the FCA's rules: the four Consumer Duty outcomes, the three cross-cutting rules and the FCA's guidance on vulnerable customers (FG21/1). Each one says where to look in the FCA Handbook. The FCA does not publish them as a list of eight.
How the scoring works: the questions each AI reader answers, how answers are marked, how the lenses and the verdict are worked out, margins of error and small samples.
Dutified diagnoses, the firm writes
Dutified finds the problems: what works, what does not, for whom and why. It does not promise better copy and it does not write replacement wording for you. Each finding comes with a short fix pointer, such as "Give the deadline as a date" or "Show the charge in pounds". Your own writers and compliance team choose the words.
If you want one, you can ask for a suggestion on a single finding that quotes your words. It is marked "Suggestion", and you accept it, edit it or ignore it. Nothing is changed, re-tested or approved for you: a new version starts only when you start one, and it includes only the suggestions you accepted or edited.
Reading a report
- Every report opens with three lines: what worked, the one thing to fix first and who struggled most. More about the three lines.
- Under a finding, it may say what fixing it could do. That is an estimate, not a result. How estimates work.
- A new version is read by the same AI customers, so any change comes from the words. About retests.
- Findings on how easy your own communication is to read for disabled people go under "Accessibility". What we check.
Feedback, not advice
Dutified reports how modelled readers responded. It is feedback, not legal or compliance advice. Decisions stay with your firm.
Reports say what happened with readers, such as how many missed the deadline. They never say whether a communication meets a rule. A rule finding says what the text does and where to look in the FCA Handbook; it is not a ruling. The person at your firm decides every change and every approval.
Your choices for reports
Vulnerable customers. One panel on firm setup holds two choices:
- Pass mark for vulnerable customers: Standard, the default, or Stricter. Choose Stricter if you want vulnerable customers to meet the same bar as everyone else.
- Weight given to vulnerable customers: Standard (1x), Higher (1.5x) or Highest (2x). In the headline scores, each reader in a vulnerable circumstance counts that many times. It never lowers the pass mark.
What your reports show. Scores and verdict, or feedback only. Feedback only says in plain words what works, what does not and why. It has no percentages, scores or verdict colour. The scores are still kept, so a report can be switched later.
Before anyone sees scores for the first time, they tick a box to say they understand what the scores are. What the box says.
Your firm administrator sets these on firm setup. The person who sets up a test can change them for that test. Every report says which it used, and every change goes on the governance trail. More about these choices.
What the colours mean
Every score has a word as well as a colour. Each lens and each customer group is scored like this:
| Green | 80 or more out of 100. |
| Amber | 60 to 79. |
| Red | Below 60. |
The overall verdict
The verdict is a set of gates, not an average. It is worked out from the whole run like this:
| Red Many readers struggled | Any one of: understanding below 55%; customers in vulnerable circumstances below 45%; avoiding harm below 55%; more than 15% of readers would do something that leaves them worse off; the text does something a critical rule looks for (a rule finding is a prompt to check, not a ruling); or key points are missing from the communication. |
| Green Most readers understood it | All of: understanding at least 75%; customers in vulnerable circumstances at least 65%; avoiding harm at least 70%; no more than 8% would act in a way that leaves them worse off; and no red rule finding. |
| Amber Some readers struggled | Anything in between. |
In this version the verdict does not look at each customer group separately, because during the pilot most groups are small samples. A group can score red while the verdict is green, so always read the grid of lens by customer type as well. How the verdict lines work.
No customer data
No customer data ever enters Dutified. Firms may give only an optional customer mix (percentages by age, channel, circumstance).
What you test is the communication itself, written for customers in general. Before you upload it, take out names, addresses, account and policy numbers, postcodes and anything else about a real person, and use a placeholder such as "Dear [Customer name]". We run a simple check and ask you to remove anything we spot, but the check is a help, not a guarantee. Dutified never asks for a customer list, customer records or details of any customer.
Your data
No customer data ever enters Dutified. Firms may give only an optional customer mix (percentages by age, channel, circumstance). The communications you upload and every result are stored in Dutified's own database. Each firm's work is kept apart: people at your firm see only your firm's work, and every screen checks that before it shows anything.
The Dutified team can see your tests so they can check every finding before you get the report. When someone at Dutified opens one of your tests, that goes on your firm's governance trail.
To run a test, the words of the communication (and your key points) are sent to Anthropic's Claude models, which play the AI customers and mark their answers. While Dutified is being built and tested, this copy of Dutified reaches those models through the Dutified team's own Claude plan (the claude command line), not through Anthropic's business API. It is for building and testing with made-up and demo communications only. The live service will use Anthropic's API under its commercial terms for business use, which say Anthropic does not train its models on that data. Nothing else is shared with anyone.
A period for keeping tests has not been set for the pilot yet. The Dutified team will agree one with your firm; until then tests are kept. If you find customer details in a test after it has run, ask the Dutified team to erase its words. That takes the words out of the app: the communication, its title and key points, what each AI customer saw and said, quoted passages and suggestions, and any upload. The scores and decisions stay. Three things it does not reach: the governance trail, which cannot be edited, so a note on it that quotes the words stays; the database backups, the newest 30 of which are kept, so an erased test drops out of them as they are replaced; and the copy the AI provider keeps under its own terms.
Dutified does not send email itself yet. Invites, report notices and approval requests all appear in the notifications list inside the app, so check it when you sign in. An invite is the one thing a new person cannot see in the app, so the Dutified team passes the invite link on to them by hand.
What it costs
Nothing you do in the app adds a charge, and the app never shows prices. What you pay is set out in your agreement with Dutified. Ask your Dutified contact if you have a question about it.
What to do with a report
Start with the quick answer at the top, then the eight lenses by customer type, who struggled and why, and the words that confused people. For each finding, read its fix pointer and say whether your firm will change the words or keep them (with a reason). Your writers change the words in a new version, which you retest when you choose. Your approvers decide, and every step is kept on the governance trail.
The consumer model
Every result is produced by a numbered version of the consumer model, and every report records which version ran it, with the settings and AI models used, so a result can be traced to exactly what produced it. The AI customers answer freely, so a rerun gives close but not identical numbers. The version in use now is 1.0.
Past reports never change, even when the model moves on.
How we know it works, and what we have not yet shown
Dutified keeps a benchmark of 20 made-up communications (letters, emails, texts, web pages, an app screen, a statement, ads and a call script), each with the verdict an experienced reviewer would expect: 8 red, 4 amber and 8 green. A model version passes when at least 18 of the 20 come out with the expected verdict and none is two colours out (green for an expected red, or red for an expected green), with 100 AI customers on each.
No version has passed yet. The model in use now, 1.0, has only been through shorter practice runs with fewer AI customers, where 12 and then 13 of the 20 came out as expected. Later versions reached 14 of 20 on full runs. Until a version passes, treat a verdict as a well-reasoned second opinion, not a measured result, and read the reasons behind it. Every report says this beside the verdict.
Not yet shown: the AI customers are a model. Their answers have not yet been compared with what real customers understood from the same communication. That comparison is planned with pilot firms.
Need a person?
Dutified is run by Dan Ilett. If your firm is not using Dutified yet, request access and say what you would like to test. Dan will get back to you.
Request access