Skip to main content
Course contents

Module 0: Start here

Client data and what to leave out

20 minutes. Last checked 9 November 2026.

Lesson 2 of 3 in this module

Everything in this lesson is fictional: the firm (Ashdown Wealth Planning), its clients and their fact-finds. Do not paste real client data to follow along.

What you will do

Cut a fact-find extract down to what one task needs, then turn it into a referral note for the firm's mortgage adviser.

When to use it

A fact-find holds far more than any single task needs: income, health, family, debts and every plan the client has. This replaces pasting the whole fact-find into a chat tool, or keeping AI away from client work because nobody at the firm is sure what is allowed.

Two UK GDPR rules bear on this. The first is Article 5.

Personal data must be adequate, relevant and limited to what is necessary for the purpose (Article 5(1)(c), data minimisation).

In practice that means you choose the task first, then keep only the fields the task uses. A referral note about a remortgage needs the client's income and the loan. It does not need their health, their children's names or their pension funds.

Health details, and anything about a client's family or personal circumstances, are the most sensitive part of a fact-find. Leave them out unless the task cannot be done without them. Where it cannot, follow your firm's procedures for those, and keep them out of this prompt.

Swapping the name for [CLIENT] does not make the rest anonymous. A date of birth, an employer, a job title or a postcode can still point to one person, so those go too, or become something generic such as an age.

The second is Article 28.

A controller must only use processors that give sufficient guarantees of appropriate technical and organisational measures (Article 28(1)).

Article 28 also sets out what the contract with a processor has to do.

Processing by a processor must be governed by a contract or other legal act, in writing (which includes electronic form), that requires the processor to process personal data only on documented instructions from the controller (Article 28(3) and 28(9)).

So before client data goes into a tool, check whether the terms make the provider your processor. Do not assume the answer either way. The check is in the terms, not in the tool's adverts or its settings screen:

  1. Find out which plan your firm is on with that tool: a personal plan someone signed up for, or a business plan the firm signed up for. Terms differ between plans, so read the terms for the plan you actually use.
  2. In those terms, look for a data processing agreement or addendum (often called a DPA).
  3. Read who it names as controller and who as processor, and whether the provider processes data only on your documented instructions.
  4. If you cannot find a DPA, or it does not say these things, treat the check as not passed. Keep client data out of that tool and ask whoever looks after data protection at your firm.
  5. Note what you found, and the date, where your firm keeps its records about the tools it uses.

The data settings. ChatGPT and Claude each have a setting that controls whether your chats are used to improve their models. The paths below are for the personal plans. Menus move, so check the path on the day.

ToolWhere the setting is
ChatGPT (Free, Plus or Pro)Settings, then Data controls, then turn off "Improve the model for everyone"
Claude (Free, Pro or Max)Settings, then Privacy, then turn off "Help Improve our AI models"

Check whether the setting is on before you start. Turning it off changes what the provider does with your chats, but it is not a contract, so it does not settle the processor check. For any other tool, Copilot included, look for a data or privacy section in its settings, and read the terms for your plan.

On your next working day, before you paste any part of a fact-find:

  1. Write the task in one line, such as "a referral note for our mortgage adviser".
  2. Copy the fact-find extract into a plain document first, not into the AI tool.
  3. Delete every field the task does not use: health, other family members, other plans, assets and debts the task does not touch.
  4. Replace what identifies the client with a placeholder in square brackets, such as [CLIENT] or [EMPLOYER]. Turn a date of birth into an age.
  5. Check whether the tool's terms make the provider your processor, using the steps above.
  6. Paste the prompt below with the cut-down extract. Put the real details back yourself, outside the AI tool.

What step 3 changes, on part of one fact-find:

Whole fact-find section

Mr [CLIENT], born 14 May 1980, salary £58,000. Type 2 diabetes, diagnosed 2023. Two children at [SCHOOL]. Workplace pension with [PENSION PROVIDER]. Mortgage balance £182,000, fixed rate ends 31 March 2027.

The name is out, but the date of birth, the school and the health detail are still there, and the referral needs none of them.

Cut down for the referral

[CLIENT], aged 46, salary £58,000. Mortgage balance £182,000, fixed rate ends 31 March 2027.

Only the facts a remortgage referral uses are left.

The prompt

Prompt parts:

  • Role
  • Context
  • Task
  • Inputs
  • Constraints
  • Format
  • Questions first

  1. Part 1. Role: You are helping a UK financial adviser at a small firm write an internal referral note for the firm's mortgage adviser.

  2. Part 2. Context: The client wants to remortgage when their fixed rate ends. The fact-find extract below has been cut down to what this referral needs. Placeholders in square brackets, such as [CLIENT] and [EMPLOYER], stand for details I have left out on purpose.

  3. Part 3. Task: Write the referral note.

  4. Part 4. Inputs: The cut-down fact-find extract, pasted below.

  5. Part 5. Constraints: Use only the facts in the extract. Keep every placeholder exactly as written, and do not guess what it stands for. Do not give mortgage or investment advice, and do not say which deal or lender would suit the client. Where the referral needs a fact the extract does not give, write "not in the extract".

  6. Part 6. Format: A short note under three headings: "What the client wants", "The facts we hold" and "Not in the extract". One line per item.

  7. Part 7. Questions first: If anything you need is missing, ask me before writing.

[paste the cut-down fact-find extract here]

Why this prompt works

The Context part carries this prompt: it tells the tool the extract was cut down on purpose and what the brackets mean, so the gaps come back listed under "Not in the extract" instead of filled with guesses.

Sample input

Paste this under the prompt, in place of "[paste the cut-down fact-find extract here]".

Sample input

FICTIONAL SAMPLE FOR COURSE PRACTICE. Not a real client.
Ashdown Wealth Planning, fact-find extract, cut down for a mortgage referral

Client: [CLIENT], aged 46
Employment: employed by [EMPLOYER], salary £58,000 a year
Partner: [PARTNER], aged 44, self-employed, income not yet confirmed
Home: owned jointly with [PARTNER], mortgage balance £182,000
Current mortgage: fixed rate ends 31 March 2027
Goal: remortgage when the fixed rate ends, and keep the monthly payment about the same
Other debts: car finance, £240 a month, ends June 2028

Example output

Example output. Yours will differ in wording.

(Fictional, written for this course.)

What the client wants

  • To remortgage when the fixed rate ends on 31 March 2027
  • To keep the monthly payment about the same

The facts we hold

  • [CLIENT], aged 46, employed by [EMPLOYER], salary £58,000 a year
  • Home owned jointly with [PARTNER], mortgage balance £182,000
  • [PARTNER], aged 44, self-employed
  • Car finance of £240 a month, ending June 2028

Not in the extract

  • [PARTNER]'s income: not in the extract (the fact-find says it is not yet confirmed)
  • The current monthly mortgage payment: not in the extract
  • The value of the home: not in the extract
  • The current lender, and whether an early repayment charge applies: not in the extract

What to check

  • Every placeholder, spelt as you wrote it. [PARTNER] should not come back as "the partner" or as a guessed name. A changed placeholder is easy to put back wrongly.
  • No advice. The note should name no lender, no deal and no product. Choosing those is the mortgage adviser's job, after their own fact-find.
  • Nothing you cut has come back. The tool cannot see what you deleted, but it can invent something that sounds like it, such as "no health issues noted". Delete any line like that.
  • The "Not in the extract" items. Each should be a real gap. Fill the ones you know from the file yourself, outside the tool, and delete any that do not apply.
  • Your cut-down extract, read once more. Before you paste, look for anything that still points to one person: an unusual employer, a job title only one person holds, a rare hobby.
  • The processor check, done first. If you have not checked whether the tool's terms make the provider your processor, the extract should not have gone in.
  • The real details, put back by you. Swap each placeholder back in the client file, outside the AI tool.

Variations

  1. One line, not the extract. If you only want help wording one sentence of the note, paste that sentence alone. The smallest paste that does the job is the right one.
  2. A colleague's extract. When a colleague hands you an extract, ask which task it was cut down for. A cut made for one task can hold fields another task does not need, so run step 3 again for yours.
  3. A note to the current lender. Change the Role and Task to "a note asking the current lender for a redemption statement". That note needs the lender and the client placeholder, not the salary, so cut the extract again before you paste.

Try it

About 10 minutes. The extract below has not been cut down. First, list what you would delete or replace before pasting it into any AI tool. Then write the whole prompt for a new job: a note to the firm's administrator asking them to book the client's annual review and send the pre-review questionnaire.

Fictional fact-find extract to use:

Sample input

FICTIONAL SAMPLE FOR COURSE PRACTICE. Not a real client.
Ashdown Wealth Planning, fact-find extract, NOT cut down

Client: Mrs Eleanor Price, born 2 February 1961
Address: 4 Mill Lane, Kingsbridge
National Insurance number: QQ 12 34 56 C
Health: hip replacement in 2025, recovering well
Family: widowed; one son, Daniel, who has power of attorney
Income: State Pension and a workplace pension from her late husband's scheme
Investments: stocks and shares ISA, about £96,000, on the Westmoor platform
Annual review: due each March
Preferences: meets by video call; wants papers by email, not post

This is a free sample from AI for Financial Advisers. See the full course.

Already bought it? Go to your course.