AI note-taking for mortgage broker fact finds

The fact-find appointment is where advisers lose the most time in the least visible way. The meeting takes an hour. Writing it up properly takes another thirty or forty minutes, and because it happens after the client has gone, it slides to the end of the day, then to tomorrow, then to whenever there is a gap.

Automatic note capture attacks exactly that gap. It does not make the meeting shorter. It makes the write-up start from something rather than nothing.

What the note is for

Before choosing any tool, be precise about the job the note does, because a fact-find note serves two different masters.

The first is your own memory. Six weeks later, when a lender queries the source of a deposit, you need to know what the client said about it and when.

The second is evidence. Your file has to show the basis on which advice was given: the circumstances gathered, the needs identified, the options considered, and why the recommendation followed. Under MCOB and your record-keeping obligations, that trail has to be intelligible to someone who was not in the room.

A transcript alone does not do the second job. Nobody wants to read nine thousand words of conversation to find out why an offset was ruled out. What you want is a structured note that a file checker can follow, with the raw record available underneath if a dispute ever needs it.

Consent, and how to ask for it properly

You are recording a conversation about someone's income, debts, health and family. Handle it accordingly.

Ask at the start, on the recording, before the substantive discussion begins. Say what is being recorded, why, how long you keep it, and that they can say no. Keep it short and unembarrassed; almost nobody refuses when the reason is explained plainly.

If they do refuse, you need a working alternative that does not make them feel awkward, and the adviser needs to be comfortable running the meeting the old way.

Also update your privacy notice. A recording and its transcript are personal data, and the fact-find conversation will usually include special category data the moment health comes up in a protection discussion. Your retention schedule should say how long recordings are kept as distinct from how long the summary is kept, and in most firms the recording should have the shorter life.

What these tools get right and wrong

Transcription of a clear one-to-one conversation in a quiet room is generally good. Two people talking over each other, a client on speakerphone in a car, or a room with three family members is measurably worse.

The specific failure to watch for is not gibberish, which you would spot. It is confident precision that is slightly wrong. Numbers are the danger zone. "Four thousand two hundred" and "forty-two hundred" and "four two hundred" are easy for a machine to render inconsistently, and a figure in a summary looks authoritative regardless of whether it was heard correctly.

Names of lenders and products are a second weak spot, because a model asked to summarise will happily normalise an unfamiliar name into a familiar one.

Summaries also tend to smooth. A client who hesitated, contradicted themselves and then settled on an answer will appear in the summary as having simply given the answer. That hesitation was information. If it mattered, the adviser has to put it back.

Making the review real

The point of failure in every firm that adopts this is the review step, because a well-written summary invites agreement.

A few things that help.

• Review the same day, while the meeting is still in your head. A note reviewed a week later is a note rubber-stamped.
• Check the numbers against the source document rather than against the transcript. Income comes from the payslip, not from what was said about the payslip.
• Read the summary asking what is missing, not whether what is there is correct. Omissions are the common defect, and they are invisible if you read for accuracy alone.
• Make the adviser edit rather than approve. A workflow where you have to type something means you engaged with it. A tick box means very little.
• Keep the version the adviser produced as the file version, clearly marked as adviser-reviewed, with a timestamp and a name against it.

That last point is what turns this from a productivity tool into something defensible. If you are ever asked to show that a human confirmed the note, you want a record that answers it in one screen.

What should never come from the summary

Draw the line at anything that constitutes a judgement.

The needs and priorities assessment, the reason a product was recommended, the treatment of anything the client said that suggests vulnerability, and the affordability conclusion all belong to the adviser. A tool can remind you that a topic was discussed. It cannot decide what the discussion meant, and a suitability rationale generated wholesale from a transcript is a genuine liability, because it will read plausibly whether or not it reflects your actual reasoning.

Vulnerability deserves a specific mention. Indicators surface in fact-find meetings in passing and in ambiguous language: a comment about a health scare, a mention of a recent bereavement, confusion about a term explained twice. Summarisers compress exactly this kind of aside out of existence. If your firm relies on the summary to capture it, it will be missed.

Deciding whether it is worth it

Time one thing before and after: the gap between the appointment ending and a complete note existing on the file. If that gap goes from two days to two hours, you have gained something that shows up in case progression, file quality and how quickly you can answer a lender.

If the gap stays the same because notes now sit unreviewed instead of unwritten, you have bought a queue rather than a saving, and it is better to know that in month one than in year two.