← All posts
FAQProductsAI

Our clients keep asking us for advice. Can software answer?

Monday 9 November 2026 · 3 min read · Rishi Gupta

A business owner asked me this on a call this week. His team finds companies for people who want to buy one, and hands the buyer a report on each business they find. Every time, the buyer comes back with questions: how would I structure this deal, what is it worth, what should worry me here. Answering those is most of his week and none of it is on an invoice. So he asked whether we could put something in front of clients that answers instead of him.

The question under the question

He was imagining a sort of virtual director his clients could open up and ask what it thought of a deal, and get a verdict back. My answer was that part of it is absolutely buildable, and the work worth paying for is deciding which part, because getting that line wrong is how these projects end up embarrassing somebody in front of a client.

Split your last twenty questions into two piles

This is the whole exercise and it takes about twenty minutes. Write down the last twenty questions clients actually asked you. Not the ones you imagine they ask. The real ones, in their words. Then put each one into one of two piles.

  • pile one: your answer is the same every time, and you are so bored of saying it that you have a stock phrase for it
  • pile two: your answer depends on something you have to weigh up, and two clients with the same question get different answers

Pile one is a product. It is also usually bigger than people expect, somewhere around two thirds of the list, and it is the pile quietly eating your week. Pile two is you. It should stay you, it is the reason they are paying, and the moment you let software guess at pile two you have handed your judgement to something with no stake in being right.

Why the general tool is not enough

He is already doing a version of this. They paste the call recording, the accounts and the buyer's criteria into a general AI tool and it writes them a report. It works, and he said himself that the next step is making the report better. The trouble starts when a client asks it something sitting outside what it was given, because it answers anyway, in the same confident voice it uses when it is right.

The fix is not a cleverer model. It is boundaries. The thing answers from your documents, your criteria and your own past reports, and when a question falls outside them it says so and points at a human. Being told to go and ask someone is not a failure of the software. It is the most valuable sentence in it, because it is the one that protects every answer that came before.

The advice you give away for free, in the same words, week after week, is the first thing worth building. Write the list of twenty this week. It costs you one coffee and it tells you what you are actually selling.

Got a version of this problem? Twenty minutes minimum, seven questions, and a one-page Bottleneck Blueprint within 48 hours. If it's not a fit, we'll say so.

Book a call → Not ready to talk? Take the eight-question AI assessment → Questions about cost, data and ownership? Thirty-five straight answers →
Keep reading
Is our data used to train the AI?
Thursday 12 March 2026 · 3 min
What is the difference between automation and AI?
Thursday 22 January 2026 · 3 min
Which AI do you actually use, and does it matter?
Thursday 13 November 2025 · 3 min