How do I choose between consultants whose proposals look the same?

Stop comparing the documents, because AI now helps write most of them and they will keep looking alike. Ask every consultant, in the same words, to bring one past report with the client's name removed, and spend twenty minutes asking how it was built: where the hypothesis came from, why those questions, what the client said that they did not believe, what they threw away, and what they are still unsure about. Write down after each meeting which answers named real meetings, files and changed minds, and hire from that list.

What is a hypothesis in a business diagnostic?

It is the consultant's best guess about what is really wrong in one area of the business, written before the data comes in, so the data can prove or disprove it. A good one is one sentence about how the business works and one sentence about what exists and what is missing. In one restaurant diagnostic the pricing hypothesis said, in short, that dish prices were set on ingredient cost alone, so delivery commission, packaging and competitors' prices never entered the price. Every research question after it was written to prove or disprove that sentence.

Should I worry if a consultant uses AI to write the report?

No. What matters is where the AI stops and who checks it. Ask the consultant to take one fact from the sample report and show you where it came from, a document or a meeting note. If every fact can be traced back to a source and a person checked it, AI only made the writing faster. If nobody can say where a number came from, that is the thing to worry about, with or without AI.

What should I do if a consultant will not share a past report?

Accept it, since many reports are confidential, and ask for one of two things instead. The first is the list of questions or the data request from a past engagement, which shows how they think without showing any client's numbers. The second is a live scoping of your own problem: two or three possible explanations, the one they would bet on and why, the data they would need, who in your company has it, and what that data will not show.

When will I know if a consultant's advice was right?

Usually only when you execute it, which can be months after the fee is paid. That is why the first meeting matters more than the proposal. You can shorten the wait by asking, before you sign, what the first visible sign of progress should be and by when, and writing it down together. Then you have a checkpoint much earlier than the end result.

Five Consultants, Five Identical Proposals (How to Tell Who Did the Thinking)

You met five consultants, and their proposals and sample reports read the same, because AI helped write all of them. Here is a twenty-minute test for the first meeting: five questions about one past report, and what a good and a weak answer sound like.

You decided you need outside help, and now there are five proposals on your desk. You met all five consultants and all five were pleasant. Each one sent a sample report from a past client, and the reports read the same: same headings, similar recommendations, same confident tone. When you asked how they work, all five said the same sentence: "We use AI for efficiency."

So which one do you hire?

The documents will not tell you. A junior consultant with one prompt and a senior who worked the problem for weeks can now send you reports that look the same, and the junior may not even understand what his own report says. What is left is the meeting. I sit on the consultant's side of that table, so here is the test I would want a founder to run on me, and on the other four.

In an earlier post I wrote about five questions to ask before you sign, about past failures and how success gets measured. Those are about the engagement. The five below are about one finished report, and they take twenty minutes.

"We use AI" tells you nothing

All five are telling the truth. In our own diagnostics AI writes the first draft of the report. Every fact goes into a register first, with its source, the report is written from that register, and then a consultant checks it. Hiding AI would not make a consultant better, and using it does not make him worse.

What AI cannot do is the deciding. Before any report is written, someone has to take your words from the first meeting and decide what the real problem probably is, which questions will prove or disprove it, and which answers to trust. With a wrong decision at the start, AI is rubbish in, rubbish out, and the rubbish comes nicely formatted.

None of those decisions show in the finished report. So you have to ask about them.

The one thing to listen for

You cannot judge a consultant's method, and you do not need to. Anyone can hear where an answer comes from.

An answer from the work names something: a meeting, a file, a person, a moment when the consultant changed his mind. An answer from the page repeats the report in other words, a bit slower.

Keep that distinction in your head for the whole meeting. The five questions are only ways of making it show.

How to set up the meeting

When you book the meeting, ask each consultant to bring one finished report from a past client, with the client's name removed. Tell them you want twenty minutes to ask how it was built. Ask all five the same way, so you compare the same test.

In the meeting, keep the report closed on the table. You are not there to read it. You are there to hear what is behind it.

A report that came from one prompt has nothing behind it. Nothing was dropped on the way, because there was no way. A report someone worked on leaves a pile: a hypothesis that looked right in week one and was dead by week two, an AI answer thrown out because it did not match what the owner said, eight recommendations cut down to three, a number redone by hand because it felt wrong. None of that is in the report. All of it is in the consultant's head, and that is what the five questions bring out.

Two identical final reports side by side. Behind one is nothing, behind the other a pile of crossed-out hypotheses, a rejected AI answer, cut recommendations and a number recalculated by hand.

The five questions, and what the answers sound like

A closed report on a table surrounded by five cards: where the hypothesis came from, why these questions, what was not believed, what was thrown away, and what is still not sure.

1. "Where did the hypothesis come from?"

A hypothesis is the consultant's best guess about what is really wrong, written down before the data comes in, so the data can prove it or kill it. Ask why this one and not another, and which other explanations were on the table.

A good answer starts with the owner's own sentence from the first meeting and shows how it was narrowed down. It sounds like this: "The owner said sales. We also looked at pricing and at delivery times, and delivery dropped out after we saw the order book." A choice was made, and you can hear it.

A weak answer repeats the hypothesis from page two, slower, or says "the data showed it". If no other explanation was ever on the table, nobody chose this one.

2. "Why did you ask for these things, and not others?"

Every diagnostic asks the client for some data and not for other data. Ask what the consultant decided not to ask for, and which questions turned out to be a waste.

A good answer ties each question back to the hypothesis and admits that a few were wasted. Someone who designed the questions knows which ones did not help. You can check one more thing yourself in a minute: are some of the questions about the outside, about how comparable businesses do it? If every question looks only inside the company, the report can tell you what it found in the numbers, but not whether that is good or bad, because there is nothing to compare it with.

A weak answer is "they were all useful". Someone who got the questions from a model thinks they were all good.

3. "What did the client tell you that you did not believe?"

This is my favourite one. In a first meeting with a window and door manufacturer, the owner told me the glass step was his bottleneck, and he wanted to buy a machine for it. The notes from that meeting say: "We need to compare the PVC production versus the glass capacity to understand if this really is a bottleneck." The owner's claim became something to check before it could become a finding.

A good answer sounds like that: a claim the client made, and how it was checked with the company's own production or sales data. Asking the owner again only gives you his answer again.

A weak answer is "the client knows the business best". A consultant who believed everything his last client said will write down everything you say, and you will pay a fee to read your own opinion back, nicely formatted.

4. "What did you throw away?"

Ask for one answer AI gave that was rejected, one recommendation that was in the draft and is not in the report, and one number the consultant checked by hand.

A good answer has specifics. On a restaurant business we started one area with nine hypotheses. One stayed. The other eight turned out to be ways of proving it, so they became questions instead. That is what a pile looks like when someone describes it.

A weak answer is "nothing, it all held up". If nothing was thrown away, nothing was judged.

5. "What are you still not sure about?"

A good answer points to the pages where the report could be wrong. In that same first meeting with the window manufacturer, the owner said his cost overview "probably won't be 100% accurate". That went into my notes on day one, before I had seen a single cost figure. In our diagnostics, data that cannot be found is marked Not available and goes into the report as a limitation, so the owner can see where to be careful.

A weak answer is equally confident on every page, and so is the person handing it over. Saying "I am not sure" in a sales meeting feels like a risk to a consultant. The one who takes that risk is showing you where to check his work.

Good and weak answers side by side

Question	A good answer sounds like	A weak answer sounds like
Where did the hypothesis come from?	The owner's sentence, the other explanations, what killed them	The hypothesis from page two, said again
Why these questions?	Which ones were wasted, and some that look outside the company	"They were all useful"
What did the client say that you did not believe?	One claim, and the data used to check it	"The client knows the business best"
What did you throw away?	A rejected AI answer, a cut recommendation, a number redone	"Nothing, it all held up"
What are you still not sure about?	The pages where the report could be wrong	The same confidence on every page

When they cannot show you a report

Confidentiality is a fair reason not to bring a past report. Then ask the consultant to do the same thing live, on your problem.

A good consultant can scope on the spot. He asks before he answers. He gives two or three possibilities for what is going on and says which one he would bet on, and why. He says what data he will need, who in your company probably has it, and what that data will not be able to show. No laptop is needed for any of it.

A weak one presents the method: the phases and the slide with the arrows.

You may also hear "we do not give our thinking away before the contract". Consultants were taught not to give the diagnosis away in the sales meeting, and it made sense when the report was the scarce thing. Today the report is the cheap part, and a consultant who holds back his thinking in the first meeting looks exactly like one who has none.

Can they fake it with AI?

On paper, yes. Live, for twenty minutes, I honestly don't think so. To invent a convincing pile of dropped hypotheses and rejected answers, a consultant has to understand the problem well enough to know what a wrong answer looks like. At that point he is not faking any more.

Before your next meeting

Ask for one past report, client name removed, and say you want twenty minutes on how it was built.
Keep it closed and ask the five questions in the same order to every consultant.
Listen for names: a meeting, a file, a person, a changed mind. Not the report said again.
If there is no report, give them your problem and ask them to scope it live.
Write two lines after each meeting on which answers came from the work, before the five start to blur together.

The only test you get before you pay

A recruiter who hires the wrong person finds out in a week, and the cost is one candidate. You find out if the advice was right only when you execute it. By then the fee is paid, the months are gone, and there is no second interview.

So the first meeting is the only test you get before you pay.

If you want to see where your business stands before you talk to any consultant, the free business health check takes under five minutes and scores you across clarity, structure, growth and AI.