AI Consulting

Whether we are the right people, and when we are not.

The shortest chapter here. Who we work well with, who we turn down, and the three behaviours that end an engagement on our side rather than yours.

We work well together when
  • You can name the one job in two sentences
  • You have hit the edge of spreadsheets and off-the-shelf tools
  • You want to know what is true more than you want to be right
  • Somebody in the business can decide, usually you
We say no when
  • You want a tool you already bought validated
  • You want AI so the business can say it has AI
  • The number does not clear the price
  • You need it next week
Both columns are real and we have used the right-hand one. The detail on each is below.

Where this works well

Four situations. If you are in one of them, a conversation is probably worth half an hour of your time.

  • You can name the job.Not “we need to modernise”. One task, one person, one recurring frustration you could describe to us in two sentences.
  • You have run past what spreadsheets and off-the-shelf tools can carry. Not because they are beneath you, but because you have genuinely hit their edge and the workarounds have workarounds.
  • You want to know what is true more than you want to be told you are right. We will sometimes say the thing you were hoping we would not.
  • Somebody in the business can decide. Usually you. A diagnosis that has to travel through four people before anything can change is a slower and more expensive thing, and we would rather say that at the start.

Where we say no

These are not warnings. They are refusals, and we have made all of them.

  • You want a tool you have already bought to be validated. If we look and it is wrong for you, we will say so. If that answer is not welcome, the engagement has no useful outcome and we should not take your money to reach it.
  • You want AI so the business can say it has AI. With no problem underneath, there is no project. There may be a marketing exercise, and we are not the right studio for it.
  • The number does not clear the price. If the job costs less per year than the work costs to do, we will tell you and point you at a subscription. The calculator two chapters back does this without you having to talk to us at all, which is the point of it.
  • You need it next week. Diagnosis before build is the whole method. A studio that skips it to hit your date is selling you the thing you asked for rather than the thing you need, and you can find one of those easily.

What ends it on our side

Three behaviours end an engagement, and we would rather you read them here than discover them:

  • Repeated renegotiation of an agreed price. Once is a conversation. Repeatedly is a statement about how the work is valued, and we take it as one.
  • Unlimited scope inside a fixed price. Scope changes get repriced. They do not get absorbed quietly, and they do not get absorbed loudly either.
  • Disrespect toward the work or the people doing it. No contract size changes this.
None of the three is about money. All three are about whether the next six months will be worth either side’s time, which is a question better answered in week one than in month five.

Where we are, and why that is fine

We should be straight about our own position, because you are going to weigh it anyway.

We are a small studio based in India, working with owner-led businesses here, in the Gulf and in the United States. Our published proof is operational rather than financial: 501 of 501 price changes in a client system carrying a written reason, counted by database query on 26 August 2026, and a similar count of status updates that used to be phone calls.

What we do not have is a measured rupee of client revenue. We have said that plainly since we withdrew sixoutcome figures from this site — three of them contradicted by our own forensic audit of a client database, and three that had no instrument behind them at all and were therefore unmeasurable rather than merely unmeasured.

That is a genuine weakness and you should price it in. A studio that has published a number it later had to remove develops a strong preference for numbers it can point at, which is why the rest of this guide reads the way it does.

If that lands, the next three chapters are the mechanics: how the first project gets chosen, what the diagnosis is made of, and what it costs.

Where these numbers came from

Every figure on this page, with what it is and where it is from. If a number is illustrative rather than measured, it says so here and it says so in the text.

  1. 501 of 501Price changes in a client system carrying a written reason. An invariant rather than an average: the denominator only rises, and one exception would break it.Counted by us with a database query against the live client system on 26 August 2026. Recorded in lib/proof.ts in this site source.We counted it
  2. 6 figures withdrawnClient outcome numbers we published and then removed, three of them contradicted by our own audit and three with no way to measure them at all.Our own forensic audit of a client production database, 15 August 2026, recorded in lib/proof.ts in this site source.We counted it