What actually happens, from the first call to a working thing.

The call: thirty minutes, free

You bring one painful task. Not a strategy, not a wish list. The job you already had in mind when you started reading this guide.

We spend the half hour on four things: what the job is, who does it and how often, what it costs you by the arithmetic in chapter two, and which of the five tests in the previous chapter it passes or fails.

You leave with one of three answers, and we say which on the call rather than in a document a week later:

• This is worth building, and here is roughly what it would take.
• This is real but it is not an AI problem. Usually a process that needs writing down, a step that should stop happening, or a subscription you can buy this afternoon without us.
• The number does not clear the cost. The most common answer, and the one the calculator lets you reach without ever speaking to us.

There is no proposal document, no deck, and no follow-up sequence. If the answer is no, the honest thing is for the conversation to end cleanly.

Why there is no paid audit stage

This is the part that differs most from how the market is usually structured, so it is worth explaining rather than asserting.

The common shape is a priced readiness assessment first, then a separately priced roadmap, then a pilot, then an implementation. One published example we read on 27 August 2026 lists a readiness assessment at 2,000 to 8,000 dollars over two to four weeks, and a strategy and roadmap stage at 8,000 to 25,000 dollars over four to eight weeks, before anything is built.

That structure is not dishonest and there are contexts where it is right. Our objection to it for a business your size is narrower: it means the first thing you buy is a document, and a document cannot be wrong in a way you can detect. If the assessment says the wrong thing, you find out one stage and one invoice later.

So the diagnosis is free and short, and the first thing you pay for is a thing that runs. If our diagnosis was wrong, it becomes obvious inside weeks rather than inside quarters, and it becomes obvious to you rather than to us.

A free diagnosis is not generosity. It is the arrangement that puts our judgement at risk instead of yours.

The first fix, week by week

If the answer on the call is yes and you decide to go ahead, most first fixes ship in four to eight weeks. Roughly:

Week one: counting

Before anything is built, we establish the number as it stands today. How many times the job happens, how long it takes, how often it goes wrong. Sometimes this is a database query. Often, in a small business, it is two weeks of tally marks, and that is a perfectly good instrument.

This week is not preamble. It is the only chance to measure the before, and every chapter of this guide that talks about telling whether something worked depends on it existing.

Weeks two to five: building the narrow thing

One job. The one from the call. Built so that a person stays in the loop wherever a wrong output would reach a customer or an account, which is usually most places.

You see it working before it is finished, on your own data, not on a demo set. That is the same test we told you to apply to anyone else pitching you, in chapter one.

Weeks five to eight: switching the old way off

The part that is skipped most often and matters most. A named person turns the old route off. If the old route stays open, the new one is optional, and optional loses.

For context on speed rather than as a promise about you: in the MIT study discussed in chapter three, the best performers reported around 90 days from pilot to full implementation, against nine months or longer at large enterprises. The advantage of being small is real, and it is mostly about how few people have to agree.

The number, and who counts it

At the end you get the same count as week one, run the same way. That is the whole reporting apparatus, and it is deliberately unimpressive.

Two things we will not do. We will not present a figure we cannot show you the query for. And we will not report a percentage improvement against a baseline that was estimated rather than counted — we have published figures we later had to withdraw, and the chapter on whether we are the right people says so out loud.

If the number did not move, you will hear that from us. It is a worse conversation and a better business.

What happens after

Nothing automatic. There is a monthly arrangement where we re-run your number by hand, read it to you, and cover up to four hours of fixes, which do not roll over. It only makes sense if the first fix actually paid. No lock-in, no minimum term. If it did not pay, the right outcome is that you stop.

What all of it costs, in both currencies, is the last chapter.