Not every company needs AI. Let's find out whether yours does

Not every company should be deploying AI this year, and some of the ones that should are not ready yet. That sentence costs us work, and we still open every conversation with it.

Three conditions have to hold before a deployment makes sense. You can check all three yourself, in half an hour, without talking to anybody.

Condition one: enough volume

Automation pays for repetition, not for difficulty.

A process that happens once a quarter can be the most irritating thing in your company and still never pay back. There is nothing to recover the setup from.

Rough threshold: below a hundred repetitions a month, or twenty leads, payback is measured in years. Above four hundred, in months.

Run it on one process: how many times a month, how many minutes per repetition, what an hour of that person's time costs. Under $250 a month, leave it.

Condition two: a process that stands still

Every change to a process means recalibrating the system, and recalibration costs.

If your offer changes monthly, procedures are mid-rebuild, or the team is still arguing about how something should work, a deployment will set a temporary version in concrete.

The point is not that the company should stand still. The point is that the one process you automate should be stable enough that a description written today is still true in three months.

Condition three: data somebody can read

This is the condition that wrecks the most projects and gets quoted the least.

The system answers from what you give it. If your pricing exists in three versions, your procedures live in people's heads, and the freshest document is attached to an email from last year, you will get faster-produced confusion.

The test: take the ten questions customers ask most. For how many of them does one document exist that everyone in the company agrees is the current one?

Below seven, you start with tidying rather than deploying. That is good news, because the tidying helps your people regardless.

When we say no

Three situations where we advise against working together even though we could sell it.

Volume under the threshold. We say so on the first call, because three numbers make it obvious.

An organisational problem dressed as a technical one. If two people describe the same process differently, no system resolves that. It just cements one of their versions.

An expectation that the system will make decisions. Hiring, performance reviews, settling disputed claims. Regulation aside, the cost of one bad call there outweighs the savings from a hundred good ones.

What to do if the conditions do not hold

Do not buy. It sounds obvious and it is the most broken rule in this industry.

At low volume, stay with a decent template and come back when you grow. With an unstable process, stabilise it first, or you pay twice. With disordered documents, start by deciding which file is authoritative, and do it whether or not AI is ever involved.

Each of those three improves the company on its own. None of them needs a supplier.

What the call looks like if the conditions do hold

Forty-five minutes and one of three answers.

You talk to the owner rather than a salesperson, and it is the same person who runs the project afterwards. We start with your numbers, not with a demo.

The call ends with a price range broken into lines: integrations, tidying data, build, testing, running costs. Agreed before signing, never during.

The first scope covers one process, not a department. After it you have data and you decide whether to continue. Your code sits in your repository from day one, so you can leave and take all of it. The warranty covers repairs at no extra charge.

The consultation is free, including when it ends with us telling you not to do it.