Fix the process before buying the tools

A familiar script: the company resolves to embrace AI, so there is a project, a shortlist, a purchase, a training day and a sign-off. Six months later the tool is live. And then? And then everyone keeps working exactly as before — there is simply one more website that needs a password.

The problem is not the tool. It's the order of operations. Most companies run AI transformation the way they once ran an ERP rollout, but AI is a different animal: it doesn't move your existing process onto a computer, it forces you to re-answer "who should do this task — and, come to think of it, should this task exist at all". Get the order wrong and it doesn't much matter what you buy.

The order we recommend has four steps. The first three have nothing to do with tools.

Step one: draw the process

It sounds banal, but in our observation most SMEs have never once drawn their core process end to end. Where an order comes from, whose hands it passes through, where it waits for approval, where it gets reworked, who ultimately answers for the result — those details live in the habits of a few veteran staff, each with a slightly different version.

Drawing it out is often the single most valuable step of the whole transformation. Before any AI is involved, the waste becomes visible: duplicate checks, spreadsheets whose origin nobody remembers, signatures that exist purely "just in case". Some steps can be deleted halfway through the drawing.

Step two: find the waiting and the repetition

With the process on the wall, resist the urge to ask "where can we use AI". Look for two things first: waiting, and repetition.

If a quotation takes three days door to door, the time someone is actually working on it is usually under an hour; the rest is waiting — for data, for approval, for somebody to be free. Repetition is blunter: the same paragraph typed twenty times a week, the same numbers copied into three systems. The two wastes have different cures. Waiting is fixed by rearranging the process; repetition is where AI belongs. Keep them straight and you stop wandering the house with a hammer, looking for nails.

Step three: re-divide the work — and dare to subtract

Only now comes the division of labour: which steps AI drafts and a person approves, which stay fully human, and which are simply cancelled. The third category is the hardest to say out loud and the most valuable. Plenty of steps exist for one reason only: they always have. A transformation is one of the few moments when "do we still need this?" can be asked out loud with no hard feelings. Don't waste it.

Step four: now choose the tools

After the first three steps, selection becomes surprisingly easy, because you finally know what you're solving. If anything, we'd caution against falling in love with any tool. Tools are consumables — replacing them every two or three years is normal. The redrawn process and the working habits your team builds are the assets. Give the assets the larger share of your budget and your affection.

Three field notes

• Pilot where it hurts but can't kill. Pain supplies motivation; survivability gives you licence to fail.
• Two to four weeks per cycle. Quantify what you can. What you can't, judge by one honest question to the people using it: do you want to keep it next week? If not, cut it.
• Track where the saved time goes. Time saved that simply evaporates is the most common and most regrettable ending. Saving time is only the first half; reinvesting it in customers and product is where returns actually happen.

Process is the skeleton; tools are the clothes. Clothes can be changed often — the skeleton has to stand straight first.

Fix the process, then buy the tools. In the right order, almost any reasonable tool will do; in the wrong order, the most expensive one merely gives old habits a newer interface.