The three-week death
Week one, everybody signs up, because the demo was genuinely impressive. Week two, two people still open it. Week three, it is a line on the card statement that nobody wants to be the one to cancel.
Nobody decided to stop. It just stopped. And the conclusion drawn afterwards is almost always the wrong one: AI does not work for a business like ours.
22%
of companies that experiment with AI get real value from it. The rest bought before they looked.
What actually went wrong
The tool was dropped on top of a process nobody had ever written down. It automated the confusion, faster. Everyone felt it, nobody could name it, so it quietly died.
Step 1 — Pick one task
Not the biggest problem in the business. The most repetitive one. Those are rarely the same thing, and starting with the big one is how projects stall.
| Good first task | Bad first task |
|---|---|
| Happens every week | Happens twice a year |
| Same shape every time | Different every time |
| Annoys more than one person | Only bothers you |
| You can describe it in a sentence | It needs a meeting to explain |
If a task fails two of these tests, put it aside. Not forever — just not first. The first one has to work, because the first one is what buys you permission to do the second.
Choose the boring task. Boring means repeatable, and repeatable is where the money is.
Step 2 — Write it down as it really is
One page. Actual steps, in the order they actually happen — including the ugly ones. Especially the ugly ones.
- What starts it? A call, a form, a message, somebody remembering
- Every step after that, in order, one line each
- Who does it, and what they need open in front of them to do it
- Where it waits — the gaps where nothing happens are the real cost
- How you know it finished
Write the real version
Not the version you would describe to a customer. The one with the spreadsheet somebody keeps on their desktop, the WhatsApp group, and the step that only works because Marta remembers to check on Fridays.
Most owners have never seen their own process on one page. It is usually a surprise.
Step 3 — Cut, before you buy
This is the step that pays for the whole exercise, and it costs nothing. Three kinds of step can almost always go:
The ghost step
It exists because of somebody who left. Nobody has questioned it since.
The double entry
The same information typed into a second place by hand.
The just-in-case check
A review of work that has not been wrong in two years.
Cut those and re-read the page. A good number of businesses reach this point and discover the process is now short enough that they no longer need the tool at all — which is a perfectly good outcome, and a free one.
Automating a step you should have deleted is paying monthly for a mistake.
Step 4 — Now go shopping
With one page in your hand you are a different buyer. You stop asking «what can it do» and start asking «can it do these seven steps».
| Ask the vendor | What you are really testing |
|---|---|
| Show me these exact steps | Whether it fits your work, not the demo's |
| What happens when it gets it wrong? | Whether a human can catch it in time |
| Who on my team touches this daily? | Whether adoption has an owner |
| What does month two look like? | Whether anyone has thought past the sale |
The stickiness test
A tool sticks when using it is faster than the old way on the first day — not after training, not once everyone adjusts. If day one is slower, week three is silence.
1 page
is the whole prerequisite. Everything expensive comes after it.
One page beats one subscription
Pick the task that repeats and annoys. Write it on one page as it truly happens. Delete the ghosts, the double entries and the just-in-case checks. Then, and only then, go and look at tools.