The mistake almost everyone makes
You start with the task that hurts most.
It makes sense — it is the one keeping you up, so it feels urgent. It is also, almost always, the task that only you can do. And a task only you can do runs on knowledge nobody ever wrote down.
So the project starts with an excavation: getting what is in your head into a form a system can use. That takes weeks, nothing works in the meantime, and by the time anything ships the team has quietly gone back to doing it by hand.
Start somewhere else.
Why the first one matters so much
It is not really about the task. It is about proving to your team that these projects finish. A first attempt that visibly works buys you permission for the second. A first attempt that drags kills the second before it starts.
The four tests
Take a candidate task and check it against these. You want four yeses.
Does it happen every day?
Frequency beats importance. A small task forty times a week beats a big one twice a month, because the savings compound and you find out fast whether it works.
Does the answer have the same shape?
Same shape, not same content. Quoting can pull twenty different numbers and still pass — as long as the rules for combining them are the same.
Does being late cost you a customer?
This is the one that turns hours into money. A reply eleven hours later is rarely just late — it is usually a customer who already called someone else.
Could someone else do it?
You want a yes. A task anyone could do given the information is a task a system can take over cleanly.
If a candidate fails two or more, pick another one. There is always another one, and forcing the wrong task is how first attempts die.
The four-week plan
Nothing here disrupts what you are already doing. The system runs alongside until it earns its place.
| Week | What happens | Who does it |
|---|---|---|
| 1 · Watch | Collect every instance of the task for a week, with the real input and the real output. Twenty examples is plenty. | Your team, 2 min a day |
| 2 · Draft | The system produces answers and nobody sends them. Every output gets marked: correct, close, wrong. | You, 15 min a day |
| 3 · Assist | The system drafts, a person approves, the approved answer goes out. Real customers. | Whoever handles it now |
| 4 · Run | Routine cases handled alone. Anything unusual gets flagged and waits. | Nobody, until flagged |
Week 1 is not wasted time
It looks like doing nothing. What it actually does is surface the exceptions — and everyone describes their process as if there are none. There always are, and finding them in week 1 costs nothing.
Week 2 decides everything
This is the week people skip, and it is the one that saves the project.
- Right eight times out of ten by Friday → it will be nineteen out of twenty within a month
- Right three times out of ten → the task failed test 2. You picked the wrong one
- You cannot tell → you are not collecting enough examples. Twenty minimum
Better to learn that now than after launch, in front of customers.
What should never be on this list
- The conversation with somebody who is angry
- The decision that changes what you charge
- The exception you make because you know the customer
- Anything where being wrong is expensive and being slow is not
If your candidate is one of these, it will fail test 2 or 3 anyway. The point of handing over the routine is to have time for exactly these.
After the first one
The second task takes about half as long. Not because the system is smarter, but because you are: you know what to collect in week 1, you know what «good enough» looks like in week 2, and your team already trusts the process.
50%
less time on the second task, and less again on the third. That compounding is the real reason to start small.
How to know it worked
One number, measured before and after: how many times that task reaches a person. Not satisfaction, not time saved in theory. How many times somebody had to touch it.
If it does not fall by at least half in month one, stop and look at why. Usually the answer is that the task had more exceptions than anyone admitted, and the honest move is to narrow the scope rather than push harder.
Want us to pick yours?
We do a free Process Diagnosis: we look at how your team actually works, run these four tests over your real processes, and tell you the one to start with and why. No cost, no commitment, nothing to install.