Five steps from audit to production. And an honest no.

Most descriptions of this kind of work are vague. Vagueness is easier to sell. This is the opposite. Five steps, each with a deadline. And the places where we stop.

Step 1 · The mapping (free, at most 3 hours of your time)

We sit down for 45 minutes with the people who do the work. Not management alone. Whoever sits with the invoices knows things about the invoices that nobody else does.

We ask three things about every task:

• How often it happens.
• How long it takes each time.
• Where the time goes: finding, typing or waiting.

The third question decides everything. If the time goes on waiting for a colleague, we cannot help. If it goes on finding the same field in the same contract, we can.

Documents are the clearest example. But the mapping is also where the rest comes up: calls you do not get to, the inbox, quotes, data entry between systems, follow-up, invoicing and reminders, reports. Tell us what repeats. Then we find the piece that saves the most.

The whole step costs you at most 3 hours. The rest of the time is ours.

Step 2 · The audit in writing (within 5 working days)

You get a document. Not a presentation. A document you can read alone and hand to someone who was not in the room.

It contains:

• Every task, with your own figures for how often and how long.
• How much of it a machine can realistically carry, and why not the rest.
• What the task costs you a year today, and what can be recovered.
• One recommendation on where to start.
• One fixed price for the first piece, or an honest no.

We deduct review time before we show you the number. Someone still reads the draft. We assume 15% of the saved time goes back into that review, exactly as in our calculator. A savings figure without that deduction is not honest.

The report is yours. Even if you never hear from us again. You can take it to another supplier or build it yourselves.

Step 3 · The agreement, before anything is built

Data processing agreement first. Not as a formality at the end. As step three of five, because it decides whether the system may go into production at all.

At the same time we write down:

• Which data the system may touch, and which it may not.
• That processing happens in the EU.
• What is logged, so you can see afterwards what was sent where.
• What a human must approve, and what may run on its own.

No data moves before the agreement is signed.

Step 4 · One workflow (2–4 weeks after the agreement)

One. Not three.

It is built on what you have. If an integration to your system exists, we use it. If it does not, we work in a mailbox and a folder. It is not elegant. It works, and it is cheaper than replacing a system your people already know.

Every output lands in an approval queue. A list where a human sees what was produced and presses yes or corrects it. The format is boring on purpose. You should be able to empty it with a cup of coffee in hand.

We also build the part that refuses. If the system cannot decide something, it flags it instead of guessing. A price it cannot find. A name spelled three ways. Those flags are not a fault. They are the system working.

Step 5 · We measure, you decide (4 weeks in production)

After four weeks in production we measure against the figures from the audit. Same task, same unit, same arithmetic. You get one number.

Then there are three outcomes:

• It works. We take the next workflow.
• It partly works. We adjust and measure again.
• It does not work. We say so, and we stop. There is no proposal for a phase two.

In MIT's 2025 review of more than 300 AI initiatives, 95% showed no measurable effect on profit and loss. Very few of them had a number to measure against. That is why step five exists.

When we say no

It happens. It is just as important to write down.

• When the task does not repeat enough. Then there are not enough hours to recover.
• When the time goes on waiting, not on working. Then the problem is something else, and we are not the answer.
• When nobody can own it afterwards. An automation nobody looks at stops working.
• When we cannot see how we would measure it. If we cannot prove the value, neither can you.

An honest no in week one is cheaper for you than a half-finished system in month six. It is also the only reason a yes from us means anything.

The next step

You tell us what repeats. We map it for free and write it down within a week: hours, kroner, where to start, and one fixed price — or an honest no. It takes at most 3 hours of your time.

Get my free AI audit