Method

One day on site. I follow one run from intake to invoice and write down who types what into which system, where things sit and wait, where work is redone, who has to chase whom. No pinboard, no workshop round. The people who run the process should be able to keep working while I do it.

I measure with timestamps from the systems where there are any, and with a watch where there are none. Anything estimated I write down as estimated, because in the meeting with the managing director every figure gets checked.

Result

The results briefing: the drawing of the process, the legs with their times, and the proposals that follow, each with its effort, its expected effect and its price.

The second sheet is the same picture, nothing moved, only marked up. I trace the critical path heavy, break out the waiting time and number every exception loop.

Nothing gets solved here yet. Without a measurement, the leg that gets automated is the most visible one. The most expensive one is almost always another.

The measurement is not a separate engagement and costs nothing extra. It comes out of what I wrote down on the day.

Result

In the briefing the legs are ordered by cost: which one takes how many hours, and roughly what it costs to get rid of it. You choose from it. Or you leave it, and still hold a measured drawing of your own process.

One workflow from prototype to production, at a fixed price. I work in your systems, with write access, and deliver something that runs at short intervals. If an assumption turns out wrong in week three, that is a result.

A rebuild includes what is missing from most offers. Evals, meaning a measurement of how often the model is wrong and what happens then. A runbook describing what to do when it jams, written so somebody can read it at 7am. And a person at your end I hand over to, who carries it on.

After the handover everything sits in your repository. Whether you call me again depends on whether you need me, not on a licence.

Result

Running code in your environment, the evals for it, the runbook and a handover, all in your repository. Which rights belong to whom is settled in the contract.

Nothing, if you want nothing. What is built runs at your place and belongs to you.

If you want somebody reachable when it jams, there is the operations engagement: on call for what I wrote, plus a capped allowance of days for continued work. No retainer for availability alone, and cancellable without anything being switched off.

Where this work differs from consulting is at the end. A hand-over here is a roadmap with a date on it, not a closing meeting with documentation attached.

In the largest engagement I am currently in, that date is fixed: at the end of 2026 the whole system — device software, cloud, the test bench on the factory floor, the AI agents inside it — goes over to the client’s in-house team. The roadmap there included recruiting and onboarding an internal engineering lead.

The same pattern sits in older cases. A conformity-assessment database for regulated measuring instruments, carried on by the client’s own developers after sign-off. A catalogue system with which a manufacturer maintains its own decors and therefore calls me less often. Both cost me follow-up work.

I can still stay if you want me to, rather than because you depend on it.

A process that genuinely hurts

Not the one that is easiest to describe, but the one where the most time is lost.

Access to the people who run it

For the whole day, and not only to management. How a process really runs is known by the person who runs it every day, including everything the written procedure does not mention.

Somebody who can grant access

One person with authority, not three weeks of lead time for an account. Missing access is in practice the most common reason a week is lost.

A willingness to let me into production

With an audit trail and with review, and every change has to be revertible. Without that access what is left at the end is a slide deck.