I get called in when a small company is about to spend money on AI, and I try to find out whether that money would fix anything.
Most of the time it would not, at least not yet. The request arrives as “we need an agent,” and underneath it is a routing problem, an unwritten policy, or two people who each believe the other one owns a decision. Software makes all three faster, not better.
So the work starts with a diagnostic instead of a build: a few weeks of reading how the operation actually runs, then a written recommendation that sometimes says “automate this,” and often says “write this down first, then automate the third step only.” When automation is the answer, I build it or hand a spec to whoever should. When it is not, I say so in writing and the engagement ends smaller than it was scoped.
Everything I learn doing this gets written up here: long-form after a project closes, short notes the week they happen. That is the whole marketing strategy.
Three phases, in that order, with a written deliverable at the end of each. You can stop after any of them; most of the value is in the first.
Two to three weeks. I sit with the people doing the work, map the decisions before the workflows, and find where the hours actually disappear, usually in handoffs rather than tasks.
We fix the process on paper first: owners, exception paths, sign-off rules. Then, and only then, we choose tooling, against your seat count and your on-call reality rather than a feature grid.
Smallest useful build, in production, with a named owner and a handover page written in week one. Six-month follow-up included, because that is when the real result shows up.
These exist so we can both tell early whether this is a fit.
You know the number before we start and it does not move because the work turned out to be interesting. A change of direction means a new, separately priced piece.
I take no referral fees, resale margin, or partner status from any tool I recommend. It is the only way the recommendation is worth reading.
Deliverables are documents you can forward and argue with, not slides that need me in the room to make sense.
If a checklist and one fewer approval layer solves it, that is the recommendation, even when it makes the invoice smaller.
Everything is built in your accounts, under your billing, documented for your people. No dependency on me after handover.
Three shapes. Anything else is one of these with a different cover page.
| Engagement | What happens | Length |
|---|---|---|
| Diagnostic | Interviews, decision map, one-page findings with a ranked shortlist and a “don’t do this” section. Ends in a document, not a proposal. | 2 to 3 weeks |
| Design & build | Process redesign, SOPs, tooling decision, then the smallest useful automation in production with a run book and a named owner. | 4 to 8 weeks |
| Second opinion | A half-day read of a proposal, quote, or half-built stack you have already got, with a written yes / no / not-yet and the reasoning. | One session |
The service page is not up yet. Until it is, the writing here is the honest way to judge whether this method is a fit.