Home · Offering · AI Strategy & Architecture
Service 05 · AI Strategy & Enterprise Architecture
From prototype to platform — without the detour through three dead ends
After the first successful prototype comes the expensive phase: ten departments want the same thing, each builds it differently, and nobody has decided what is platform and what is application. Architecture is the answer to that question, not a diagram.
Where it usually gets stuck
- Too many candidates, no order.Without scoring by benefit, data situation and risk, the loudest use case wins, not the best.
- Every department builds its own platform.Three vector databases, two model providers, no shared operations. The effort is incurred three times.
- Decisions without a rationale.Whoever doesn’t write down why an option was rejected will have the same discussion again in a year.
How I work
- Take stock.What’s running, what’s been promised, which data really exists — kept separate from what’s on the slides.
- Assess the portfolio.Benefit, data situation, risk, effort. The result is a sequence, not a wish list.
- Target picture and split.What is platform, what is application, where the boundaries lie and who is accountable for them.
- Build the first slice.A vertical cut through all layers, close to production. It proves the architecture or refutes it — both are valuable.
How you measure it
Whether, after the first slice, a second team uses the same platform instead of building its own — and whether the decisions of back then can still be justified.
Frequently asked questions
Will we end up with a document or a system?
Both, but in that order: the architecture emerges from the built cross-section, not before it. A target picture that has never met code rarely holds.
How long does something like this take?
For a stocktake with an assessed portfolio I count in weeks, not months. The first cross-section follows immediately.
Do you work with our teams or instead of them?
With them. The goal is that the blueprint is carried on without me afterwards.
Getting started
- One plant, two weeks, one number you can rely on.Connect what’s there. Two weeks of measuring instead of guessing. Afterwards, whether the expansion pays off is stated in euros — not in a presentation.