A costed AI plan, including what not to build
A senior engineer spends two to four weeks inside your problem and looks at your real data. You get a sequenced, costed roadmap — and an honest list of the ideas that aren't worth it yet.
A plan you can defend.And a build spec.
No frameworks from an analyst. No deck you can't act on. A costed roadmap grounded in your real data, detailed enough to hand straight to a build team.
A scored opportunity map
Every candidate use case on one impact-versus-effort grid, each tied to a real process, an owner, and a measurable metric — not a brainstorm list.
A costed ROI model you can edit
Build cost in engineering weeks, recurring inference and infra cost at real volume, and the value side with every assumption stated and a break-even line. A spreadsheet, not a slogan.
A build-vs-buy call per use case
Build custom, wrap an existing model or API, buy off-the-shelf, or do nothing — with the specific tools and vendors we'd actually choose, and why.
A data-readiness report
What data exists, where it lives, and the quality, coverage, and labelling gaps to fix before any model earns its keep. Usually the real bottleneck, surfaced early.
An EU AI Act and GDPR risk note
Each use case classified by risk, with lawful basis, data residency, and DPIA triggers flagged — and a clear note on what needs your lawyer. Practical, not a legal opinion.
A sequenced 6–12 month roadmap
Phased, costed, dependency-ordered, and detailed enough to be the build spec — designed so the first shippable result lands fast. Plus an explicit do-not-build list.
Five steps,first win lands fast.
From scoping call to live readout — most candidate ideas get cut along the way, and that's the point.
Thirty minutes, an engineer on our side. We cover the business pressure, the data situation, and the budget. Sometimes you already know what to build and should skip straight to development — we'll say so.
Week one to two: we interview the people who own the processes and the people who own the data, get read access to the relevant systems, and look at the actual schemas, volumes, and quality — not a description of them.
Week two to three: each candidate gets a feasibility verdict, a cost model, a compliance classification, and a build-buy-or-do-nothing call. This is where most candidates get cut.
Week three to four: we sequence the survivors into phases with dependencies and a fast first win, attach a metric and a cost ceiling to each, and write the do-not-build list.
A working session walking you through the plan, the numbers, and the trade-offs — argued live, not just emailed. You own every artifact. If a build follows, the roadmap is the spec.
We'll tell younot to build it.
If any of these sound familiar, we should talk.
Founders with a board mandate
On the hook to "have an AI strategy"
- A deadline to have a plan and no shortlist
- Pressure to defend it to investors
- Real risk of spending six figures on the wrong idea
Outcome: A plan you can put in front of the board, with the bad bets already cut.
CTOs with more ideas than budget
Whose team is pitching agents, copilots and RAG
- More proposals than budget to fund them
- No outside read on what's feasible against your data
- Compliance landmines hiding in the backlog
Outcome: A sequenced shortlist you trust, with the landmines flagged.
Leaders whose pilot stalled
A POC that demoed well and died in production
- Data that wasn't ready for production
- Costs that blew up at real volume
- No one ever defined what success meant
Outcome: A grounded second attempt that survives real users this time.
Senior engineers. No handovers. No fluff.
Start your deployment.
Talk directly to a principal engineer.
No sales team.
No discovery workshops.
No procurement circus.
We scope, build and ship.
- Reply within 24h
- Engineer-led assessment
- Written proposal
- Portugal / EU timezone
No commitment. Just an engineer.