How does an AI pilot project work?

01Week 1: Scope the use case, clarify the data, define success criteria

The pilot starts not with code but with a sharp cut. In the first week we decide together what the solution has to do in the end — and what is deliberately left out.

• Scope the use case: a single, clearly described application — for example “read incoming invoices and hand them to the ERP in structured form”. Everything else is noted down but not built.
• Define the data sources: which systems, documents and interfaces the solution needs, who provides access, and which sample data serves development and testing.
• Agree on success criteria: how acceptance will be measured in the end is put in writing before the first line of code is written. That protects both sides from arguments in week 6.

02Weeks 2 to 6: Build, integrate into your systems, set up EU hosting

The longest phase is development. It doesn’t happen behind closed doors: you see interim results on your own data and can course-correct before anything heads in the wrong direction.

• Development: the solution is built along the criteria agreed in week 1 — not as a prototype, but as software intended for production use.
• Integration: connecting to your systems (ERP, DMS, CRM or line-of-business application) is part of the pilot. An AI solution with no connection to the real workflow stays a toy.
• EU hosting: in parallel, hosting is set up in the EU, GDPR-compliant with data minimisation and access control — from the start, not bolted on afterwards.

03Weeks 6 to 8: Testing against real data, acceptance, code handover

To finish, the solution is checked against the criteria from week 1 — on real data from your day-to-day work, not on hand-picked examples.

• Acceptance on real data: your staff test the solution on real cases. It is measured against the success criteria agreed in advance.
• Code handover: your team receives the complete source code — that’s part of the fixed price. You stay independent and free to switch at any time.
• AI Act check: for the use case delivered, we include a conformity check. For typical mid-market applications the effort involved is manageable.
• Six or eight weeks? If data and interfaces are ready from week 1, going live around week 6 is realistic. Deeper integration or scattered data sources push it towards week 8 — and that shows up during scoping, not only at the end.

04What you as the customer need to contribute

A pilot in six weeks only works if both sides deliver. Your contribution is concentrated at the beginning and the end:

• One fixed point of contact: a person who bundles questions and can make decisions about the use case. That keeps coordination short.
• Data access in week 1: sample documents, test data, interface credentials. Delayed access is the typical reason a pilot needs eight weeks instead of six.
• Test users for acceptance: staff from the department who check the solution on real cases in weeks 6 to 8 — they are the fastest to spot whether the result holds up in everyday work.
• IT for the handover: someone who takes delivery of the source code and hosting setup, so the knowledge lands in-house and doesn’t stay with the provider.

05How to recognise a good pilot project

Not every “pilot” deserves the name. Four traits separate a solid project from an extended demo:

• Acceptance criteria before you start: written, measurable, checkable against real data. A pilot without criteria agreed up front is a demo with an open ending.
• Production use instead of prototype: the result runs in your systems in the end and is used by your team — not on the provider’s laptop.
• A fixed price for a fixed scope: at appDev from 39,000 €, depending on the state of your data and the depth of integration. Changes to the scope are recorded as a separate step; the agreed fixed price stays stable.
• A clear next step: what comes after the pilot is named — at appDev an optional operations plan from 1,900 €/month with ongoing development and KPI tracking, as a separate module. You decide after the pilot, not before.