From pilot to scale — in three phases

An AI rollout is not a tool purchase but a guided process with phases and measurable transitions. Anyone who wants everything at once ends up in a big bang and fails on the operating structure. We developed a three-phase model from our AI adoption work. Every phase has a clear goal and a gate that releases the next step.

The three phases and their gates

Phases without transition criteria lose their way. That is why every phase has a gate — a verifiable state that releases entry into the next phase.

In phase 1 you build a core team and implement the first productive use cases. In phase 2 you onboard further areas and deepen integrations in parallel. In phase 3 you connect complex systems — ERP, data pipelines, process agents. This is more reliable than the big bang because every phase builds on proven patterns rather than on hope.

Go broad and go deep run in parallel

The most common thinking error: first reach every employee, then integrate. Wrong. From phase 2 onwards, breadth of adoption and depth run at the same time — not one after the other.

• Breadth of adoption (track A): onboard further teams and sites in a structured way, establish training and office-hour formats, roll out standard use cases.
• Champions & depth (track B): champions build internal assistants, deepen integrations and carry AI knowledge into the company.
• Steering & KPIs (track C): monthly prioritisation, two-week delivery cycles, a quarterly review with an adoption and KPI view.

Go only broad and you get flat adoption without hard ROI. Go only deep and you build lighthouses nobody uses. Only both movements together deliver the full value — the measuring impact study shows it hour by hour.

The 90-day pilot recipe

The pilot is designed for speed and learning, not for completeness.

Bring IT, product owners and data protection on board early — not shortly before go-live. The sequence: a kickoff with roles and priorities, two core workshops, the first assistants and workflows, then follow-up sessions on quality and adoption. Give your champions protected time in the calendar. Without it, adoption peters out — that is the one pattern missing in every failure.

Which unit first — the hybrid cut

The most common question is not "which model" but "in which structure do we roll out?". Three options, one recommended default.

• Site-based — when plants and countries work very differently.
• Department-based — when sales, service and procurement have clearly separated processes.
• Process-based — when end-to-end flows run across several areas.

In practice the hybrid is the most stable: start with one or two departments, then roll out across sites with proven patterns, and deepen one or two end-to-end processes in parallel. The decision criterion is simple: pick the unit that delivers visible results in six to eight weeks.

Estimate the effort honestly

Not every integration is equally expensive. A small S/M/L matrix keeps expectations realistic.

Never connect SAP fully at the start — with very many tables a full connect at the beginning is almost never sensible. Choose tables based on the use case, start with a small set and extend only where the benefit is clear. Read-only first, write actions later.

Where to read on

The three phases are the scaffolding — people carry it. How to win champions and dissolve resistance is in change management. The security frame for track B comes from governance & security, and the deep use cases are built with automation. And what the rollout is worth in the end is calculated by measuring impact.