Dossier

From AI pilot to an operating capability

Ownership, economics and the decisions that turn an experiment into a supported service.

The strategic problem

An AI pilot can demonstrate a compelling output without demonstrating a viable service. The gap usually involves workflow ownership, integration, evaluation and the cost of handling mistakes. Begin with the operating decision: which task changes, who uses the output, and who is accountable when it is wrong?

Three decisions to make early

Choose the boundary of automation, the acceptance test and the owner of the continuing service. A team that can build a prototype is not automatically the team that can support it. Assign data ownership, release approval and incident response before the first production commitment.

Build an economic baseline

Record volume, handling time, quality and exception rates in the current process. Include review, integration, training and model-use costs in the proposed alternative. A reduction in task time is not automatically a cash saving: explain whether it creates capacity, improves service or changes staffing demand.

The production gate

Ask users to complete real tasks under representative conditions. Test failure paths and rollback, not just the happy path. Require a named person to accept the remaining error profile. If the benefit depends on staff silently correcting outputs, include that work in the result.

What the board should receive

Use a compact service record: problem, owner, affected users, baseline, measured result, unresolved risks, total cost, suppliers and the next decision date. Compare initiatives on the quality of their evidence rather than the sophistication of their demos.

Discussion for EventChanger

Which use cases have survived six months of ordinary operation, and what changed between the pilot and the supported service? Invite the business owner and frontline users alongside the technology team.