AI Consulting: From Strategy to Deployment

What a good engagement produces

The output should be specific enough to act on: a shortlist of candidate processes ranked by value and feasibility, an honest assessment of whether your data and systems can support them, an architecture, and a cost and effort estimate you could take to a board.

It should also say clearly what not to do. An engagement that recommends everything has not done the analytical work.

• A prioritised, costed shortlist rather than a list of possibilities
• A readiness assessment naming specific blockers
• A target architecture and technology recommendation
• An explicit list of what was considered and rejected, with reasons

Discovery has to involve the people doing the work

Strategy built only from conversations with leadership tends to describe the process as management believes it works. The exceptions, workarounds and informal rules that determine feasibility live with the people doing the job.

If a consulting engagement has not spent time with those people, treat its conclusions with caution.

Prototype early, decide honestly

The fastest way to resolve a disagreement about feasibility is to build the narrowest possible version and look at it. A prototype that takes two weeks can settle a question that would otherwise absorb two months of meetings.

That requires a genuine willingness to stop. A prototype that cannot fail is a demonstration, not a test.

Plan the handover from the start

Decide early who will operate and change the system afterwards. If that is your team, the engagement should be structured so knowledge transfers continuously rather than in a document at the end.

The measure of a good engagement is that you are less dependent on the consultant at the end than at the beginning.