The AI pilot blueprint

The Handover Method is a scoped, senior-led AI advisory engagement that ends with the client team in possession of the runbook, the decision log and the production AI workflow. Scoped pricing, written scope, built-in progress reviews, handover before close.

It is not a perpetual retainer, not a vendor pass-through and not a discovery deck. The shape exists so the buyer can sign one document, see one outcome and walk away with a team that can keep going without us.

A bounded pilot is not a fashion. It is the shortest interval that survives leadership's review cycle and the longest interval that survives your own attention span. A bounded scope lets a CFO book the cost inside a planning window, a COO sequence team capacity around it and the advisor deliver something the business can run without us.

Too short and the baseline has no time to stabilise — the result becomes noise dressed as signal. Too long and the cost of the pilot starts to compete with the cost of just fixing the process in-house. The engagement shape must match the cadence of the business that is paying for it.

The most underrated phase of the engagement. What we write down at the start is what we get measured against at the close. Weakness here is the most common reason pilots ship a result that does not survive a board conversation.

A real baseline has three parts: a KPI read from your operating system (not a vendor dashboard), a measurement protocol agreed with your finance leadership and a written agreement on what 'improved' means. We close baseline with those three artefacts and a signed amendment to the SOW. If any of them is missing, the next phase does not start.

Discovery is where most consulting firms invoice without producing anything the client can own. Our version is short, ends with a written operating-model diff and is paid for out of the pilot fee — not billed separately. If discovery is worth invoicing on its own, the engagement has already failed.

The output is not a slide deck. It is an annotated workflow diagram, a list of the three or four interventions we would actually make and a go/no-go recommendation. If discovery shifted the picture, the progress review is where the SOW is re-aligned in writing — and where leadership sees the same story we do.

Every SOW we write includes a progress review where baseline, scope and target are re-confirmed against what discovery actually showed. Direction is adjusted in writing if discovery shifted the picture, and the decision log captures the call. This is not a refund clause and not a renegotiation. It is a scope-reality checkpoint that keeps the SOW honest about the work as it actually runs.

Procurement teams always want this written in, because it is the piece that converts a multi-phase commitment into a sequence of explicit alignment decisions. The cost of writing it is small. The cost of not writing it is the cost of an engagement that drifts away from what leadership signed.

This is the part that looks like delivery. What it actually is: an operating-model change running inside the business, read against the baseline on a regular cadence.

Frequent measurement is non-negotiable. The AI layer is the noisy part of the system — it gets better or worse depending on data drift, operator behavior and seasonality. If we read the KPI once at the end, we have no way to separate model drift from process drift, and no way to explain the curve to the CFO. Multiple cadence snapshots beat one big reveal every time.

The handover clause in every SOW states: by the close of the engagement, your team owns the runbook, the decision log and the operator playbook. We are not the embedded advisor six months later. We are a phone call when the operating model changes.

Handover is the acceptance criterion we are most protective of. It is the reason clients do not end up in a managed-service tail — and the reason we do not end up in one. Both outcomes are worse for long-term trust than charging a fair pilot fee up front.

In consulting, 'done' usually means 'the invoice is paid'. Ours has three acceptance criteria, all written into the SOW — the target KPI was read from the client's operating system by the client's finance team against the agreed baseline; the client's team runs the workflow without the advisor on the call; and the decision log is complete enough that a new operator could read it cold and understand why the workflow is shaped the way it is.

If any criterion fails, the engagement is not done. We extend at our cost until they pass — and the decision log captures why each criterion took the path it did.

Three pilot failure modes we see across the industry. Baseline theatre — the client team writes a baseline that is easy to beat; the measurement protocol is owned by someone too junior to defend it in a CFO conversation. Prevention: baseline artefacts have to be agreed with finance leadership before the next phase can start.

Demo-to-nothing — the pilot ships an impressive demo but never gets into the operating system. Prevention: the progress review forces a written commitment to the production path before the back half of the engagement begins.

Handover-by-handshake — the engagement closes on a phone call and the client 'owns' a Slack channel of context instead of a runbook. Prevention: handover is a SOW acceptance criterion, not a gesture at the end of the last meeting.

The scoped pilot is the shape of a commitment the business can actually make. Every slot above is engineered to close on a handover — not to keep the advisor in the room. That is The Handover Method.

If you want to see how this maps to your own operating problem, the 30-minute call is the right next step. Bring the workflow, the cost pressure or the stalled initiative. We will map it against the blueprint on the call.

These are the engagement shapes a CFO can pick from when an AI initiative needs to be commercially safe and operationally real. The Handover Method (a scoped pilot) is one shape — not the only shape — and it fits a specific window: a workflow with meaningful operating exposure, an outcome the business can verify in a defined window and a team that can absorb a runbook before we leave.

When the answer is none of those, an RFP route or an in-house build are the right calls and we will say so on the call. The point of naming the shapes is to make the choice readable instead of leaving it to vendor narrative.