Approach / The delivery brief

Five things to agree on before a pilot starts.

Before the first sprint, the client and delivery team need a shared brief. Explore each field: the question, the risk, and an example from my work.

01 / Delivery brief

The business problem

The question: What is costly, slow, risky, or otherwise worth changing?
When it is missing: The pilot optimises for what the demo does well, and afterwards nobody can say what it fixed.
From the work: On the AI account, recurring reporting was the cost that could be named and measured. That is where the automation started, not with the most impressive model.

02 / Delivery brief

The users

The question: Whose workflow needs to change, and what support will they need?
When it is missing: Adoption gets measured after the fact, once the people who were supposed to change have already found a workaround.
From the work: Three enterprise AI implementations went live on time and were adopted because the recruiters using the system were part of the delivery plan, not its audience.

03 / Delivery brief

The constraints

The question: What do data access, security, integrations, time, and competing tools allow?
When it is missing: The first integration meeting becomes the moment the timeline is rewritten.
From the work: The Digital Twin work depended on partner data and a fixed price. Both were written down before the first sprint, and the first POC landed in about five weeks.

04 / Delivery brief

The success signal

The question: What evidence would be credible enough to act on?
When it is missing: Every stakeholder brings a different chart to the review, and the pilot ends in a second pilot.
From the work: Margin and delivery metrics were the shared numbers on the AI account. Tradeoffs were argued from them while the team still had room to act.

05 / Delivery brief

The decision

The question: What will the evidence allow the customer to stop, fix, expand, or buy?
When it is missing: A pilot that cannot be stopped is not a pilot. It is an unbudgeted rollout.
From the work: Scope Sentinel is built the same way: when the evidence is missing it says so, instead of drafting a plausible answer for someone to approve.