[blog]
The scoping conversation that decides, before any work starts, whether an initiative can succeed.
Not "what can AI do here?" but "which number, reported by whom, do you want to move?" It sounds obvious. In practice most AI initiatives can't answer it, and that's why they end as pilots.
The number must already exist in a report someone reads. Trading volume in a segment. Prepaid recharge rate. Cost per case. Days to decision. If the metric has to be invented for the project, the project has no owner.
The owner is the person whose budget or bonus improves when the number moves. Not the CDO, not the innovation lead, not the vendor. If that person isn't in the scoping conversation, we ask for them to be. If they can't be found, we say so and stop.
Uplift without a baseline is a story. We design the measurement before the intervention: a held-out segment, a region, a set of agents, or where randomisation is impossible, a matched pre-period adjusted for seasonality. The design is written down and agreed.
A result at which the team stops. Agreeing it in advance removes the sunk-cost argument later and makes bolder ideas safe, because a cheap, fast no is a legitimate outcome. A fair share of Prove sprints end this way, and they're the cheapest lessons a client ever buys.
A one-page scope: metric, owner, control design, kill criterion, and cost to prove. Everything else — models, platforms, architecture — is chosen to serve that page.
[related]
[share]
Copy the address bar link, or send this article to a colleague who owns the metric.
Tell us about the decision or workflow you want to change. We'll come back with an honest view on whether it's worth proving, and what it would take.
Start a conversation