How to scope an AI engagement that doesn't disappoint
Most disappointing AI engagements were destined to disappoint at the scoping stage. Here is how to avoid that.
Disappointment is engineered at scoping
The most common pattern in failed AI engagements is not bad delivery. It is a scope that was vague, output-defined rather than outcome-defined, and missing the controls plan, the production target, and the capability-transfer expectation. Once that scope is signed, everyone is just executing against the misalignment.
What to insist on in scope
A named production target, not 'a pilot'. A controls plan delivered in parallel, not after. An evaluation harness as a delivered artefact. A capability-transfer plan with named internal owners. A value-realisation review built into the closing weeks.
What to walk away from
Any scope without a controls plan. Any scope where the consulting team owns the model and the data permanently. Any scope that defines success as 'delivered deck' rather than 'system in production'. Any scope that does not name the second use case the architecture should support.
Phased commitments protect both sides
A discovery sprint followed by an explicit go/no-go for embedded delivery protects the buyer's budget and the consultant's reputation. The teams comfortable with that structure are usually the ones worth working with.
Key takeaways
- Disappointing engagements are usually mis-scoped, not mis-delivered.
- Insist on production targets, controls plans, evaluation harnesses, and capability transfer.
- Walk away from output-defined scopes and permanent-IP-ownership models.
- Phased commitments with explicit go/no-go protect both sides.
Related insights
View allAI consulting trends for the next 24 months
The hype cycle is consolidating. Here is what serious AI consulting will look like through 2027 — and what to watch out for.
Read AIAI adoption frameworks: from pilot to operating model
Pilots are easy. Operating models are hard. The gap between them is where most AI programs quietly die.
Read AIBuilding an AI governance model your CISO will sign off on
Governance done well is enablement. Governance done badly is theatre. Here is how to design the first kind.
ReadWant to talk through this in your context?
We'll bring the relevant playbook from this article into a 30-minute working session — focused on your team and your numbers.