The approach
AI-led. Delivery-proven. In that order.
AI programmes rarely fail in the model; they fail in delivery. This is a four-stage approach — Assess, Prove, Embed, Sustain — that treats every AI problem as the operations problem it usually is.
Why AI programmes stall
Four failure modes show up again and again, and none of them are technical:
- Ownership — a pilot has a sponsor; production needs a named owner, a service level and a runbook. Without those three, the model works and the workflow never changes.
- Process — AI applied to an unstable process amplifies the instability. Straighten the sequence and the handoffs first, and the same automation compounds.
- Governance — data boundaries, retention, audit and vendor review decided in week one cost days; discovered in month six, they cost the programme.
- Adoption — the tool ships and the team keeps working the old way. Behaviour change is a workstream with a name on it, not an announcement.
Four stages, four gates
Each stage ends in a decision you control.
You can stop after any one of them and still be further ahead than when you started.
Assess
An AI readiness assessment across six dimensions — data, process, governance, skills, tooling and economics. Output is a scored baseline and an opportunity matrix, ranked by return and effort.
Prove
One workflow. One measurable number. Six to eight weeks. Security and compliance gating happens here, at the front, not after the build.
Embed
Production build with an owner, a service level, monitoring and a runbook. This is where the delivery network comes in — and where most programmes quietly stop.
Sustain
Role-based enablement so the team actually uses it, plus a fractional AI operations lead until the capability is in-house.
Questions
Common questions
Why do most AI pilots fail to reach production?
They fail on operations, not on the model. The four recurring causes are no named owner after the demo, an unstable process underneath, security and compliance arriving last, and adoption being assumed rather than managed. Each shows up as a model problem and gets solved as an operations problem.
What are the four stages of the approach?
Assess (a six-dimension AI readiness assessment), Prove (one workflow, one measurable number, in six to eight weeks with compliance gating up front), Embed (a production build with an owner, service level, monitoring and a runbook), and Sustain (role-based enablement plus fractional operations leadership).
Can I stop after one stage?
Yes. Each stage has a decision gate and stands on its own. You can stop after any one of them and still be further ahead than when you started — there's no obligation to continue to the next.
Get started
Tell me what you're trying to get AI to do.
One conversation is usually enough to tell us both whether there's a real return in it. No pitch, no obligation.
sandeep@sandeeprajhans.com