How an AI programme ends up with no owner
It rarely starts as a programme. Marketing trials a content tool. Finance builds a spreadsheet assistant. Operations runs a proof of concept with a vendor. IT licenses a chat platform. A data team prototypes a document classifier. Each was a reasonable experiment, sponsored by someone enthusiastic, and each reported promising early results.
Eighteen months later there are a dozen initiatives, overlapping tools, several vendor contracts, and no single person who can list them all. None has moved into a team’s normal way of working, because each pilot needed the same things nobody owned: data access, a security review, integration capacity, and a manager willing to change a routine. The executive sponsor is asked what the AI budget achieved and cannot answer with confidence.
This is a portfolio problem, not an adoption problem. Coaching staff on any single tool will not fix it.
What the work involves
A complete inventory. I talk to whoever started each initiative and collect the same facts for every one: purpose, user team, tool or vendor, cost to date and committed, data it touches, evidence of results, and current status. Shadow pilots that never went through approval are included, without blame; you cannot govern what you cannot see.
One test for every initiative. Each is reviewed against the same questions. Is there a baseline, so improvement can be shown? Is there a user team that wants it? Is there an approved route to the data it needs? Is there a person who will own it in daily work? What does it depend on that it does not control? Applying one test makes initiatives comparable for the first time.
Shared blockers. Typically three or four dependencies, such as a data-access approval process or a security review queue, block most pilots at once. Mapping them shows where one decision unblocks several initiatives.
Triage with the sponsor. A working session where the sponsor, with the evidence in front of them, decides to continue, merge or stop each initiative and confirms an owner for each survivor. I prepare a recommendation for every item, but the decisions are yours.
The signature deliverable, illustrated
You receive a portfolio triage, named workstream owners, stop decisions and a sequenced execution backlog. Illustrative extract from a triage table:
| Initiative | Evidence | Owner | Dependency | Decision |
|---|---|---|---|---|
| Contract clause summariser (legal ops) | Baseline and reviewer scores on 40 contracts | Legal operations manager | Document store access approval | Continue; first in sequence |
| Two separate meeting-notes tools (sales, HR) | Usage only, no baseline | None found | Overlaps the licensed chat platform | Merge into platform; stop both contracts at renewal |
| Demand-forecast prototype (operations) | Demo on sample data | Former sponsor has left | Data warehouse integration | Stop; revisit with a new owner and data route |
Illustrative example. It shows the format, not a client’s portfolio.
The execution backlog sequences surviving work by dependency and readiness, and the review format gives the sponsor a one-page monthly view: owner, next milestone, blocker, and whether evidence is improving.
How acceptance is judged
The sponsor accepts the diagnostic when every known initiative has a recorded decision and reason, every continuing initiative has an owner who has accepted the role, shared blockers have named owners, and the first monthly portfolio review has been held using the new format. A shorter list is a normal and healthy result.
Ownership and handover
The sponsor owns the portfolio and chairs the review. Each workstream owner owns their initiative’s backlog and evidence. I hand over the inventory, the triage record, the backlog and the review template. If the programme needs someone to run it part-time while internal owners grow into the work, a fractional AI enablement lead can be scoped separately.
When this is the wrong page
If the issue is a single licensed tool that staff are not using, adoption rescue is narrower, quicker and aimed at a different sponsor. If one specific AI application or codebase cannot ship and you need to decide whether to salvage it, use AI application rescue. For a single pilot’s go, revise or stop decision, the free pilot scorecard may be enough.