Eleven projects, no map
Four departments, eleven separate AI initiatives, three overlapping vendor conversations, and two teams that had independently built the same document classifier without knowing about each other.
This is the standard result of an organisation reacting to pressure from the top without a mechanism for coordination.
The audit
Five weeks across process, data, systems, capability and risk.
The data findings were the ones that reset expectations. Two of the highest profile initiatives depended on production data that was recorded inconsistently across sites. Fixing it was a longer job than the project itself.
The systems findings were the ones that saved money. Three of the eleven initiatives were asking for capability that already existed in a platform they were licensed for and not using.
Sequencing
We scored the portfolio and deliberately put the boring items first.
The two that got funded were a quality inspection workflow with a clear measurement baseline and a maintenance scheduling assistant. The second had a narrow scope and a low cost of error. Neither was the most exciting item on the list. Both were things the organisation was able to succeed at while it was still learning how to run this kind of work.
Nine were stopped. Each got a written reason, kept in the roadmap document, along with the condition that justifies revisiting it.
Governance
Two pages. What data can leave the network, who approves a workflow that touches a customer or a payment, what gets logged, who reviews monthly.
We have written forty page governance frameworks before and they do not get read. The two page version is on the wall in the operations office.
Where it went
Both funded programs reached production within the twelve month window. The stopped list held, which is the outcome the COO cared about most and the one that is hardest to attribute to the work.