Procurement at an independent school
Five stakeholder groups, three systems that didn't talk to each other, and one administrator absorbing all the friction. The answer was not a piece of AI software.
The situation
A school brought me in to look at how classroom supplies got ordered. The presenting complaint was familiar: teachers waited too long for materials, and nobody could see how much budget was left until it was already spent.
Underneath that, one person — the purchasing administrator — was manually processing over 200 transactions a month. For each one she searched a vendor catalog, checked a budget she had to reconstruct by hand, placed the order, then re-entered the same transaction into a separate payment platform and again into the school's financial system. Three systems, none of them connected, all requiring the same data.
She was not the problem. She was the compensating control for a process that had never been designed — it had accumulated.
What I actually did
- Interviewed every group, not just the sponsor. Teachers, the purchasing administrator, the accounting director, and school leadership. Their priorities genuinely conflicted, and any recommendation that ignored that would have died on contact.
- Mapped the real workflow. The documented process and the actual process were different. An earlier process diagram showed teachers ordering directly — they didn't, and administration had no appetite for them to.
- Tested the vendor claims. I checked what the platforms could actually do against what the marketing said they could do, before anyone signed anything.
- Built a phased roadmap with a deliberately low-risk first phase, plus two divergent options for later expansion depending on how phase one landed.
What I found
The most useful finding was a negative one. The obvious move — push ordering out to every teacher — was the wrong move. Leadership said plainly that they wouldn't ask staff to take on budget responsibility, and teachers didn't want it. A technically elegant solution that nobody would adopt is not a solution. So the first phase changed nothing about the teacher experience except that it got faster.
What I recommended
Phase one kept the familiar intake process entirely intact and concentrated on removing the manual re-keying and giving the administrator real-time budget validation at the moment of ordering. It was designed to be reversible, to require no new logins for teaching staff, and to be worth doing even if no later phase ever happened.
Phases two and three were sketched as options, not commitments — to be revisited only after phase one had proven itself in daily use.
Where AI came into it
It didn't. There was no model, no chatbot, no "AI-powered" anything in the recommendation. The problem was a process and integration problem, and it deserved a process and integration answer. Adding AI would have added cost, fragility, and another system for one person to maintain.
That is the honest shape of a lot of this work.
A note on what's claimed here
This engagement covered discovery, analysis and recommendation. The efficiency figures modeled during the project — a substantial reduction in per-transaction handling time and in teacher wait time — are projections from the recommended design, not audited post-implementation measurements. I'd rather describe the work accurately than quote a number I can't stand behind. Client name withheld pending their permission.
Recognize any of that?
If one person in your organization is quietly holding a process together, that's usually worth half an hour of conversation.