Nobody ever got fired for running a pilot. People get fired for rollouts that fail loudly.
That’s the framing that’s been missing from the AI-in-education conversation. Districts keep being told to “integrate AI,” and they keep responding the way you’d expect risk-averse institutions to respond: with big, slow, expensive plans that take a year to approve and another year to roll out — and that scare everyone involved, from the superintendent down to the teacher who’ll actually have to use it.
The answer isn’t to move faster. It’s to move smaller. A 90-day pilot, done right, proves AI works before anyone has to promise it works. And it does it without putting anyone’s career on the line.
Why Pilots Work
A pilot is the lowest-risk way to introduce something new, and it works because of a simple asymmetry: the downside of a small failure is contained, but the upside of a small success is enormous. If your pilot fails, you’ve lost a semester and a little credibility — and you learned what doesn’t work, which is itself valuable. If it succeeds, you have evidence, and evidence is the currency that convinces boards and budget committees.
The mistake districts make is treating AI like a district-wide initiative from day one. That’s not a pilot. That’s a gamble.
The 90-Day Path
Here’s the shape of a pilot that doesn’t scare anyone:
Month 1 — Audit and Prepare. You don’t spend money in Month 1. You audit what you already have — the underused servers, the tools already in the building. You pick ONE pilot class, not a rollout. You stand up a single virtual desktop host on the hardware you own. And you set a baseline: how does that class work today, before you change anything? You need that baseline to prove improvement later.
Month 2 — Pilot and Train. Deploy one AI-assisted workflow in that single class, end to end. Train the faculty on using the tool, not on theory. Introduce a simple governance model so students can use AI productively without abandoning academic integrity — and document everything. What broke, what worked, what students said. You’re building your evidence base in real time.
Month 3 — Evaluate and Expand. Measure against the baseline. Did workload drop? Did student outcomes improve? Collect student and teacher voice — the people living it have the best data. Then plan expansion: from one class to a program, from a program to a district — but only with evidence in hand.
Why This Works With Leadership
This approach succeeds with administrators because it never asks them to bet the farm. It asks for permission to try one thing, in one classroom, for one semester, using resources the district already owns. That’s a request almost no superintendent can say no to, because the cost of saying yes is tiny and the cost of saying no is “we’re the district that refused to even try.”
And when you come back in ninety days with baseline-to-results data, you’re not asking for a leap of faith. You’re asking them to fund something that’s already working.
The Point
The AI-in-education conversation has been stuck on “should we?” long enough. The way out isn’t a bigger plan. It’s a smaller one. One class. One workflow. Ninety days. Evidence instead of promises.
That’s a pilot a superintendent can say yes to — and a program you can build from.