Pilots and Experiments

Why This Page Exists

A pilot is a deliberate, time-bound trial with someone accountable for reporting what happened — including when it does not work. This page explains how to run one well, so that what gets learned in one department does not have to be relearned in six others.

The most expensive pattern in institutional technology adoption is not failure. It is undocumented failure — the same approach tried independently in four departments, abandoned four times for the same reason, with nobody having written it down.

A pilot that ends in "this did not work and here is why" is a genuine contribution. It is frequently more valuable than a success, because it saves other people the attempt.

Pilots currently underway across the district are being compiled and will be listed here. If you are running one, please tell the ARC — details at the bottom of this page.

Running a Pilot Well

"We are going to try AI in our department" is not a pilot; it is a period of time during which some people use some tools. When it ends, nobody can say whether it worked.

Write down the actual question first, and make it answerable: Does this reduce the time spent on intake by a measurable amount? Do students who use this score differently on the same assessment? Does this cut the email backlog?

Then decide, in advance, what result would make you stop. Pilots that cannot fail tend not to end.

Before a pilot involves real people's data or real institutional processes, confirm what applies. Depending on what you are doing that may include data classification, contract and procurement review, security review for a new vendor, accessibility requirements, and records retention.

These apply whether or not the tool costs money. A free tool processing student information is not exempt because it was free.

Choosing the Right Tool covers what to check and who to ask. Doing this at the start typically costs a week; discovering it at the end can cost the whole pilot.

If students are affected — by a tool in their course, a change in how their work is handled, or a service they interact with — they should know, in plain language: what is being tried, what happens to anything they enter, and what their alternative is if they would rather not participate.

An opt-out matters most when participation is not really voluntary, which in a required course it is not.

A short writeup is enough: what you tried, what you expected, what happened, what you would tell someone considering the same thing.

Resist the pull toward a success narrative. A pilot reported as "promising, needs further study" when it actually did not work is worse than no report, because the next department reads it as encouragement.

Tell the ARC About Yours

If you are running or planning an AI pilot anywhere at Maricopa, the ARC would like to know — at the start, not only at the end. Early visibility means we can connect you with someone already doing something similar, flag a requirement before it becomes a problem, and make sure the result reaches people who need it.

Email ai-resource-center@domail.maricopa.edu with what you are trying, who is involved, and what question you are trying to answer. It does not need to be formal.

When a pilot concludes, a few paragraphs on what happened is enough to make it useful to everyone else.