Sign in or create a free account to save this Card to your Deck.

Practice Card · Evidence of progress

Define one or two 90-day management AI pilots

TWO GOOD PILOTS BEAT TEN UNFINISHED EXPERIMENTS.

A management AI pilot should be small enough to supervise and consequential enough to teach the company something useful. One or two well-defined pilots create clearer evidence than a portfolio of loosely governed experiments. Each pilot needs a business problem, current baseline, accountable manager, defined users, permitted inputs, intended AI capability, human decision boundaries, risk controls, success measures, and a review date. Ninety days is a planning horizon for structured experimentation, not a promise that every use case should reach full scale in that period.

Expanded teaching

Expanded teaching

Choose a management problem

Start with work a manager already owns: preparing decisions, coordinating work, improving a workflow, supporting a team, monitoring a result, or removing a known source of friction. Avoid pilots selected mainly because a tool is novel.

Keep the portfolio small

One or two pilots allow leaders to pay attention to adoption, exceptions, data quality, workflow design, and business results. A long list of simultaneous experiments creates activity while making it harder to learn which implementation choices matter.

Write the pilot charter

Define the current method and baseline, expected business result, owner, users, scope, inputs, AI capability, human approvals, prohibited actions, risks, training needs, measures, stop conditions, and review dates.

Manage the 90 days as stages

Use early weeks for process and baseline validation, the middle period for controlled use and adjustment, and the later period for comparison and decision. Do not spend most of the quarter configuring technology and then evaluate on a few days of use.

Where AI can help

AI can help structure the charter, generate test cases, organize evidence, summarize user feedback, and compare results with the baseline. Managers own the workflow, staff experience, business decision, and whether the pilot should scale.

Graduate, revise, or stop

A pilot should end with a decision. Scale when benefits are supported and controls are adequate. Revise when the problem or design remains promising but the evidence is incomplete. Stop when the business case, readiness, or risk does not support continuation.

Diagnostic inquiry

Questions to ask, not conclusions

Can each proposed 90-day AI pilot be explained as a bounded management experiment with a business baseline, accountable owner, human decision boundaries, measurable result, and explicit end-of-pilot decision?

Questions that distinguish possible explanations

  • Which pilot can be supervised well enough that failures and exceptions will be visible?
  • What larger dependency could make the result impossible to interpret within 90 days?
  • What condition would cause you to stop rather than continue because the organization has already invested time?

Practical application

Practical application

Select no more than two management AI pilots for the next 90 days. Write a one-page charter for each with the problem, baseline, owner, scope, users, inputs, human approval points, risks, expected result, measurement method, stop conditions, and final decision date.

Ask Atria

Talk through this Card

Talk it through with us

If this Card describes your week, a free conversation is the fastest way to see what it means for your company and what to do next.

Book a free conversation