What should a Chief AI Officer do in the first 90 days?

The goal of the first 90 days

One owner. One policy. One measured win in daily use. Everything else waits.

That sounds small. It is supposed to. The most common way a new AI officer fails is by starting ten things and finishing none. One finished, measured win earns the authority to do the next two.

Days 1 to 30: map

The first month is about getting the real picture and retiring the fastest risk. Do these in roughly this order:

• Agree with leadership on who owns AI (you) and the one or two numbers that matter most this year.
• Interview the people who do the repeated work. Build a repeated-work inventory together: task, who does it, times per month, hours per month, whether it is written down, and the risk if a draft is wrong.
• Find out which AI tools staff already use, officially or not. No blame. You need the real picture.
• List which vendors already use AI on your organization's data.
• Draft the one-page AI use policy and get it adopted.
• Pick one deployment: high-volume, low-risk, easy to measure. Record its “before” number.

Days 31 to 60: ship one win

The second month is about getting one workflow into daily use the right way.

• Write the step-by-step rule for how the chosen task should run with AI in it.
• Set it up on approved tools only, with the policy's data rules built in.
• Run it with the people who do the work, not for them. Adjust the rule every time they get stuck.
• Decide exactly when you will measure the “after” number, and put it on the calendar.

Days 61 to 90: prove it and hand it off

The third month turns one win into something the organization owns.

• Measure the “after” number on the date you set. Report it straight, even if it is smaller than you hoped.
• Save the working version in a shared playbook: steps, prompts, and owner.
• Train the next group of staff on it hands-on.
• Present a one-page readout to leadership.
• Pick the next two deployments from your inventory.

How to pick your first AI deployment

Look for work that happens often, follows a pattern, and would not hurt anyone if a draft were wrong, because a person still reviews it.

Leave anything high-stakes (legal, medical, or financial decisions, or anything sent without review) until you have a proven win and a working policy.

Good first candidates:

• Drafting routine emails and newsletters.
• Summarizing meetings into action items.
• Turning long documents into short briefs.
• First drafts of proposals or grant sections.
• Sorting or tagging incoming requests.

What goes in the 90-day readout

One page for leadership. No jargon, no screenshots of chat windows. Cover these seven things:

• What we set out to do: the owner, the number, and the deployment we chose.
• What we did: policy adopted (date), deployment in daily use (since when), staff trained (how many).
• What changed: before number, after number, cost, annualized result.
• What didn't work, and what we learned.
• Risk: what the policy now covers, and anything still open.
• Next 90 days: the next two deployments and the numbers they'll move.
• The ask: time, budget, or decisions needed from leadership.

Mistakes that sink the first 90 days

I see these over and over, and I have made a few of them myself.

• Buying tools before a policy. Tools are easy to buy and hard to unwind. Policy first, then tools that fit it.
• Pilots with no before number. If you didn't measure first, you can't prove anything later. Measure on day one.
• The hero problem. If the system only works because you are there, you built a job for yourself, not a capability.
• Chasing the demo. Boring, repeated work pays first.
• Starting over for every new tool. Write rules that survive tool changes.

What this looks like in practice

A few systems I run myself: an automated news pipeline that lets one person run a daily sports media brand, an RFP discovery system with a source list, a fit filter, a proposal kit, and scheduled AI runs, and a daily command brief that pulls from Gmail, Calendar, and Drive with no manual entry.

None of them started as “AI projects.” Each started as a repeated task I wrote down first. That is the pattern the first 90 days are designed to teach your organization.