Fractional CAIO 30/60/90-day plan

A fractional Chief AI Officer who cannot tell you what happens in the first 90 days is selling a vibe. Here is a plan you can hold them to. Hours stay in the 8 to 20 a month range. If they need a war room for 90 days, you hired a project, not a fractional executive.

This is the cadence we use on Fractional AI Leadership. Steal it even if you hire someone else. Pair it with when to hire a Fractional CAIO so you do not buy the plan before you need the seat.

Days 1 to 30: baseline, access, and a kill list

The first month is not a vision offsite. It is inventory.

Map the AI that is already happening: official pilots, unofficial chat tools, vendor trials, and the one spreadsheet everyone pretends is not the system of record. Write down who owns each, what data it touches, and whether it is allowed.

Score the company across data, automation, adoption, and governance. You want a number you can repeat in a board pack, not a 80-page appendix. If you need a fast first pass, the AI Readiness Assessment is the same four dimensions.

Pick the first workflow candidate. High volume, high frustration, measurable in hours or errors, lives in systems you already own. Write the baseline before anyone builds: cycle time, touches, failure rate.

Publish a kill list. Two vendors or tools you will stop this quarter, and one you will keep on a short leash. If nothing gets killed, the CAIO is collecting logos.

Name the monthly operating rhythm: who attends, what decisions get made, what the board sees. Name the human who will use the first workflow's output. If that person cannot be found, stop. You do not have a first project.

Days 31 to 60: one workflow live, with a human in the loop

Month two is proof, not a second strategy.

Stand up the first workflow under the CAIO's oversight. Success metrics were written in month one. Do not change them because the demo looked nicer than the baseline.

Keep a person in the loop. The point is to watch the model fail on real tickets while the blast radius is small. Log what was retrieved, what was sent, and what the human changed.

Close build-versus-buy on that path. If the honest answer is buy, help negotiate. If the honest answer is build, scope a fixed delivery as a separate decision. Do not smuggle a six-month implementation into the retainer.

Stand up the minimum policy: which tools are approved, which data never goes to a public model, who can promote something out of a pilot. Governance that freezes the company is a fail. Governance that exists only in a slide is also a fail.

By day 60 you should be able to say, in one paragraph, what shipped, what it cost, and what you will not do next.

Days 61 to 90: own the number or end the retainer

Month three is where the engagement earns a renewal.

The first workflow is either live with a before-and-after number or honestly dead. Dead is acceptable. Hidden is not.

The roadmap for the next two quarters has owners and rough cost, not 40 initiatives. The board pack has one AI owner, one spend line, and the kill list with dates.

Decide the shape of month four. Raise hours if a second workflow is real. Convert the next build to a fixed-scope engagement. Or help write the job description for a full-time hire and step back. Or end. Month to month exists so ending is cheap.

If none of that is true at day 90, you hired a guest speaker. Stop paying.

What you owe the CAIO

Access to the people who run the messy process. A sponsor who can say no to a vendor in the room. The list of tools already in use. One hour on the calendar every month that does not move. Without those, a 90-day plan is fiction.

You do not owe a data lake, a new platform, or a reorg. Those show up later if the first number moves.

What to put in the contract

Write the 30/60/90 artifacts into the first invoice cycle: scored baseline, kill list, first workflow baseline, live-or-dead pilot, board pack line. Hours cap. No license kickbacks. Delivery, if it happens, is a separate statement of work.

If a shop will not agree to artifacts, they are selling availability. Availability is not ownership.

Talk to an engineer if you want this plan applied to your stack, or to hear that you only need the first 30 days as a consulting sprint.