The First 90 Days of a Chief AI Officer

You have ninety days to prove the seat should exist. Nobody will tell you that, and everybody is counting.

The Chief AI Officer job is new enough that no one can hand you a map, and watched closely enough that everyone already has a private opinion on whether the title was a good idea. Here is the part that surprises new officers: what sinks them is almost never the AI. It is the setup around it. The budget that lives somewhere else. The projects they cannot kill. The number nobody agreed to. You can be the best AI mind in the building and still be gone by month eighteen because you spent the first ninety days on the wrong things.

A Chief AI Officer's first ninety days are not for writing an AI strategy. They are for putting a meter and a gate on AI spend, securing a real mandate with budget and kill authority, finding the one place AI turns into money, and shipping a single measurable win before the goodwill runs out. The strategy is the easy part, and it comes last.

Here is how I learned that, expensively.

What $35,000 taught me about day one

The first time I ran an AI rollout at an AI SaaS company, I did the exciting version. I gave fifteen people open access to the models and told them to go find value. Smart people, real problems, no guardrails. It felt like leadership. It felt like trust.

A month later the bill was thirty-five thousand dollars and the pipeline was zero.

Not a slow leak. Fifteen people running experiments that all felt productive, every one of them reasonable on its own, and not a single dollar came out the other end. Someone was summarizing documents nobody read. Someone was rebuilding a tool we already owned. Someone had a loop calling the most expensive model to check its own work. All of it looked like progress on a Friday, and none of it moved a number.

Here is the part that stung. A meter would have caught it in the first week. A gate would have caught it on day one. I had neither, because I was busy being visionary instead of being accountable.

So now I build backwards. Before anyone on my watch touches a model, two things exist: a meter on the spend, so I can see the money moving in real time, and a gate on the output, so nothing expensive ships without a human deciding it was worth it. The strategy can wait. The bleeding cannot.

That thirty-five-thousand-dollar month is the most useful thing that ever happened to how I run a first quarter. It is also the whole lesson, compressed. The job is not to have a vision. It is to put the governor in before you touch the gas.

The clock is shorter than you think

Why ninety days and not a year? Because the honeymoon is short, and it is not yours to extend.

Look at the job that came right before this one. The Chief Data Officer averaged a tenure near two and a half years, well under a CFO or a CIO, and the goodwill tended to run out around month eighteen. As Harvard Business Review put it when the Chief Data Officer was the new hot seat, the honeymoon "often ends sharply at about the 18-month period, when they are held accountable for achieving major transformational change." Michael Watkins built an entire discipline on the same idea for every executive transition: the mistakes you make in the first three months can sink the whole tenure.

Do the arithmetic. If the reckoning lands around month eighteen, and real change takes longer than that to show up in earnings, then the only lever you fully control is the first ninety days. That quarter is what buys you the rest. Waste it and you spend the back half of the honeymoon explaining a plan instead of pointing at a result. I cover why that precedent should worry every new officer in why Chief AI Officer hires fail.

You will be handed a title, not a mandate

Here is the trap almost everyone walks into. The company gives you the badge and quietly assumes the badge is the job. It is not.

IBM studied more than six hundred Chief AI Officers and found only sixty-one percent control their own AI budget. So roughly four in ten are on the hook for AI outcomes with no authority over the money that produces them. Most of them, fifty-seven percent in the same study, were promoted from inside, which sounds like an advantage and hides a trap: you already have the relationships, so everyone assumes you have the authority, and nobody thinks to actually hand it to you.

Heidrick and Struggles, who place these executives for a living, put the softer version plainly. "Too often," they write, "AI adoption within an organization becomes a responsibility that no one and everyone owns." They add that a lot of these seats are just older roles rebranded, which "blurs the definition of true AI leadership."

Read those facts next to each other. Four in ten cannot spend, many of them cannot kill, and the role itself is often a rename. That is not a job. That is a person to blame when it does not work. So your real first-day task is not a strategy. It is closing the gap between the title and the authority, while the goodwill is still high enough to ask. More on who should actually own AI in who should own AI, and on what the role owns in what a Chief AI Officer actually owns.

The ninety days, run backwards from a number

Most new officers run this forward. Strategy, then tools, then a pilot, then, if there is time left, a result. Run it backwards, starting from the result.

Days 1 to 30. Meter it, then follow the money. Before you do anything that feels visionary, put the meter and the gate in. It is cheap, it takes a week, and it is the only thing standing between you and a thirty-five-thousand-dollar surprise. Then listen. You get a grace period where you can ask the questions you will not be able to ask in month six, so spend it mapping one thing: where AI already touches money here. Not where it is technically interesting. Where it moves revenue, margin, cost to serve, or price. Sit with the people doing the work, not just the ones who bought the tools. Meet the CFO before you meet a single vendor, because the CFO already knows where the money is thin and will tell you in twenty minutes what a vendor spends a quarter obscuring. By month's end, write one page: what "AI turns into money" means at this specific company, this year. If you cannot write that page, you are not ready to spend a dollar, and you are nowhere near ready to hire.

Days 31 to 60. Get the mandate in writing, in three parts, then pick one fight. Vague authority is how the seat dies, so ask for it by name, and get it in an email a senior person replies "yes" to:

A reporting line to the CEO or the board.
A budget number you control, not one you request each time.
Written authority to kill any AI project, including ones you did not start.

That last one is the tell. If they will hand you a budget but not the right to kill things, you have a spending seat, not a leadership one, and you want to learn that in month two, not month eighteen. If they will give you none of it, you have your answer about the job, and it is better to know while you can still renegotiate or leave clean.

Once the mandate is real, pick one workflow to rebuild. One. Not a platform, not a bake-off, one workflow that touches the money you mapped in the first month. The value comes from redesigning how the work gets done, not from the software you bolt onto it, and treating the quarter as a procurement exercise is the most common way it gets wasted. That is one of the failure modes that ends these seats. While you are at it, kill one bad AI bet in the open. Turn off something everyone assumed was untouchable. Nothing makes a new mandate real faster than watching you use the kill switch you just asked for.

Days 61 to 90. Ship the proof twice, on a number the CFO already tracks. One win is a fluke. Two is a pattern, and a pattern is what earns you the next quarter. Instrument it so every run carries its own number: a per-customer cost to serve, a margin point, a piece of pipeline that closed. Then take that number to the board in the language they already speak, not a demo they can clap for and forget. The demo proves the model runs. The number proves someone got paid. That gap between a working model and a paying business is the whole game, and I wrote it up in the commercialization gap.

What burns the quarter

The ways this goes wrong are boring, predictable, and almost always self-inflicted. Open model access with no meter, which is how you get a thirty-five-thousand-dollar month with nothing to show for it. The ninety-page strategy nobody reads and the board quietly resents paying for. The tool bake-off that ends in a bigger bill and the same P&L. The governance framework built in month one, before there is a single thing worth governing. Chasing the frontier model when a cheaper one already clears the bar. And the quiet killer, the one that ends the most seats: ninety days in, you still cannot name the number you moved, so neither can anyone trying to defend you.

The day-90 scorecard

Three things sitting on the table at day ninety. One shipped commercial win, on real work, that someone other than you can point to. A mandate in writing, with a budget and the right to kill. And one metric the CFO now watches because of you.

Have those three, and the seat is real. You earned the rest of the eighteen months to make it matter, and you turned month eighteen from an ambush into a checkpoint you set the terms for. Have a strategy deck and a signed tool contract instead, and you are already behind, no matter how good the deck is.

The role was never about knowing the most about AI. It is about turning it into money before the goodwill runs out, with a meter running the whole time so your own spend never blindsides you again. Ninety days, one win, and a gate on everything, built in that order.

This is one spoke of the Chief AI Officer guide. Start with the pillar: Chief AI Officer: the complete guide.

Frequently asked questions

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

Put a meter and a gate on AI spend before anyone touches a model, secure a real mandate (a reporting line, a budget you control, and written authority to kill projects), map where AI already touches revenue and margin, and ship one measurable commercial win before the honeymoon ends. The AI strategy comes last, not first.

What is the biggest first-quarter mistake a new Chief AI Officer makes?

Two versions of the same mistake: handing out open model access with no meter, which is an easy way to burn tens of thousands of dollars with nothing to show for it, and treating the title as the mandate. IBM found only 61 percent of Chief AI Officers control their own AI budget, so authority is not automatic. Spending the quarter on a strategy deck instead of a governed, measurable win is what stalls the seat.

Should a new Chief AI Officer start with governance or a commercial project?

Both, in a specific order. Put the lightweight governor, a spend meter and an output gate, in first, because it is cheap and it stops the bleeding. Then go straight to one commercial project that moves a number. A full governance framework built before there is anything to govern is theater.

How does a Chief AI Officer get real authority in the first 90 days?

Ask for it in writing, by name: a reporting line to the CEO or board, a budget you control, and explicit authority to kill projects. Roughly four in ten Chief AI Officers do not control their AI budget, so it is not given automatically. If the company will not put the budget and the kill switch in writing, that is the answer, and it is better to learn it in month two than month eighteen.

What does success look like at day 90 for a Chief AI Officer?

One shipped commercial win on real work, a mandate in writing (a budget plus the right to kill), and one metric the CFO tracks because of you. A demo and a strategy deck do not count.

How is a Chief AI Officer's first 90 days different from a normal executive transition?

The role is new, so there is no settled playbook and the seat is watched more closely, and the AI itself can quietly burn money from day one if it is left ungoverned. The honeymoon is also short: the Chief Data Officer precedent shows transformational results get demanded around month 18, so the first quarter has to produce a governed, measurable win, not just a plan.

Sources

Executive-transition framework and the first-three-months thesis: Michael D. Watkins, "The First 90 Days, Updated and Expanded," Harvard Business Review Press (May 2013). Chief AI Officer budget authority (61 percent control the AI budget) and internal-promotion share (57 percent): IBM Institute for Business Value with the Dubai Future Foundation and Oxford Economics, "Solving the AI ROI puzzle" (July 2025), 600-plus Chief AI Officers across 22 geographies; figures are self-reported and correlational. AI-leadership ownership and role-relabeling quotes: Heidrick & Struggles, "2025 AI, Data, and Analytics Officers Report" (published February 2026). Chief Data Officer tenure and the eighteen-month honeymoon: Tom Davenport, Randy Bean, and Josh King, "Why Do Chief Data Officers Have Such Short Tenures?", Harvard Business Review (August 2021); the roughly two-and-a-half-year average tenure is corroborated by MIT Sloan Management Review and Korn Ferry C-suite tenure data. All survey figures are self-reported by executives, not audited results.

About the author

Jeff Brokaw is a Certified Chief AI Officer and technical-commercial operator who rebuilds the commercial layer of technically complex companies and ships AI in production, not slideware. He runs a multi-model agent system in production, governing where a model's guess becomes a company commitment, the exact mandate this article says most hires never get. He ran the commercial side of two media companies tied to more than $850M in associated exits, took an AI fintech from zero to $1M ARR in nine months, and authored the go-to-market behind a $114M institutional raise that came together in under 30 days.