Appointed this year, with no team: a 90-day plan for a new Chief AI Officer

If you were handed the Chief AI Officer title this year, you have plenty of company. What you probably do not have is a team.

What the data says

On September 12, the contact-data company Lusha published an analysis of people holding the Chief AI Officer title in its database. Of 1,180 incumbents it found, 532, or 45%, had been appointed since September 1, 2025. Lusha says every count is a floor: it matched exact titles only, and the newest job changes are the least likely to have shown up yet. This is company-reported vendor data, so treat it as a direction, not a census.

Two more findings from the same report matter for how the job is shaped:

• About two-thirds of AI leadership titles sit in companies with fewer than 200 people.
• The share of recent appointments is highest in companies of 200 to 10,000 people, at 52% to 53%, against 41% in small companies.

Put together: the role is new, it often sits in small companies, and mid-size companies are adding it fastest.

A second report from the same week shows what tends to be missing when they arrive. Gartner’s survey of internal audit, covered by Help Net Security on September 15, found that 93% of audit leaders and auditors report some level of AI use, while 54% of the chief audit executives polled have not started measuring the value of it. That is one function, not a whole company, but the shape is familiar: use comes first, and measurement comes much later.

So a new CAIO usually walks into AI that is already in use and not yet measured, with no staff to fix either. The plan has to fit one person.

A 90-day plan for an office of one

Each 30-day stretch ends with one document you can hand to your CEO. If you only finish those three documents, you have done the job’s foundation.

Days 1 to 30: see what is running.

• Ask every department head one question: which AI tools does your team use, whether or not the company pays for them?
• For each tool, record four things: who owns it, what data it sees, whether it can act on its own (send, pay, change a record), and whether anyone approved it.
• Write a two-page acceptable-use policy: approved tools, data that must never go in, when a person must check the output, and who to ask. Have counsel review it before it goes out.
• Deliverable: the inventory and the published policy.

Days 31 to 60: choose where to go first.

• Collect the use cases people are asking for. Expect more than you can handle.
• Score each one with the rule below.
• Pick no more than three to support this quarter. Say no, or not yet, to the rest in writing, with the reason.
• Deliverable: a ranked list with the top three and their named owners.

Days 61 to 90: prove it and report it.

• For each of the three, write down before you start what “better” means: hours saved, turnaround time, error rate, or something your CEO already tracks.
• Agree a one-page monthly report with your CEO and send the first one.
• Deliverable: the first monthly report.

A scoring rule you can use tomorrow

Score each use case from 1 to 3 on four questions:

1. Value. If it works, does it change a number the CEO already watches? (3 = yes, clearly; 1 = nice to have)
2. Data. How sensitive is what it touches? (3 = public or internal only; 1 = customer, employee or regulated data)
3. Action. Can it act without a person approving each step? (3 = drafts only; 1 = it sends, pays or changes records)
4. Effort. Can it run on tools you already have? (3 = yes; 1 = new vendor and integration)

Add the scores. Start with the highest totals. One override: anything that scores 1 on both Data and Action goes to a formal review before it goes anywhere else, whatever its total. That keeps a high-value idea from walking past the two questions a review exists to ask.

Worked example (illustrative, not a client). Picture a 400-person distributor. Sales wants an assistant that drafts follow-up emails for a rep to send: Value 3, Data 2, Action 3, Effort 3, total 11. Finance wants an agent that pays approved invoices on its own: Value 3, Data 1, Action 1, Effort 1, total 6, and it triggers the override. The first goes into the quarter’s top three. The second gets a review and, at most, a supervised pilot later.

What to do this week

• Send the one-question inventory request to every department head.
• Block two hours to draft the acceptable-use policy, and book time with counsel to review it.
• Ask your CEO which three numbers they would want to see from AI a year from now. Build your scoring and your report around those.
• Put the three 30-day deliverables on your calendar now, so the plan survives the rest of your calendar.

You do not need a department to run an AI office. You need a short list of what is running, a rule for what comes next, and a report someone reads every month.