Fractional AI Officer Rates: Day Rates, Monthly Retainers and What Sits Inside Each

Start with ours, because we publish them and most of this market does not. A one-off starter build is $1,500 to $2,500, fixed. A two-week production sprint that puts one workflow live is $5,000, fixed. Retainers start at $5,000 a month. That is the entire ladder, it sits on our pricing page, and nothing about it moves depending on how the discovery call goes.

Now the honest caveat. Those are our numbers, not the market's. One shop's rate card dressed up as an industry benchmark is the same trick as the ranges, just with better manners. What we can hand you instead is the arithmetic, and the arithmetic works on any quote you get, ours included.

The unit that matters is a day of senior attention. Not a month, and not whatever the tier happens to be called. Once every quote in front of you is expressed as dollars per working day, they become comparable for the first time. Some of them stop looking like a bargain the moment they are.

Because the ranges are missing the denominator. Five specific problems, and each one is enough on its own to make a number useless:

• No unit of time is attached. A retainer is quoted per month. A month is not a quantity of work. Two days a month and three days a week are both "a monthly retainer" and they differ by a factor of six.
• The currencies and units are mixed. Day rates get quoted in one currency, retainers in another, hourly rates in a third, all inside the same article, and nothing is normalized.
• There is no methodology. No sample, no source, no count of engagements behind the band. The number is a recollection of what somebody once charged.
• Every publisher is a seller. We are one too, which is exactly why our own prices go in a chart with a citation rather than in a sentence you have to take on faith.
• Nobody publishes the exclusions. Tier lists are all inclusions. The line that decides your actual annual spend is the one nobody prints.

This is a support post inside a bigger argument. If you are still working out whether the seat belongs in your company at all, what it owns and who should not hire one yet, read our piece on the fractional AI officer first and come back here for the money.

Divide the monthly number by the days actually committed in writing. That is the whole method, and it takes ten seconds. The reason it works is that it forces the second number into the open, and the second number is the one the quote was designed to leave vague.

Do the same to the full-time route so you have something to compare against. Built In puts the average US chief technology officer at $224,550 in base pay, which is a fair proxy for the most senior technical leader at a mid-market company. Base pay is not what the seat costs you. The Bureau of Labor Statistics measured wages and salaries at 69.9% of total employer cost in private industry in March 2026, with benefits taking the other 30.1%. Gross the base up by that share and the seat costs $321,245 a year before a recruiter fee, before equity, and before the months it sits empty.

Then count the days honestly. 2026 has 261 weekdays. Eleven US federal holidays land on a weekday, and a normal PTO allowance takes fifteen more. Call it 235 working days, and a day of that person costs $1,367.

That is the reference point. Any quote sitting above it is charging you a premium over a permanent hire for a part-time version of the job, and the provider should be able to say out loud what the premium buys.

Three shapes, and the tier names in this market almost always map onto one of them. What separates them is not seniority. It is capacity, and specifically whether anybody attached to the retainer is allowed to build.

Advisory buys you a standing call, a roadmap document, vendor shortlists, a board briefing, and a named person your team can ask. It is real work and it is worth paying for when the problem is genuinely that nobody has decided anything. It ships nothing.

Embedded leadership adds the parts that only work if someone has authority: sequencing the backlog, killing the pet projects, writing acceptance criteria that a vendor can be held to, and running the weekly where those decisions get made in front of everyone. Still ships nothing, but the decisions are now good enough that shipping is possible.

Leadership plus build puts engineers behind the roadmap the same person wrote. This is the shape we sell, and it is why our retainer floor is where it is. It is also the only one of the three where month three produces something running against real data instead of another document.

The building. Almost always the building.

A leadership retainer produces decisions. Decisions are not systems, and the gap between the two is where the second invoice lives. In our experience it is the bigger of the two, it lands a few months after the first, and it is the line that was missing from the model the buyer approved.

So price it up front. On our ladder, one scoped automation built and shipped is $1,500 to $2,500 fixed, and a full workflow live in production is a $5,000 two-week sprint. Two of those over a year is $10,000, which is two months of the retainer floor. If a provider cannot give you an equivalent number, what they are quoting is not the cost of the program, it is the cost of the first stage of it.

Three other lines get left out often enough to be worth naming. Model and inference spend stays on your card, and it moves when a pilot gets popular. Licenses for whatever the roadmap says to buy stay on your card too. And evals, the tests that tell you whether the thing is still right six weeks after launch, are a build item rather than a leadership item, so they fall through the crack between the two invoices more often than anything else on this list.

Five things, in rough order of how much they move the number.

Whether build capacity is attached. This is the big one, and it moves the number in the direction people do not expect: a retainer with engineers behind it costs more per month and less per unit of delivered work. The custom builds we ship come out cheaper per working day than the meetings that decide on them.

Committed days versus available days. "Available for questions" costs almost nothing to promise. Four days a month, named in the contract, costs real capacity, and it is priced accordingly. If a quote is cheap, check which one you bought.

Regulatory surface. Health, finance and anything with a regulator attached carries review, documentation and sign-off work that a general engagement never touches. It is real work, it is slow, and it gets priced.

Accountability. A provider who signs up to outcomes prices differently from one who signs up to attendance, because one of them can lose. That premium is the most defensible one on the list.

Who is actually doing the work. Ask whether the senior person on the call is the senior person on the account in month four. If the answer is soft, you are paying a day rate for a name and getting a different person's days.

Ask these in the order they are written. Each one exists because the answer to it changes the number, and because a provider who cannot answer it in a sentence has not costed their own offer.

1. How many days a month, in the contract? Not "typically", not "as needed". A number, written down. Everything else on this list is downstream of it.
2. Who is the named person, and are they on the account in month four? Get the name in the agreement. Turnover on a fractional seat is the quiet way a day rate doubles without the invoice changing.
3. What does month three produce? A running system, a sequenced backlog, or a document. All three are legitimate answers. Only one of them is worth a build-tier price.
4. Who builds what you decide, and what does that cost? If the answer is "a partner" or "your team", ask for the number anyway. An unpriced second invoice is not a smaller invoice.
5. What is explicitly excluded? Licenses, model spend, evals, integration work, out-of-hours incidents. Make them say it out loud, then put the list in the agreement.

Run a quote through those five and the range collapses into a price. Which is the whole reason this page exists.

When you do not have enough decisions to fill the days.

A company running one AI pilot, with a technical leader who has capacity and an owner who already has budget authority, does not need to buy leadership by the day. It needs somebody to build the thing. Buying four days a month of strategy for a backlog with three items on it is how a $60,000 year turns into a folder of slides, and we would rather tell you that now than invoice you for it later.

The reverse case is just as real. If AI decisions are already being made across four departments with no arbiter, no rate card saves you, because the cost of the wrong sequence is bigger than the cost of the seat. That is the situation where part-time AI leadership pays for itself in the projects it stops rather than the ones it starts.

The test we use is countable rather than a feeling. More than one pilot live, no single owner, and a willingness to hand that owner budget and a veto. Miss the third condition and no day rate is the right day rate, because you would be buying advice nobody in the building is obliged to take.