Fractional AI Leader, Consultancy or Head of AI

Between $18,895 and $405,541, for the same job description. That is a 21x spread on one line item, and almost none of it comes from how much work gets done.

Those four bars are four different products wearing one title. At $18,895 you are buying about thirty days of outside opinion a year from somebody with no authority over anything. At $405,541 you have a named executive with a seat at the leadership table who cannot start for four months. The middle of the chart is where most companies with fifty to three hundred people actually live, and it is the part nobody prices honestly.

Because rates are the product. Consultancy.uk, which tracks the industry for a living, puts it plainly: firms treat their fee structure as a trade secret, which is why the published ranges you find are either somebody's own rate card or a guess dressed up as research.

So skip the rate cards and go to the records where somebody was legally obliged to write the number down.

The first is a federal pay table. NIST publishes what the US government will pay an expert or consultant per day, pegged to the OPM General Schedule and effective 11 January 2026. The ceiling for a GS-15 grade expert is $629.84 a day. That is the most the federal government will hand a senior outside advisor for a day of work without special justification.

The second is bankruptcy court. When a company files Chapter 11, its advisors have to disclose their rates to the court, and Debtwire pulled those disclosures across 38 large cases. Alvarez and Marsal billed $1,550 an hour for a single professional in Northvolt AB. FTI Consulting billed $1,525 an hour in Big Lots. AlixPartners posted the highest blended rate of any debtor-side financial advisor at $1,079 an hour, which is the honest number to reason with because a blended rate is what a firm actually averages once the juniors are counted.

Run those two records against each other and the spread is absurd. A day of senior consulting is $629.84 if the buyer is the US government and $8,632 if the buyer is a company in front of a bankruptcy judge. Same grade of person. Same eight hours. That 13.7x gap is not a quality gap, it is a bargaining-power gap, and it is the entire reason a retainer quote is meaningless until you know who is on the account.

So ask any consultancy one thing before you sign, and it isn't the price. Ask who exactly does the work, and at what grade. The answer decides which end of that spread you are buying. We wrote out the arithmetic that turns any monthly quote into a cost per working day separately, because it is the one calculation that survives contact with every provider in this category.

$84,296, roughly, and nothing has been built yet.

That is the retained search fee. Executive search firms bill 25 to 33% of the candidate's estimated first-year compensation, staged across engagement, shortlist and completion, with expenses pushing the top end toward 35%. Built In puts US technology leadership base pay at $224,550 with another $56,435 in additional cash, so first-year cash lands near $280,985 and a 30% fee comes to $84,296.

Then the salary starts. Base is not the cost of an employee: the Bureau of Labor Statistics priced benefits at 30.1% of total employer compensation for private-industry workers in March 2026, with wages making up the other 69.9%. Divide $224,550 by 0.699 and the loaded seat is $321,245. Add the search fee and year one is $405,541.

None of that includes the tooling, the model spend, the two engineers this person will immediately ask for, or the twelve weeks of ramp before they know which of your systems is load-bearing. That is the cost of the chair, not the cost of the work.

And the fee is the smaller problem.

Twelve to sixteen weeks of active search for a C-suite hire, and that is the clean version. Talentfoot's 2026 placement benchmarks put a director-level search at six to twelve weeks, a VP at eight to fourteen, and a C-suite role at twelve to sixteen or more, measured from the intake brief to an accepted offer. Their advice is to budget about four months for the top of that range, because board involvement and scope changes stretch it further.

An accepted offer is not a start date. Notice periods sit on top. Then ramp sits on top of that.

Meanwhile the other two routes have a search window of zero. That gap matters more than the money to most companies, because budget survives a four-month wait and momentum does not. Every week the seat stays empty, another department buys another tool and starts another pilot nobody will own.

Every provider in this category says the same sentence: consultants recommend, we own it. It is a good sentence and it is unfalsifiable, so it tells you nothing. Replace it with three questions that have a yes or a no on the end.

Who signs the AI budget? If the answer is a committee, nobody owns AI and you have bought advice. IBM's Institute for Business Value surveyed 2,000 CEOs across 33 geographies between February and April 2026 and found 76% of organizations now have a Chief AI Officer, up from 26% a year earlier. Fifty points in twelve months isn't fifty points of new function. Much of it is a title landing on somebody's existing job description. The budget question separates the two.

Who can kill a department's pet pilot? This is where consultancies structurally cannot compete, and it is not a criticism of consultancies. The department that funded the pilot is also the client, and no retained firm kills its own sponsor's project twice. A part-time leader with a written veto will, because the veto is what you hired.

Who is still in the room in month seven? A consultancy is there if the statement of work was renewed. A full-time hire is there, though month seven is roughly when they arrive if you started the search today. A fractional leader is there on the same cadence they started on, which is the quiet advantage of a retainer relationship over a project one.

Every engagement ends. What survives it is the part nobody prices.

A full-time hire leaves everything and takes the context. The systems stay, the vendor relationships stay, and the reasoning behind eighteen decisions walks out with the person, because it was never written anywhere except their head. Good hires do this too. When one person is both the decider and the documentation, there is no second copy.

A consultancy leaves a deck and an invoice trail. Sometimes it leaves working software, in which case ask hard about who holds the repository and the model credentials. A firm whose commercial model depends on the next statement of work has no reason to make itself unnecessary, and the good ones will tell you that outright.

A fractional engagement is judged on exactly this. The engagement is supposed to make itself smaller: a written roadmap, a decision log your CFO can read, evaluation harnesses on whatever is live, and a named internal owner for each system. If month nine looks identical to month three, the arrangement failed regardless of what shipped.

Ask every provider what you keep on the last day. Then ask to see one of those things, from a client they already have. The pause tells you more than whatever comes after it.

Each route has one characteristic failure, and knowing which one you are exposed to is worth more than shaving 10% off a quote.

The full-time hire fails on utilization. You budgeted $405,541 for a leader and then handed them a backlog that needed an engineer, so a $321,000 executive spends month four writing Python. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027 on escalating cost, unclear value or weak risk controls, and the same firm reckons only about 130 of the thousands of vendors claiming agentic products are real. A leader who cannot tell those apart is expensive in a way the salary line never shows. If the shortfall is build capacity rather than judgment, an engineer embedded in your team is the cheaper correct answer and it isn't close.

The consultancy fails on scale mismatch. Accenture booked $19.3 billion in new work in the third quarter of its 2026 fiscal year and logged 104 client bookings of $100 million or more year to date. Nothing about an operation that size is built to care about a retainer the size of the third bar on that chart, and the partner who sold it won't be the person on your calls.

The fractional route fails on refusal to delegate authority. If you will not hand over the budget and the veto, you have bought a very expensive opinion, and Deloitte's numbers say that is common: only 21% of organizations have mature governance for autonomous agents even while three quarters plan to deploy them within two years. Governance without authority is a document.

Run the questions in this order and two of them usually end the conversation before you price anything.

Start with the product question. If AI sits inside the thing you sell, hire full time and stop reading comparison posts, ours included. Nobody part-time is going to own a capability your customers pay for. The $405,541 is the price of the category you are in.

If AI is something you run rather than something you sell, ask whether you will actually hand over a budget and a veto. If the honest answer is no, you do not want a leader at all, you want a decision. Buy a short audit that ranks what to automate first and skip the retainer entirely. This is the cheapest correct answer in the whole comparison and almost nobody recommends it, because it is worth a few thousand dollars to whoever recommends it.

Only if both gates pass does the fractional route become the right shape, and then one more question splits it: do you need things built, or only sequenced? A fractional AI officer who hands you a roadmap and no delivery capacity leaves you exactly where the roadmap says you should not be. That is why the leadership seat we run sits next to build capacity rather than instead of it.

There is a fourth case the whole category ignores, and it is more common than any of us admit: you do not have an AI leadership problem, you have two workflows that are eating a person a week.

Buying leadership for that is like hiring a head of transport because the delivery van needs a service. Get the two workflows built, watch them run for a quarter, and then work out whether the pattern repeats often enough to need somebody sequencing it. Our own numbers make the size of that difference obvious: a scoped starter build runs $1,500 to $2,500 fixed and a two-week production sprint is $5,000 fixed, against $60,000 for a year of the leadership seat at our retainer floor. If two builds solve it, twelve months of leadership is procrastination with a budget line.

The test is whether you can name three distinct AI decisions that need making in the next two quarters. Not three tools you want. Three decisions, each with a real trade-off and a real owner who currently does not exist. Fewer than three and you are early. More than five and you probably needed someone six months ago. The full set of situations where the fractional model is the wrong purchase runs to six, and this is only one of them.