Corporate AI enablement is priced three ways, and the three are not variations on a theme. Per-seat buys access for a headcount. Per-cohort buys delivery hours for a group. A retainer buys an ongoing presence.
Pick the wrong one and you pay a fair price for something that was never going to move your team’s output. Most pricing guides answer the amount question and skip the structure question, which is the one in front of you when three quotes arrive in three shapes.
The three AI enablement pricing models
| Structure | Unit you pay for | Scales with | Fails when |
|---|---|---|---|
| Per-seat | A named person’s access, yearly | Headcount | Access was never the constraint |
| Per-cohort | A fixed block of delivery hours | Sessions and roles | The room is half empty |
| Retainer | Monthly availability of a facilitator | Calendar months | Nobody inside has ownership |
Per-seat is what every catalog platform uses, and the only model that publishes a number. Coursera shows a rate for its self-serve Team plan up to 499 users, then stops: at 500 users and above the price is “tailored to your company.” A company of 50 to 500 people sits on that seam, which is why two firms of similar size quote wildly different per-head figures and both tell the truth.
Per-cohort inverts the arithmetic. You buy a session, or a series, and the marginal cost of the twelfth attendee is zero. Price moves on how many groups need their own material, not how many badges exist in the building. The fee covers preparation, hours in the room, and any follow-up, which is why two cohort quotes at the same headline price can differ by half in hands-on time.
Retainer pricing buys months, the standard shape for embedded delivery work and the least standard shape for teaching people. For dollar figures by team size, see the companion piece on what an AI enablement program costs.
What each model pays your vendor for
A price structure is an incentive contract. Read it that way.
Per-seat pays for enrollment. Revenue is complete the moment the seats are provisioned. Whether anyone logs in is your problem, and every renewal talk is about seat count rather than skill. Not bad faith, just what the contract rewards, and it is why platform vendors invest hard in catalog breadth and lightly in whether a given person can now do a given task.
Per-cohort pays for a filled room. The vendor wants attendance, and wants the session to go well enough that you book the next one. A session that teaches nothing does not get rebooked. The weakness is the cliff: once the last cohort runs, nobody has a stake in the habit.
A retainer pays for presence. Month eleven bills the same as month two. With no clause handing capability to named people inside your company, the structure rewards your dependence.
Buy the one whose rewarded behavior matches the outcome you need.
Where per-seat stops being cheap
Per-seat always wins on cost per covered head. That is arithmetic, not an argument: a few hundred dollars a year per person covers a whole company for less than any live program covering a third of it. It loses on cost per hour of supervised practice.
Take a 120-person company. A catalog subscription covers everyone, but nobody is scheduled and the material is optional. A cohort program covering the same 120 people runs four half-day sessions, one per role group, inside the market band for that size.
Per covered head the catalog is several times cheaper. Per hour of scheduled work on your own documents it buys nothing.
The crossover is not about company size. It is about what you are buying:
- Coverage (everyone gets reference material, self-directed learners use it): per-seat is correct and cheap.
- Capability (output has to change in named roles by a date): per-cohort is correct, and per-seat will not get there at any price.
The failure mode is buying coverage while believing you bought capability: licenses in place, dashboards fine, work unchanged. Why that gap persists is the case for treating enablement as its own line item.
One number that makes quotes comparable
Reduce every quote to cost per trained person per supervised practice hour. Three shapes, one unit.
ATD’s 2026 State of the Industry report draws on 2025 data from 340 organizations. It put average direct learning spend at $846 per employee and formal learning hours at 16.7 per employee per year. Divide one by the other: about $51 per formal learning hour, across every subject a company trains on. Hold your quotes against that.
| Quote | Spend | Supervised hours | Cost per person-hour |
|---|---|---|---|
| Catalog seats, 120 people | mid five figures yearly | 0 scheduled | undefined, recurring |
| Four half-day cohorts, 120 people | low five figures once | 480 person-hours | about $25 |
| Retainer, three months, two facilitated hours a week | varies | 24 facilitator hours | compute and ask |
A cohort program near $25 per person-hour buys scheduled practice at half the market rate. A catalog seat has no denominator until somebody sits down with it, and a person who completes two hours in a year paid more per hour than the live session cost.
Ask every vendor for that denominator: hours in the room, hours of preparation against your material, hours of follow-up, each as its own line. A blended fee hides which of the three you are short on, and it is almost always the last.
Why retainers are rare in training
A 2026 pricing benchmark from AI Enablement Insider surveyed 38 firms in this market, 26 of which sell training, and asked how they charge. Retainers and subscriptions were used by 27% of the training firms, against 38% of consulting lines and 39% of implementation lines. The sample skews UK and small, so read it as direction rather than a US census.
The direction is clear. Retainers have settled on work that is ongoing by nature, which building and running systems is, and have been declined for teaching, which has a natural end. Six of the seven firms using retainers also sold implementation. The retainer follows the build, not the training.
The same survey found the per-head spread you would expect from two products wearing one name. Among specialists reporting rates, four of six charged £1,000 or more per head, while two of five full-stack firms sat at £100 to £300. Cohort prices above roughly £15,000 and day rates above £2,000 belonged only to the specialists, and nine of fourteen full-stack firms ran workshops of a day or less.
A cheap per-head number here is usually a short session attached to a larger sale, and the shape of the quote says so before the content does. See what enablement services include and quietly exclude.
The benchmark finance will use against you
Direct learning spend per employee fell from $1,254 in 2024 to $846 in 2025, a drop of about a third, while formal learning hours per employee rose from 13.7 to 16.7. Companies bought more hours for less money, which ATD attributes partly to virtual classrooms cutting travel and to leaner internal teams.
That divergence is your approval memo. A quote that buys scheduled hours moves with the market; a quote that buys optional access is buying something else. A program covering 40 people at a low five-figure fee works out to a few hundred dollars per trained person, a fraction of the $846 an average employer already spends per head. Framed as a share of an existing benchmark, it is a routine request.
Hybrids that hold up, and the clause that matters
Two combinations earn their keep.
Cohort first, thin retainer second. The fixed fee builds the skill; a small monthly block of facilitation hours keeps it alive through the first quarter. Price that block in hours, not access, and let it expire by default.
Cohort fee plus per-seat reference access. Live sessions for the roles whose output has to change, catalog seats for everyone who wants to browse. The answer to “seats or sessions” is usually both, at different depths.
One clause does most of the work in either case:
Within the engagement, the provider will train and certify at least two named employees to deliver the core curriculum, and will hand over all session materials, exercises, and facilitation notes in editable form.
A vendor who agrees is selling capability. One who declines is selling a dependency with a monthly invoice. Our criteria for shortlisting an enablement partner name the artifact to demand for each promise.
Which AI enablement pricing model fits your situation
| Your situation | Structure | Why |
|---|---|---|
| Licenses bought, usage fine, output unchanged in two or three roles | Per-cohort | Only structure buying scheduled practice on your own work |
| Large scattered group, self-directed, reference material wanted | Per-seat | Lowest cost per covered head; depth is not the goal |
| One function has to change by a fixed date | Per-cohort, one role group | Concentrates hours where the outcome is measured |
| Program delivered, habit fading at week six | Thin retainer, hours-priced, expiring | Reinforcement is time-boxed, not permanent |
| No internal owner named yet | None yet | Fix ownership first; every structure fails without it |
| Automation work after enablement | Retainer or project, apart | Different product and budget |
The fifth row is the one people skip. If the work has landed on a capable person as a second job with no set-aside hours, no pricing structure saves it. A short internal diagnostic answers more than a paid scorecard does. The piece on what readiness assessments measure and what they miss sets out how.
Four questions that open up a blended quote
Most proposals arrive as one number. Four questions open it up.
- If our headcount doubled next month, what happens to this price? If it doubles, you are being sold seats whatever the cover page says. If it moves only when a role group is added, it is a cohort price.
- How many hours sit in front of our people, and how many in preparation against our documents? A vendor who cannot split the fee has not costed the work.
- What is in scope between the last session and the end of the contract? Silence means the cliff, where decay happens.
- Which two of our employees will run this without you afterward? That answer separates a program from a subscription.
Survive all four and the structure is sound.
Frequently asked questions
What are the pricing models for AI enablement?
Three, plus hybrids. Per-seat charges a yearly rate for each named person’s access. Per-cohort charges a fixed fee for a block of delivery hours, with the marginal attendee free. A retainer charges monthly for availability. Catalog platforms use the first, live providers the second, embedded firms the third.
Is per-seat or per-cohort cheaper?
Per-seat is cheaper per covered head, every time. Per-cohort is cheaper per hour of supervised practice, usually by a wide margin, because per-seat buys no scheduled hours. Decide which denominator matters and the comparison answers itself.
Should we pay a monthly retainer for AI training?
Only for time-boxed reinforcement after a program, priced in facilitation hours and set to expire rather than renew. A 2026 benchmark found retainers at 27% of training firms against 39% of implementation lines, so an open-ended training retainer is a shape the market has moved away from.
How do I compare three quotes in three different formats?
Convert each to cost per trained person per supervised practice hour. Ask every vendor to split the fee into preparation, delivery, and follow-up hours, then hold the result against the roughly $51 per learning hour implied by the ATD 2026 figures.
Do catalog seats like Coursera or Udemy Business replace a workshop?
No. Seats buy optional access to recorded courses, which suits self-directed learners who know what they want. A program buys scheduled practice on your own documents with a named owner afterward. If licenses are in place and output has not changed, more optional access does not address it.
Does the price include our AI licenses?
No. Seats for ChatGPT, Claude, Gemini, or Microsoft Copilot are contracted with the tool vendor and billed separately. Microsoft 365 Copilot lists at $30 per user per month on an annual commitment, on top of a qualifying base license, so the license line can exceed the services fee.
Can we negotiate the structure rather than the price?
Yes, and it is the better negotiation. Turning a per-seat proposal into a cohort fee for a narrower group, or trimming an open-ended retainer to a three-month block of hours, beats a discount on the original shape.
Who should own this internally once the contract ends?
Someone with standing to change how work gets done in one function, with hours formally set aside: usually a COO, a VP of Operations, a chief of staff, or a head of learning. Naming that person before you sign makes the handover clause enforceable.
Key takeaways
- Per-seat buys access, per-cohort buys delivery hours, a retainer buys months. Three products, one label.
- Each structure pays your vendor for its own behavior: enrollment, a filled room, or continued presence. Buy the incentive that matches your outcome.
- Per-seat wins on cost per covered head and loses on cost per supervised practice hour.
- Normalize every quote to cost per trained person per practice hour, against the roughly $51 per learning hour implied by ATD’s 2026 figures.
- Retainers are the rarest structure in training and among the most common in build work. Treat an open-ended one as a question, not a default.
- One clause carries most of the weight: two named employees trained to deliver the curriculum, with editable materials handed over.
If you have quotes in front of you and cannot tell which structure you are being sold, bring them to a free AI-readiness call. We will split the fees into hours, name the model behind each, and say which role groups need scheduled practice. For the wider sequence from license rollout to measured capability, start with our AI enablement playbook.
Arthur Wandzel
