Buy one real session, for money, with a cohort that looks like your company, and agree the company-wide price per head before you find out whether it worked. A pilot workshop is not a cheaper rollout and not a test of whether AI helps your team. It is an audition, and the vendor is what gets tested. Treat it that way and a few thousand dollars tells you whether a fifty-thousand-dollar commitment is safe. Treat it as a free trial and you have bought a sales demo.
Search this question and you get pages about piloting software — useful pages, but they answer how to test whether a model earns its place in a workflow. You are not testing a model. You have the licenses, and what you are uncertain about is a facilitator, a curriculum, and whether any of it survives contact with four hundred people. Whether to buy that capability or build it in-house is worked through in our guide to buying versus building AI enablement capability, and the reason license counts stopped standing in for readiness is argued in the case against buying licenses and calling it capability.
What a Pilot Workshop Actually Tests
Every published pilot framework assumes the technology is the variable. For a training purchase it is not. ChatGPT, Claude, Gemini, and Microsoft Copilot all work. The uncertainty sits on the other side of the table, and a pilot workshop answers four questions about a supplier:
- Can this facilitator hold a room of your people? Not volunteers at a conference. Your skeptical operations manager, your two people who have never opened an assistant, and your one enthusiast who will try to take over.
- Does the material come from your work or from their deck? A session built on your contracts, your reports, and your weekly rhythm teaches something transferable. One built on generic examples teaches that the examples were generic.
- What happens after the last slide? Most of what you are buying is the week after. A pilot shows whether anything exists there.
- Will they tell you the truth about the result? A vendor who reports that two of your twelve participants did not change their habits is one you can scale with. A vendor who reports a satisfaction score is selling the next session.
None of the four is answerable from a proposal, a case study, or a reference call. All four are answerable from one session you paid for.
Why the Pilot Advice You Find Does Not Fit
The standard guidance is sound for its own subject and misleading here. Three pieces of it break on contact with a training purchase.
“Collect four to six weeks of baseline data.” Reasonable before you route support tickets through a model. Impossible when the CFO wants the program decided this budget cycle. A training pilot gets its baseline the cheap way: pick one task the cohort already repeats weekly, have them timestamp it twice before the session, and accept that the measurement is coarse.
“Scope tightly to one workflow with one owner.” Correct for an agent, actively harmful for a workshop. Narrowing to your most AI-ready team produces a result that cannot survive multiplication. Centric Consulting describes the failure mode well: the conditions behind a pilot win, clean data and one determined champion, tend to disappear at scale.
“Run eight to twelve weeks.” That is a software pilot’s clock. A facilitated session plus a measurement window lands closer to six to eight weeks, and most of that is waiting rather than working.
The most-cited number in this field needs the same care. MIT’s NANDA initiative reported in The GenAI Divide: State of AI in Business 2025 that roughly 95% of enterprise generative AI pilots stall before reaching scaled adoption. It is a working paper with a modest sample, and “failure” meant the absence of rapid profit-and-loss impact. Read it as a warning that pilots go nowhere by default, not as a probability for your own.
Pay for It
The instinct is to ask for a free session. Resist it, for three reasons unrelated to being generous to a vendor.
A free session is a demo, and a demo is pre-built for somebody else. The distinction is drawn plainly in LearnFrame’s guide to training program costs: a demo is something the supplier already built for another buyer, while a pilot runs on your content and your review process. The same guide notes that most suppliers will build a first piece at a reduced rate before you commit, and that if nobody has offered, you should ask.
A free session also selects for the wrong person. Unpaid delivery comes out of a sales budget, so the firm sends whoever closes best. You get an excellent ninety minutes and learn nothing about the facilitator who would run your rollout.
And it costs you the right to complain. Paying buys standing: you can ask for the agenda to change, for your own documents to be used, for a named facilitator in the agreement, and for an honest write-up. Those asks are reasonable from a client and awkward from a prospect.
The sum is small against what it settles. ATD’s 2025 State of the Industry report put average direct learning expenditure at $1,254 per employee for 2024. A pilot at a few hundred dollars a head sits inside one person’s annual learning budget, spent to decide the shape of everyone’s.
What a Pilot Should Cost
Market pricing for a facilitated AI enablement session runs roughly $2,000 to $15,000, set mostly by length, group size, and how much material is built from your own documents. A pilot sits in the lower half of that band: one session, small group, not a program. Three reference points to price against, verified on their own pages in October 2026.
| Benchmark | Figure | What it tells you |
|---|---|---|
| Market range, facilitated session | $2,000–15,000 | A pilot above the midpoint is a program, not a test |
| Pryor Learning, public AI Fluency seminar | $199 per seat, three hours, $20 off for groups of three or more | The floor for open-enrollment seats with no customization |
| ATD average cost per learning hour used, 2024 | $165 | The industry’s own per-hour yardstick |
The last two are the useful pair. Open-enrollment seats are cheap because nothing is built for you: the catalog session teaches a method and your documents never enter the room. A private pilot at $5,000 for twelve people is about $417 a head, roughly $139 per learning hour — under the ATD average, for material built on your own work.
Then add the room. A $90,000 salary carries roughly 30% in employer burden, about $56 an hour across a 2,080-hour year. Twelve people for three hours is 36 hours, close to $2,000 in payroll on top of the fee. The same session for a 300-person company is 900 hours, or about $50,000 before anyone invoices you. That ratio is the argument for piloting: the fee scales linearly and the payroll scales brutally. Whether to pay at all, now that free material has gotten genuinely good, is worked through in our comparison of free AI training against paid workshops.
Keep the Pilot and the Rollout Comparable
A pilot predicts the rollout only to the degree the two resemble each other. Four things must be held constant, and vendors quietly vary all four.
The facilitator. Name them in the agreement. Firms that win on a founder’s delivery and staff rollouts with contractors are common, and the pilot is where you catch it.
The source of the material. If the pilot uses your documents and the rollout uses a standard deck, you measured something you will not receive.
The session length. A three-hour pilot does not predict a ninety-minute rollout session. Shorter is not proportionally worse; it is differently worse.
The mix of people. Buyers get this one wrong most often. A cohort of your keenest adopters produces a flattering number and no information. Build the cohort to look like the company: a couple of enthusiasts, a majority of ordinary users, and at least two who are quietly unconvinced. The quiet two are the most valuable seats in the room, because they are the margin on which a rollout succeeds or stalls.
What a single session cannot test is whether the capability is still there in March. The trade-off between a platform’s drip of content and a facilitated session’s sharper, shorter half-life is examined in our look at retention across the two formats.
What to Fix in Writing Before It Runs
Five items, agreed before the session, each far harder to settle afterward.
1. The rollout rate. The clause nobody writes about, and the one that pays for itself. Fix the company-wide price per head now, while the vendor does not yet know the pilot succeeded. A pilot that goes well without a pre-agreed rate is a negotiation you enter having already told the vendor you want what they sell. Ask for part of the pilot fee to be credited against that commitment while you are at it; many suppliers will credit half.
2. The three thresholds. What result scales, what buys a second pilot with changes, what stops. Written down before you pick the cohort, because criteria chosen afterward are criteria chosen to make the pilot look successful.
3. Who decides. One named person with the authority to say no. A pilot without a named decision owner becomes a pilot that gets discussed quarterly.
4. What happens to your documents. You are handing a supplier real contracts, reports, and customer correspondence. Where it is stored, who sees it, what it may train, and when it is deleted belong in the agreement, not in a reassuring sentence on a call. The full set of terms worth arguing over is laid out in our review of the contract terms that decide whether training lands.
5. The write-up. A short, honest document: what each participant could do afterward that they could not before, and who did not move. Specify it in advance, or you get a completion rate and a satisfaction score. As Devlin Peck’s pilot kit puts it, pilots reporting only completion and pass rates earn polite nods and no budget.
The measurement design behind those thresholds — how many observations you need, how to discount the novelty window, how to get a comparison group almost for free — is set out in our rules for evaluating a pilot before you scale it.
What the Vendor’s Answer Tells You
The request is itself a test, and most of what you need arrives before anyone books a room.
| Their answer | What it tells you |
|---|---|
| “Happy to. What will you measure?” | They have done this before and expect to be judged. The strongest signal available. |
| “We’ll do one free.” | Sales budget, best closer, generic material. Decline and offer to pay. |
| “We only deliver company-wide.” | Either a program-shaped method or an unwillingness to be tested on one session. Ask which. |
| “A pilot is priced at our standard per-head rate.” | Defensible, but fix the rollout rate at the same time. |
| “We’ll use our standard curriculum.” | The pilot will not predict anything. Either your documents enter the room or it is theater. |
| “Anything derived from your documents remains ours.” | A procurement problem, not a detail. Resolve it before the session. |
A supplier who asks what you intend to measure, then argues with your definition, is the one to buy from. That conversation is the product.
When to Skip the Pilot
Three situations make piloting a waste. Under about twenty-five people, the pilot is the rollout; run the session, measure it properly, and accept that you have bought the program. Inside one homogeneous team, the cohort is a sample of one job, and the second session tells you nothing the first did not. Against a closing budget window, a six-to-eight-week pilot that ends after the money expires converts a decision into a delay.
SHRM’s 2026 survey of 223 L&D executives found that over one in three named skills training their top priority, and that the function’s hardest problems cluster around budgets, ROI, and deciding which skills matter. One pilot is a good answer to all three. Serial piloting is not. If this would be your third pilot of the year, the problem is not the vendor.
Frequently Asked Questions
What should a paid pilot workshop cost?
Roughly $2,000 to $7,000 for one private session with ten to fifteen people, built partly on your own material. The full market range for a facilitated session is about $2,000 to $15,000, and a pilot above the midpoint of that band is a program being sold as a test.
Should I ask for a free pilot instead?
No. A free session comes out of a sales budget, which means the best closer rather than the facilitator who would run your rollout, and material built for somebody else. Paying buys your own documents in the room, a named facilitator, and the standing to ask for changes.
How many people should be in the pilot?
Ten to fifteen. Below ten, one absence distorts the result. Above fifteen, you are running a small rollout, the facilitator cannot give individual attention, and you learn less per dollar.
How long before I decide?
Six to eight weeks from session to decision. The session is one day; the rest is a measurement window, so you read changed behavior rather than enthusiasm. Deciding the week after is deciding on the mood in the room.
What if the vendor refuses to run a pilot?
Find out whether the method genuinely depends on scale or whether they simply will not be judged on one session. Some sequenced programs do lose their logic when cut to a single workshop. A supplier who cannot explain which applies has answered the question.
Who should be in the cohort?
A cross-section, not your volunteers. A couple of daily users, a majority of ordinary ones, and two or three who are unconvinced. Build it to resemble the company you intend to roll out to, including the parts you would rather not.
What do I measure?
One task the cohort already repeats, timed before and after, plus a short note from each participant on what they can now do that they could not before. Usage dashboards and satisfaction scores measure attendance and mood, neither of which survives a conversation with finance.
What if the result is inconclusive?
Report it as inconclusive and say what would make it decisive. An honest inconclusive pilot with a named next test beats a flattering one, and it is what your three pre-written thresholds exist to make reportable.
Can I pilot two vendors at once?
Yes, and it is the strongest version of this exercise if you can afford two fees. Hold cohort size, session length, and source documents constant, run them two or three weeks apart, and compare the write-ups rather than the rooms. There is a reason to want the second one: after a single supplier has run your pilot, you have absorbed their structure and their vocabulary, and you are not really going back out to tender. Price that in.
Key Takeaways
- A pilot workshop tests the vendor, not the technology. The assistants work; the facilitator, the material, and the week afterward are the uncertainties.
- Pay for it. A free session is a demo built for somebody else, delivered by whoever closes best.
- Expect $2,000 to $7,000 for a private session of ten to fifteen people — near or under ATD’s $165 average cost per learning hour used.
- Fix the company-wide rate per head before the pilot runs. That clause stops a good result from becoming an expensive one.
- Build the cohort to look like the company, skeptics included, and write the three thresholds before you pick the people.
Where to Start
Write the three thresholds first, before you contact anyone: what result scales, what buys a second attempt, what stops. Then ask two suppliers for one paid session with twelve of your people, your documents, a named facilitator, and a fixed company-wide rate. How they answer narrows the field faster than any proposal.
If you want a second opinion on those thresholds before you send that email, book a free AI-readiness call. We run these sessions, so the conversation tends to be specific about what a pilot can and cannot settle — including the cases where you should not buy one.
Arthur Wandzel
