A train-the-trainer arrangement for AI enablement is a handover specification. What you buy is a named list of artifacts, in editable formats, that your company owns and can change after the engagement ends. Judge any proposal on that list. If it describes stages, badges, and a community but never says what files land in whose hands on the last day, it has not told you whether the capability survives the vendor.
Every program selling this model offers the same five-stage diagram: train a first group, pick champions, certify them, roll out by department, keep support active. The diagram is the easy part. The questions a 200-person company has are narrower, and nobody answers them in public: who to pick, how many hours a week it costs them, what gets handed over, what goes stale and when, and at what headcount the model stops being cheaper than paying someone to show up.
The handover is an artifact list
Write the list into the contract before the first session. Nine items, each with a format and an owner.
| Artifact | Format that makes it usable | Owner after handover |
|---|---|---|
| Run of show | Editable doc, minute markers | Facilitator |
| Slide deck | Native file, not a PDF export | Facilitator |
| Facilitator notes | Per slide, including what to do when an exercise stalls | Facilitator |
| Exercise set | Each one a task, an input document, and the artifact it produces | Facilitator |
| Prompt library | Versioned shared doc, 10–20 entries | The team, not one person |
| Pre-work and access checklist | Owners and lead times | Coordinator |
| Participant handout | One page, printable | Coordinator |
| Measurement instrument | Before-and-after questions plus scoring | Whoever reports to the sponsor |
| Change rights, in writing | A clause letting staff edit, extract, and re-record the above | Legal or procurement |
The last row gets skipped most and matters most. Several published programs hand over “all the materials necessary to facilitate these sessions” and leave license terms unstated. A deck you may present but not edit is a subscription with extra steps. Ask for edit rights in the same sentence as the deliverable.
Two other rows deserve attention. Facilitator notes are the only artifact that carries judgment, which is why most handovers omit them and most second-run sessions feel thin. And the exercise set has to be tasks against real inputs rather than a topic list, because the exercise is where behavior moves or does not. Our exercise library covers what belongs in it.
Who to pick, and why the obvious choice is often wrong
The strongest candidate is rarely the most fluent person in the building. Pick for three traits, in this order:
- Calendar authority. Someone who can hold a recurring 90-minute block without asking permission. This rules out more good candidates than any skills gap.
- Credibility with the skeptics, not the enthusiasts. A facilitator the quiet half of the department already trusts can run a mediocre session well. The reverse is not true.
- Working fluency, a floor rather than a ranking. They need to have finished real work with an assistant, not to be the best at it.
There is a measurement trap here that the research names and the vendor pages do not. In the strongest available review of train-the-trainer outcomes, Nexø and colleagues note that “trainers are often selected to be trainers because they already have a high level of experience or knowledge,” which is why trainers show small gains on knowledge tests (BMC Medical Education, January 2024). If your new facilitators score well on day one, that tells you about your selection, not your program. Measure the departments they teach.
Your facilitator also needs the department’s own work in hand rather than a generic case study, which is the argument for training on last week’s real work. A cascade multiplies whatever it carries.
The hours nobody puts in the budget
A published program anchors the real cost. One US provider sells an internal AI champion track at $8,500–$12,000 for four to six weeks and states the participant’s commitment plainly: three to five hours a week with a strategist plus four to six hours of applied work (ITECS, checked October 2026). That is seven to eleven hours a week, so 28 to 66 hours of a trusted senior employee before they have taught anyone anything.
Then delivery starts. Budget three to four hours of prep per 90 minutes in the room for the first run, dropping to about an hour by the third repeat, plus a couple of hours per department to swap in that team’s own documents. A facilitator covering six departments lands near 35 to 40 hours on the first pass.
The honest year-one total for one internal facilitator is 65 to 105 hours plus the vendor fee. At a loaded $80 an hour, that time runs $5,200–$8,400, comparable to the invoice. Neither number is large. Both need approval from the person who owns that employee’s quarter, and the program dies in month three when they were not asked.
So put the reduced load in writing, to their manager. And pick two facilitators, even in a 120-person company: a single point of failure with a promotion ahead of them is the most common way this ends.
What the evidence says about cascade training
State the evidence as it reads. The 2024 review covering MEDLINE, Embase, CINAHL and ERIC screened 3,800 records down to 11 studies and concluded that “train-the-trainer models can successfully disseminate knowledge,” while also finding “there is not yet sufficient evidence to conclude whether TTT-programs are more effective compared to other programs.”
Read both halves. The cascade works as a way to reach people. It has not been shown to teach better than a vendor teaching directly, so the case for it is reach, cost per head at scale, and local relevance. Anyone selling it as better teaching is ahead of the literature.
The same review carries a sharper finding no commercial page mentions: none of the 11 studies examined how the cascaded sessions were taught, and the authors note an absence of transparency about how knowledge was assumed to move from trainer to trainee. “Certified facilitator” describes a completed process, not what gets delivered in month four. You have to measure that yourself.
The dose evidence for AI training is clearer. BCG’s survey of more than 10,600 workers across 11 countries found that only 36% of employees feel adequately trained in AI use, and that people receiving five or more hours of training are significantly more likely to become regular users, “especially in person and with coaching” (BCG, June 2025). Five hours is the bar a cascade has to clear per person, and a 60-minute all-hands does not clear it however many people it reaches. The wider case for treating this as capability rather than access sits in our AI enablement manifesto.
Three things that go stale, on three different clocks
Every program here is sold as a one-time build. Three parts of it decay, and each needs an owner and a date.
- Screenshots and click paths: weeks to months. Assistant interfaces move constantly, and a deck showing menus that no longer exist costs a facilitator credibility in the first ten minutes. Keep them in a separate appendix so one person can refresh them without touching the deck.
- Prompt library: quarterly. Prompts written against last year’s defaults often need less scaffolding now, and some are no longer needed. Review once a quarter and delete more than you add.
- Exercises and source documents: annually, or whenever the work changes. When a team’s reporting cycle or template changes, the exercise built on the old one teaches the wrong thing.
One calendar entry per clock, one name per entry. A refresh that belongs to “the AI champions” belongs to nobody.
A 90-day sequence from vendor-led to internally led
The sequence matters more than the stage names. Transfer happens when the internal facilitator teaches in front of the person who taught them, not when a certificate is issued.
| Phase | What happens | Leads | Exit test |
|---|---|---|---|
| Weeks 1–2 | Select two facilitators, confirm released hours, collect real source documents | Sponsor | Reduced load in writing for both |
| Weeks 3–4 | Facilitators attend as participants, finishing their own real work | Vendor | Each produced an artifact they would send a colleague |
| Week 5 | Walkthrough of the run of show, hard parts rehearsed rather than reviewed | Vendor | They can say what to do when an exercise stalls |
| Week 6 | Co-taught session; the internal facilitator runs two blocks | Shared | Participant artifacts match the first cohort’s |
| Weeks 7–9 | Internal facilitator teaches a full session; the vendor observes against a fidelity checklist | Internal | Checklist passed, sponsor sees the output |
| Weeks 10–12 | Second internal session, unobserved; handover signed off against the artifact list | Internal | Every artifact in internal hands, editable |
| Quarterly after | Refresh cycle plus a monthly office-hours slot | Internal | Prompt library changed since last quarter |
The fidelity checklist in weeks 7–9 is the piece the literature says is missing, and it is cheap to build: eight to twelve observable items drawn from the run of show. Did every participant open an assistant in the first ten minutes? Did each leave with a finished artifact? Was the exercise built on their own document? Score it, keep it, run it again in month six.
Keep cohorts small once the internal facilitator takes over. An AI workshop for teams runs on facilitator attention, a first-timer has less slack than a professional, and the arithmetic is unforgiving above a dozen people, as our guide to cohort size works through. The design question itself sits in our workshop design playbook.
When a cascade pays for itself
Do the arithmetic before buying the model. Hands-on AI training for a team runs roughly $2,000–$15,000 in the US market depending on length, tailoring, and travel. Call it $3,500 for a half-day with a dozen people.
A 200-person company needs about 17 cohorts of 12 for one pass, so buying delivery outright runs near $60,000. Building internal capability costs a vendor fee of $8,500–$12,000 plus 65 to 105 internal hours, and next year’s pass costs almost nothing beyond those hours. The cascade wins clearly.
Now run it at 60 people. Five cohorts, about $17,500 bought outright, against a vendor fee plus two people losing a quarter of their week for six weeks and then owning a refresh cycle forever. That is break-even in cash and worse in attention. Below about 100 people, paying for delivery is usually the better purchase, and a vendor should say so.
Three conditions make a cascade pay:
- More than one pass. Onboarding, turnover, or a second wave of departments. A one-time rollout amortizes nothing.
- Enough headcount that cohort count is the constraint, not cohort quality. Roughly 100 people and up.
- Local material that differs by department. If finance, operations, and client service work on different documents, an internal facilitator adapting examples beats an outsider who cannot.
If two of the three are false, keep buying delivery. If you are unsure which side of the line you sit on, a free AI readiness call answers it faster than a proposal will.
What published programs hand over
The programs currently selling under this heading, checked in October 2026:
| Program | Length | Published price | Named deliverables | Gap |
|---|---|---|---|---|
| ITECS AI Champion | 4–6 weeks | $8,500–$12,000 | Curriculum doc, 10–15 prompts, deck, operations playbook, 90-day roadmap | Edit rights not stated |
| Ajaia Train-the-Trainer | Not stated | Not published | Facilitation training, session guides, prompt libraries, templates | No duration, no certification criteria |
| AI for Education (schools) | 1–2 days | Not published | Materials for two to four named sessions, prompt library, video library | License terms unstated |
| GitHub advocate playbook | 30/90-day phases | Free | Facilitation guides, community hub, check-ins | Volunteers only, no measured outcomes |
One program publishes a price. None publishes change rights. And the free playbook is the most honest about time, suggesting 30 to 60 minutes a week and insisting the role must not become a second job. That is the right corrective to the build weeks above: the build is heavy, the steady state has to be light.
Four ways the handover fails
- The facilitator gets promoted. The most common ending. Two facilitators from the start, and the artifact list owned by the team rather than a person.
- Nobody released the hours. The sponsor approved the program; the facilitator’s manager approved nothing.
- The deck was handed over, the notes were not. Session two is a slide read-through, attendance drops, and the model gets blamed for a handover failure.
- Fidelity drifts with nothing watching. By month six the session has become a demo. A twelve-item checklist twice a year catches it for the cost of one afternoon.
Frequently asked questions
What is a train-the-trainer AI program?
An arrangement in which an outside facilitator teaches a small internal group to run the corporate AI workshop themselves, then hands over the materials so they can keep running it unaided. The deliverable is the material plus the rights to change it, not a credential.
How long does it take before internal facilitators can teach alone?
About 10 to 12 weeks: two weeks of selection and scheduling, two attending as participants, one rehearsing, one co-teaching, then two solo sessions. Published programs advertise four to six weeks, which covers the build and not the supervised deliveries.
How many internal facilitators does a company need?
Two at minimum, whatever the size, because one is a single point of failure. Above roughly 300 people, one per major function beats a central pair, since the value comes from local examples.
What should be in the handover package?
Nine artifacts: run of show, deck in a native editable format, per-slide facilitator notes, the exercise set with its input documents, a versioned prompt library, the access checklist, a participant handout, the measurement instrument, and a written clause letting staff edit all of it.
Can internal trainers change the content?
Only if the contract says so. Several vendors hand over materials with no stated license terms, which leaves editing in a gray area. Ask for change rights explicitly, and for native files rather than PDF exports, since the second quietly prevents the first.
Is cascade training as effective as having the vendor teach everyone?
The evidence says it spreads knowledge but has not been shown to teach better than direct delivery. Choose it for reach, cost at scale, and local relevance.
How do we tell whether the cascade is still working six months later?
Run a fidelity checklist of eight to twelve observable items against a current session, and measure the taught department rather than the facilitator. Facilitators are picked for existing skill, so their scores move least and tell you least.
What does a train-the-trainer program cost?
Published US programs for one internal facilitator run roughly $8,500–$12,000 over four to six weeks, inside the broader $2,000–$15,000 range for hands-on team sessions. Add 65 to 105 internal hours in year one, often the larger half of the real cost.
At what company size does this stop making sense?
Below about 100 people, or whenever only one pass is planned. Five cohorts bought outright costs about what building internal capability does and takes far less attention.
Who makes a good internal AI facilitator?
Someone with authority over their own calendar, credibility with the team’s skeptics, and enough working fluency to have finished real work with an assistant. Calendar authority rules out more strong candidates than any skills gap.
Key takeaways
- Buy the handover, not the badge. Nine named artifacts in editable formats, with written change rights, decide whether the capability outlasts the engagement.
- Pick facilitators for calendar authority and credibility with skeptics, in that order. Working fluency is a floor, not a ranking.
- Budget 65 to 105 internal hours per facilitator in year one, and get the released time approved by their manager, not only by the sponsor.
- The peer-reviewed evidence supports the cascade for reach, not for better teaching, and no study in the strongest review examined how cascaded sessions were taught. Build your own fidelity checklist.
- Screenshots decay in weeks, prompt libraries in a quarter, exercises in a year. One name and one date against each.
- Below roughly 100 people, or for a single pass, paying for delivery is usually cheaper. If the arithmetic is close, talk it through on a free AI readiness call before signing a build.
Arthur Wandzel
