For a non-technical team, the format that works is a three-hour facilitated session on the team’s own documents, one function per room, capped at about 15 people per facilitator, with a 90-minute clinic three weeks later. Two sittings, five or six contact hours, one job learned properly.
Almost nothing in the corporate AI workshop catalog is shaped like that. The catalog sells seats: a day of live instruction per person, a self-paced module library, a webinar with recordings. Easy to price, easy to buy, and the reason so many AI enablement budgets buy a completion certificate and no change in how Tuesday goes.
What makes a non-technical group fluent is not simpler content. It is their own work on screen while someone who knows the tools watches them do it.
The formats, ranked by what transfers per hour
Ranked by what survives to the following week, not by content covered.
| Rank | Format | What it reliably teaches | Price anchor |
|---|---|---|---|
| 1 | Private facilitated session, 3 to 3.5 hours, one function, their documents | One workflow people repeat alone | $2,000 to $15,000 market range |
| 2 | Clinic or office hours, 60 to 90 minutes, same group, 2 to 3 weeks later | Fixes to what broke in use | In the engagement fee |
| 3 | Short live sessions spread over one to two weeks | A small set of workflows, spaced | $195 list for 1.5 hours of live instruction |
| 4 | Internal champions trained to run the above | Repeatable delivery | Quoted per program |
| 5 | Live catalog course, one seat at a time | Generic fluency, to be translated later | $350 per person per day |
| 6 | Self-paced module library | Vocabulary, and a completion record | $49 per month, or $150 to $375 per bundle |
| 7 | All-hands demo or lunch-and-learn | Awareness that the tools exist | Free |
The gap between rank 1 and rank 5 is not teaching quality. Catalog courses are often taught well. It is translation work: a seat-based course runs on sample material, so each attendee carries the lesson back to their own documents alone, in a week when nobody has a spare hour.
Rank 7 is on the list because it keeps getting mistaken for training. Two hundred people watching three impressive examples is internal marketing: nobody practices, nobody is corrected, and the handful of confident users ask all the questions.
Why “non-technical” is the wrong brief
Send a vendor the brief “we need an AI workshop for non-technical staff” and most read it as a request to simplify: fewer acronyms, more analogies, a slide explaining what a language model is. None of that is the constraint.
The real constraint is that these people have no sandbox and no slack. A developer can try a tool on a side branch at 9pm. A benefits administrator has 40 open cases and a system of record she is not allowed to be wrong in. Her question is whether this letter is safe to paste into a chat window and how to spot a wrong answer.
So replace the audience label with a task. “Our six-person finance team spends Thursday afternoons turning ledger exports into commentary for the leadership meeting, and we want it done by lunch” is a brief a facilitator can design three hours around. “Non-technical” is not.
There is a return argument for starting here rather than with your technical teams. Brynjolfsson, Li and Raymond tracked 5,172 support agents at one software firm after an AI assistant arrived and found a 15% average rise in issues resolved per hour. The gains concentrated among less experienced and lower-skilled agents, while the most experienced saw small speed gains and small quality declines. The people with the least tooling background had the most to gain, which argues for sending the first session to operations, finance, HR and support rather than the engineers, who need something different anyway.
Two findings that should decide your format
Vendor pages argue formats with adjectives. Two published results settle it.
Practice beats presentation, by a wide margin. Freeman and colleagues synthesized 225 studies of undergraduate science, engineering and math courses in PNAS in 2014 and found examination performance up 0.47 standard deviations under active learning, with lecturing raising failure rates by 55%. College classrooms are not conference rooms, but that direction, across 225 studies, is the most replicated finding in instruction research, and it rules out any format whose main activity is watching.
Hours matter, but only with a person attached. BCG’s AI at Work 2025 survey covered more than 10,600 leaders, managers and frontline employees in 11 countries. Regular AI use ran sharply higher among people with at least five hours of training and access to in-person sessions and coaching, and only a third said they had been properly trained. A year later, AI at Work 2026 put close to 12,000 workers in more than a dozen markets: 36% felt adequately upskilled, and 66% got limited or no guidance on what to do with the time AI saved them.
Together they give the shape of the thing to buy: five or six contact hours, split across two sittings, mostly spent doing rather than watching, with a human in the room and a named use for the hours the team gets back. Anything failing one of those tests is a seat, not a session.
What the catalog stocks, and what it does not
Published prices and formats from the providers a buyer meets first, checked this month.
| Provider | Published format | Published price |
|---|---|---|
| Certstaffix | Half-day business-professional overview, four one-day live online titles, plus self-paced bundles of 8 to 17 modules | $350 per person per title; $150 to $375 per bundle |
| Section | 1.5 hours of live instruction and Q&A, plus recordings | $195 list, or membership at $62.50 per month billed annually |
| Google AI Essentials on Coursera | Five self-paced courses, under 10 hours total, beginner level | $49 per month in the US after a 7-day trial |
Every one of those is a per-seat product. Private group rates exist but are quoted on a call, so the comparison a buyer most needs is the one the websites cannot give. Market rates for a private facilitated session run about $2,000 to $15,000 depending on how many roles need their own material, and our breakdown of the half-day and full-day units walks the line items.
Notice what nobody stocks: the second appointment.
The three hours, in order
We run these sessions, so this part is observation rather than theory. The shape that works for a non-technical group is narrower than most agendas allow.
Before the day. Ask for three or four real documents, redacted if needed: a client email thread, an intake form, a messy export. A session built on those teaches a workflow; one built on generic samples teaches a demo.
First hour. One task, demonstrated once on the team’s own document, then repeated by everyone on their own version while the facilitator walks the room. No tool tour, no history of the field. People who have never typed a prompt produce something usable inside 20 minutes when the task is theirs.
Second hour. Verification and data rules, taught inside the task rather than as a policy slide: what to check before sending, which fields never get pasted into a chat window, what a model does when it does not know. Abstractly, this is compliance theater. Taught on the letter they just drafted, it sticks.
Third hour. Each person saves two prompts somewhere shared and names one real task to do this way before Friday. The named task is the deliverable, not the slides.
Two mechanics decide whether that happens. Cap the room at 15 per facilitator, because the product is individual correction and past that nobody gets any. Keep practice at twice the talking time. Physical or remote moves the ceiling on both, which we take apart in the on-site and remote comparison.
The clinic is where training becomes capability
Book the second appointment before the first one runs: 90 minutes, same group, two to three weeks out, diarized while the budget is open. It never gets booked later.
The useful questions do not exist on day one. They appear in week two: the prompt that worked twice and failed the third time, the summary that invented a number, the task that needed two tools rather than one. A longer first session cannot answer those; it can only cover more untested ground.
The clinic is also where that BCG finding about saved time gets handled. If the administrator now finishes her Thursday reporting by 11am and nobody has said what the extra two hours are for, the hours go back into the inbox and the gain vanishes from every measure you have. Naming the use of recovered time is a 10-minute agenda item, and it is the difference between training and AI enablement that shows up in the numbers.
Three formats that fail non-technical teams
The self-paced license. A module library bought for a team already behind on its day job produces completion rates, not capability. Good reference layer under a live session, poor substitute for it.
The prompt-library handout. A deck of 50 prompts transfers nothing: a prompt written for a task you do not have is just text. The two a person wrote themselves, on their own document, are the ones still in use next quarter.
The mixed room taught as a workflow class. Seven functions in one session means every example is wrong for six of them. Cross-functional rooms work when the output is a shared decision, such as agreed rules for safe use, and the design mechanics behind both choices are a separate craft. For teaching a workflow, one function per room.
Pick the first session by function
One task per team, chosen for frequency and tedium rather than ambition.
| Function | First task worth three hours | What good looks like after |
|---|---|---|
| Finance | Turning a ledger export into leadership commentary | A draft in 30 minutes, checked against the export |
| HR and people ops | Drafting policy answers and scorecards from source documents | Consistent answers, with the source quoted |
| Customer support | Rewriting and summarizing long case threads | Shorter handoffs, fewer repeat questions |
| Marketing | Turning a research call recording into briefs | A week of drafts from one call |
| Operations and admin | Structuring messy intake forms | Clean fields, exception list, nothing retyped |
| Sales | Pre-call research and follow-up notes | Notes written before the next meeting |
| Leadership | Stress-testing their own memo and board pack | Sharper questions, not a longer memo |
Leadership groups are the one case where shorter and more frequent beats everything, because their scarce resource is calendar, not money. Keeping that room engaged without talking down to it is its own problem, covered in our piece on executive sessions.
Five questions that expose a presentation
- How many net contact hours, and what share is hands-on? Seven-hour days often hold five hours of teaching. Under two-thirds practice and you have bought a presentation at workshop prices.
- Will you build the exercises from our documents? A vendor who never asks for your material has one curriculum for everyone.
- How many people per facilitator? The honest answer names a cap. A vendor happy to take 60 in a room is selling a webinar.
- What does the follow-up include and cost? This tells you whether you are buying attendance or capability.
- Who on your side has done this work, not just taught it? Ask what they automated last month. Practitioners answer in specifics.
Frequently asked questions
What is the best AI workshop format for non-technical staff?
A private facilitated session of about three hours, one function per room, built on the team’s own documents, with a clinic two to three weeks later. That shape gives each person enough corrected practice on one real task to repeat it alone, the only outcome that shows up next week. Managers should attend and do the exercises rather than observe.
How long should an AI workshop be for non-technical employees?
Five to six contact hours, split across two sittings rather than one long day. BCG’s 2025 survey found regular use sharply higher above roughly five hours when paired with in-person teaching and coaching, and splitting the hours lets people try the work alone in between.
How much does an AI workshop for non-technical teams cost?
Private group sessions run about $2,000 to $15,000 at market rates, depending on how many roles need their own material and whether follow-up is included. Per-seat options are cheaper and published: $350 a head for a one-day live class, $195 for 90 minutes of live instruction, $49 a month self-paced.
Can non-technical staff learn AI from a self-paced course?
A few self-starters will. Most will not finish, and finishing is not the goal. Self-paced material works as a reference layer after hands-on AI training on their own documents.
How many people should be in the room?
Cap it at 15 per facilitator, and 12 is better, which is tighter than the 25 a general session can carry. The limit is correction time rather than seating: once the facilitator cannot reach everyone during practice, the session turns into a lecture with laptops open.
Do we need separate sessions for each department?
For teaching a workflow, yes. One function per room keeps every example relevant to everyone present. Mixed rooms suit a shared decision, such as agreed rules for what data may go into which tool.
What should a non-technical team be able to do after one session?
Repeat one workflow end to end on real work, with two saved prompts tested on their own documents, a short list of checks before anything goes out, and one named task to do that way this week.
How do we measure whether it worked?
Measure the task you trained, before and after: hours to produce the Thursday report, cases closed per day, drafts sent back for rework. Completion rates and license dashboards say nothing about capability, which is why so many teams have strong usage numbers and flat output.
Key takeaways
- Buy a three-hour facilitated session on the team’s own documents, one function per room, 15 people per facilitator, plus a clinic in week three. That shape beats every seat-based option on what survives.
- “Non-technical” is not a brief. Name the weekly task you want finished faster and let the facilitator design the hours around it.
- Two published results decide the format: 225 studies in PNAS put active learning 0.47 standard deviations ahead of lecturing, and BCG found regular use rises sharply when five hours or more come with in-person coaching.
- The catalog stocks seats, not sessions, and sells no follow-up at all. Book the clinic before the first session runs, or it never happens.
- Say out loud what the saved time is for. BCG found 66% of workers get limited or no guidance on that, and unassigned hours go back into the inbox.
Count the functions you need to move this quarter, pick the weekly task each of them hates most, and book three hours against it. A free AI-readiness call will tell you which task belongs in the first session and what follows it.
Dirk Jan van Veen, PhD
