Most AI workshops are designed forward. Someone picks topics, fills a day, books a facilitator, and hopes behavior follows. It rarely does. A session changes what happens on Monday only when it was designed from Monday backward: name the thing each person owes somebody next week, then build the shortest path to them making that thing with AI, checked and usable, before they leave the room. That inversion is the whole method, and almost no agenda in print follows it. What follows is how to design an AI workshop for teams on that principle, from the pre-work that sets the ceiling through the run of show, the facilitation choices that decide the outcome, what to cut, what the market charges, and how to tell within three weeks whether your AI enablement spend moved anything.
Why most sessions produce a good day and no change
The failure is not a lack of interest. It is a design that never required anyone to do their own work.
PwC’s 2025 Global Workforce Hopes and Fears Survey, run across roughly 50,000 workers in 48 economies, found 54% had used AI for their role in the past year while only 14% used generative AI daily. The same survey found daily users far more likely to report productivity gains than infrequent users, 92% against 58%. The gap between trying a tool and using it every day is where training money either works or evaporates.
BCG’s AI at Work 2025, published June 26, 2025 from a survey of more than 10,600 leaders, managers, and frontline employees across 11 countries and regions, puts a sharper point on it. Only one-third of employees say they have been properly trained. Regular use among frontline employees has stalled at 51%. And the finding most relevant to session design: regular usage is sharply higher for employees who receive at least five hours of training and have access to in-person training and coaching.
Five hours with coaching. Not a 60-minute lunch-and-learn, and not a seat in a self-paced catalog course. BCG is reporting an association rather than a proven cause, but as a floor to design against it already rules out most of what gets booked.
Three patterns account for nearly every session that lands flat.
The tool tour. A facilitator drives while the room watches features. People leave able to describe a product and unable to use it. Watching someone else be fluent produces the warm feeling of competence and none of the competence.
The generic exercise. The room writes a prompt for a made-up company’s made-up newsletter. The skill never attaches to a real task, so there is nothing to carry back. Practitioner accounts of what separates sustained users from one-time triers land on the same factor repeatedly: the practice has to be on real work, in the real tool, with the real constraints.
No receiving party. Nothing produced in the room was owed to anyone. A draft with no reader is an exercise. A draft your director expects on Tuesday is a commitment, and commitments survive the week.
McKinsey’s state-of-AI research has found roughly a third of organizations report scaling AI across the enterprise, with the rest stuck earlier. Enterprise-wide scale is not what a 50-to-500-person company is buying; what it is buying is the end of pilot limbo on one team. A workshop can do that, but only if it is built as the first increment of a program rather than as an event. The broader case for treating capability as a build rather than a purchase is laid out in why buying licenses is not the same as building, and the surrounding program structure in the move from license rollout to measurable team capability.
Design backward from the Monday artifact
Start with one question: what does each person in this room owe somebody next week?
Not a topic. Not a skill. An artifact with a name, a format, and a reader. A renewal proposal. A board update. A QBR deck. A candidate scorecard. A vendor security questionnaire. A monthly close commentary. A support macro set. A sprint summary for the CFO.
For each artifact, write the Monday test in one sentence: by Monday, this person can produce a usable first draft of this artifact with AI, verified, in under X minutes. Then design the session as the shortest path to that sentence being true. Cut everything that does not shorten that path, including material that is interesting, impressive, or free.
This sounds obvious. It is also the step almost every agenda in print skips, which is why the agendas look alike and the outcomes do not improve. A five-phase framework tells you the shape of a day. It does not tell you what belongs in it. The artifact does.
Two rules make the method hold up in practice.
One artifact per person, three per room, maximum. Below one, nobody has a stake. Above three, the facilitator cannot hold the room’s context and the comparison round collapses into parallel monologues. A room of 12 with three artifact types gives you four people per artifact, which is exactly enough to compare outputs usefully.
The artifact must be boring and frequent. The temptation is to pick the hardest, showiest document in the department. Resist it. Pick the one produced weekly by four people who all dislike it. Frequency is what converts one guided success into a habit; a quarterly artifact gives the skill nowhere to land before it fades.
A useful side effect: once the artifacts are named, the curriculum writes itself and the shopping list shrinks. You now know which documents must be cleared for use, which account tier people need, and which failure modes you have to teach. Each function needs its own artifacts, and for a company running several at once, one company, nine different curricula covers how to split the work by role without running nine separate programs.
The pre-work pack that sets the ceiling
Every vendor page says “pre-work.” Almost none say what is in it. The pack has four items, all of which must exist before the session is booked, not before it starts.
1. The task audit. Each attendee lists their five most frequent writing or analysis tasks and an honest estimate of how long each takes today. Fifteen minutes of their time. The audit does three jobs: it produces the artifact shortlist, it gives you a baseline to measure against later, and it quietly tells each person that the day is about their work rather than about AI.
2. The document pack. Five real documents per artifact type, cleared for use in the tool. Real means a finished example of the artifact (so the room knows what good looks like), plus the raw inputs it is normally built from. Cleared means someone with authority has signed off that these files can go in. Chasing clearance on the morning of the session is the single most common reason a room spends its first hour idle.
3. Provisioned accounts, tested. Business-tier accounts, logged into once by each person before the day, in the browser they will actually use. Free personal accounts carry weaker data protections and other feature sets, so the business tier matters for more than compliance. Test the login. A room of 12 people hitting SSO for the first time at 9:05 costs 25 minutes you cannot get back. If the ground rules for what may be pasted where are not settled yet, settle them first; the gap between AI policy and training covers how to write rules people can follow in the moment.
4. A sponsor-stated outcome, in writing, in the sponsor’s words. One or two sentences from the executive paying for the day, naming what should change after the day. This gets read aloud at the open. It is the cheapest adoption lever available, and BCG’s finding that regular use is driven by leaders and managers while frontline use stalls says plainly why the signal matters.
If any of the four is missing, move the date. A session run on an incomplete pack teaches a watered-down version of the day you paid for, and nobody in the room will know what they missed.
Who belongs in the room
Cohort design decides more about the outcome than content does, and three calls do most of the work.
Group by artifact, not by department or seniority. Four people who all write renewal proposals belong together even if they sit in sales, customer success, and finance. They share inputs, format conventions, and failure modes, so every comparison teaches all four at once. A department is a billing unit, not a cohort.
Cap the room at 12 to 16 with one facilitator. The limit comes from the comparison round, which only works if the facilitator can read several outputs closely and name the differences out loud. Beyond 16, that round becomes a show of hands. If you need to train 40 people, run three sessions, not one large one. Rooms of 30 are a budget decision presented as a scheduling decision.
Decide the manager question with care. Managers in the room with their direct reports suppress the experimentation that makes the day work, because nobody wants to produce a bad first draft in front of the person who writes their review. The usual fix: run a separate short session for the manager layer first, covering the same artifacts, so they can sponsor credibly without watching the practice. The one case against a split is a small, truly peer-like team, where the split does more harm than the rank gap does.
One more note on who to include. The person most worth converting is rarely the volunteer. It is the busy, respected, mid-tenure producer whose opinion the rest of the team actually checks. Get that person a visible win in the room and the follow-through takes care of itself. Get the office enthusiast a win and you have confirmed what everyone already assumed.
The run of show for a six-hour session
Six working hours, lunch excluded, built so the room is making things for more than half of it. The split that matters is simple: at least half the clock on their own work in their own tool. Anything below that is a presentation with exercises attached.
| Time | Block | What the room is doing |
|---|---|---|
| 0:00–0:20 | Open | Sponsor’s outcome read aloud; the day’s three artifacts named; the two ground rules |
| 0:20–1:00 | Cold attempt | Everyone attempts their own artifact with AI, unaided |
| 1:00–1:50 | Comparison round | Reading best and worst outputs side by side; naming the difference |
| 1:50–2:00 | Break | |
| 2:00–3:15 | Build round | Writing the reusable prompt for each artifact, as a group, then re-running it |
| 3:15–4:00 | Lunch | |
| 4:00–4:45 | Verification drill | Checking a deliberately flawed output against source documents |
| 4:45–5:30 | Second artifact | Same loop, compressed, on artifact two |
| 5:30–6:00 | Commit round | Each person writes their next-week plan; prompt library handed over |
The sequence matters more than the timings, which flex by group. What cannot flex: the cold attempt comes before any teaching, and the comparison round comes before any prompt is written for the room.
The cold attempt
Forty minutes, no guidance, their own artifact, from whatever inputs they brought. The brief is thin on purpose: “produce a usable first draft, however you can.”
Facilitators hate this block. It looks like wasted time, several people flounder, and the output quality is mixed to poor. That is the point. The cold attempt does two things no lecture can. It creates the felt problem that makes the next hour land, and it generates the raw material the session is actually built on: a spread of real outputs on the same real task, from people with the same job.
Walk the room during it, say almost nothing, and take notes on who produced what.
The comparison round
This is the part competitors do not publish, and it is where the teaching happens.
Take four outputs from the same artifact type, chosen before the break: the best, the weakest, and two in between. Put them on the screen and read them. Then ask the room the only question that matters: what did the person with the best draft do differently?
Nine times in ten the answer has nothing to do with clever phrasing. It is context. The strong draft came from someone who pasted in the finished example, the source data, and a line about who the reader is. The weak draft came from someone who typed a request and pressed enter. Watching that difference surface from their own colleagues’ work, in 20 minutes, teaches more than a prompting module does in a full day, because the room derives the lesson instead of receiving it.
Two facilitation notes. Ask the author’s permission before showing a weak output, every time, and frame it as the room’s problem rather than the person’s. And resist giving the answer; if you name the context gap yourself, you have converted a discovery into a claim.
The build round
Now the room writes the reusable prompt, together, per artifact. Not a clever one-liner. A working block with four parts: who the AI is acting as, what the finished artifact looks like (format, length, sections, tone), what source material is attached, and what to flag as uncertain rather than guess.
Build it on screen, in one shared document, with the room arguing about the wording. The argument is the value. By the end of the block, each artifact has a prompt that the four people who own that artifact helped write, which is a far better predictor of use than a prompt handed down from a consultant’s library.
Then everyone re-runs their cold attempt with the group prompt and their own inputs. Side by side with the 40-minute version from the morning, the improvement is usually dramatic and entirely visible, which does more for buy-in than any slide about adoption.
Save every prompt to one shared file before the room leaves. That file is the real deliverable of the day.
The verification drill
Forty-five minutes, and the block most agendas drop first. Dropping it is how a team gets fast at producing output nobody can trust.
Hand out an AI-generated version of a real artifact with four errors planted in it: an invented figure, a misattributed quote, a plausible-sounding claim with no basis in the source documents, and one real error the model made on its own. Give the room the source material. Ask them to find the problems and time it.
What people learn is where to look. Numbers get checked against the source, every time. Named references get opened. Confident specifics that appear nowhere in the inputs get deleted. After one timed pass, checking stops being a vague ask to “review the output” and becomes a short routine attached to a specific document type, which is the only form in which it survives contact with a busy week.
This block is also where the room’s quality skeptics convert. Give someone a method for catching the failure mode they were worried about and the objection resolves itself.
The commit round
Thirty minutes. Each person writes three lines and reads one aloud.
- The artifact I will produce this way next week, and the date.
- The one person who will see the result.
- The thing most likely to stop me.
Collect them. They are the follow-through list, and the third line tells you where the program will break before it breaks. Vague commitments produce vague outcomes, so push for a date and a name every time.
Facilitation choices that decide the outcome
The mechanics below look small on paper. In practice they separate a day that works from a day that was fine.
One facilitator per 12 to 16, plus a floater above 10. The floater handles logins, model slowness, and the person who is quietly stuck and will not raise a hand. Without one, the facilitator spends the build round doing tech support and the comparison round loses its spine.
Two screens if you can get them. One holds the shared prompt document, which stays visible all day. The other is whatever is being read right now. A single screen forces constant swapping and the prompt library disappears from view, which is the one artifact you want in everyone’s side vision all day.
Keep the senior skeptic busy and early. Give them the first cold attempt on the hardest artifact and ask for their verdict in the comparison round. A 20-year producer who publicly concedes that one output is close to usable converts more of the room than any evidence you could present. A 20-year producer left to watch from the back becomes the story of the day.
Plan for the tool misbehaving on stage. It will. The model will be slow, refuse a harmless request, or produce a confidently wrong paragraph in front of everyone. Narrate it. “This is the failure mode we are about to learn to catch” costs you nothing and buys the credibility that a flawless demo quietly spends.
Say the limits out loud. A single session does not produce capability; it produces one guided success and a plan. Sponsors who hear that on the day stop expecting a transformation by Friday and start resourcing the three weeks that follow, which is where the capability actually forms.
Protect the hands-on clock from yourself. The biggest pull in the room is toward talking, because talking is what a facilitator is comfortable doing. Every minute of explanation comes out of the build round. If the day runs long, cut content, never practice.
What an AI workshop for teams should refuse to teach
Cutting is the least popular part of workshop design and the highest-yield. Four things get cut from a fluency session, and cutting them raises what people retain.
What goes on inside the model. Tokens, context windows, temperature, and how transformers work. Fun to know, and no help in making a verified renewal proposal on Tuesday. Nobody needs to know how a transmission works to drive to work.
Prompt-engineering taxonomy. Zero-shot, few-shot, chain of thought, as named categories. The room needs one good reusable prompt block per artifact and the habit of adding context, which it gets from the build round without ever learning the vocabulary.
Agent and workflow building. Automations, custom assistants, and connected tools are another project with its own budget and a prerequisite: people who can already judge whether a model’s output on their own documents is any good. Build automation on top of fluency, not instead of it. Where that line sits for your company, and when it makes sense to hire the capability versus grow it, is the subject of buying versus building enablement capability.
Tool-stack strategy. Which products to standardize on is a procurement decision made by two or three people, not a training topic for 14. Train on the tool you already own, in the account tier people will actually have on Monday.
What stays: their artifacts, context discipline, iteration, verification, the shared prompt library, and the ground rules for what data goes where. That is a full day, and it is enough.
Formats and their honest tradeoffs
| Format | Clock | What it can honestly produce | Where it fails |
|---|---|---|---|
| Executive briefing | 90 min | Shared vocabulary, a sponsor who can make decisions | Produces no capability; sold as training, it burns the budget |
| Half day | 3–4 hrs | One artifact per person, done once, verified | Verification gets squeezed; second artifact never happens |
| Full day | 6 hrs | Two artifacts, a prompt library, a committed plan | Attention sags after lunch without a hard format change |
| Four weekly sessions | 4 × 90 min | Repetition across real work cycles; the best retention | Hardest to schedule; attendance decays without manager cover |
| Two days | 12 hrs | Several artifacts and a trained internal facilitator | Overkill for most 50-to-500-person teams; usually a vendor’s preference |
If you can only buy one thing, buy the full day and spend the saved money on the four weeks after it. If you can schedule the weekly series, do that instead: the gap between sessions is where real work happens, and the second session’s opening question (“what broke?”) is the most valuable 20 minutes in the whole format.
The executive briefing deserves a warning. It is truly useful as a sponsor-alignment move and truly useless as training, and vendors sell it as both. Buy it for what it is.
On-site or remote
Remote delivery works, with two conditions. Cap the room at eight, and require cameras during the comparison round. The comparison round depends on reading the room’s reaction to a weak output, and a wall of gray tiles removes the signal the facilitator steers by.
On-site earns its travel cost in three situations: the cohort includes senior skeptics, the team is newly formed or distrustful, or the artifacts involve material people are nervous about putting in a tool. All three are social problems, and social problems resolve faster in a room. BCG’s finding that in-person training and coaching coincides with sharply higher regular use is consistent with that, though the survey measures association rather than cause.
For everything else, remote is the better buy. The money saved on travel funds the follow-up sessions, which affect retention more than the delivery mode does.
What it costs and what the money buys
Market ranges for hands-on AI training delivered to a team run roughly $2,000 to $15,000, which is a wide band for a reason. Four variables move the number.
Customization depth. A session built on your documents, with your artifacts, costs more than a standard deck re-skinned with your logo. It is also the only version that passes the Monday test. Ask what changes between your session and the one the vendor ran last week; if the answer is thin, so is the design.
Facilitator count. A floater is a second day rate, and above 10 people it is worth it.
Cohort count. Three sessions of 14 cost roughly three times one session of 14. Rooms of 40 are cheaper per head and produce less per head, which makes the per-head math misleading.
Follow-through. Weekly check-ins, prompt-library curation, and a measurement readout for the sponsor are real work and get priced separately. If they are not in the quote, you have bought an event.
Self-paced catalog seats from Coursera, Udemy Business, LinkedIn Learning, Pluralsight, and Skillsoft are far cheaper per person and solve another problem. Verify current pricing and curriculum on each provider’s own pages before comparing, because corporate catalogs and plan tiers change quarterly. Catalog courses train people who are already keen. A facilitated session changes how a team produces a specific document. Companies that treat the two as substitutes usually buy the catalog, watch most of the seats go untouched, and conclude that AI enablement does not work.
A useful frame for the spend: a full day for 14 people in the middle of the range costs roughly what one mid-level employee costs the company in a month. So the question is not whether the day is expensive. It is whether the day produces one artifact per person that keeps getting produced that way.
Wiring measurement in at design time
Measurement bolted on afterward measures satisfaction, because satisfaction is the only thing you still have access to. Most AI enablement budgets die at renewal for that reason alone. Four instruments, all set up before the session.
A pre-session confidence rating. One question per artifact, asked in the task audit: how confident are you making this with AI today, one to five. Re-ask it at the close and again at day 21. Cheap, and it moves visibly.
The baseline time estimate. Already in the task audit. Re-measure the same artifact at day 21 against the same estimate. This is the number sponsors quote, so collect it properly.
Artifact production count. How many of the committed artifacts were actually produced that way in the following three weeks. This is the real adoption signal, and it comes from the commit round, which is why that block is not optional.
Prompt library contributions. Additions and edits per month by someone other than the facilitator. When the library stops moving, the program has stopped.
The Kirkpatrick levels are a useful sanity check on where your evidence sits: reaction (did they like it), learning (can they do it), behavior (are they doing it), results (did the business change). Most training buyers collect level one and report it as level four. Levels three and four take a quarter and a named owner, and proving capability moved covers how to build that case without inventing numbers.
The 14 days after the room empties
Treat the session as a start condition, not a finish line. The two weeks after it decide whether anything compounds, and the plan is short enough to fit on one page.
Day 1. The prompt library goes where people already work, not into a new tool nobody opens. The shared drive folder they use daily beats a purpose-built wiki every time.
Day 2. The facilitator or internal owner sends each person their own commitment back, in their own words. Not a group email. One line, to one person, naming their artifact and their date.
Day 7. A 30-minute group check-in with one question: what broke? Collect the failures, fix two of them live, and add the fixes to the library. This session matters more than the back half of the workshop did, and it is the first thing cut from most budgets.
Day 14. Managers ask one question in their regular one-to-ones: show me the last one you did this way. Not a survey. A request to see the artifact. BCG’s data on leaders and managers driving regular use while frontline use stalls is, read practically, an argument for this one question.
Day 21. Re-ask the confidence rating and re-measure the artifact time. Send the sponsor three numbers and one example artifact. Short readouts get read.
Name the owner of this plan before the workshop, and give them hours rather than goodwill. At most companies in the 50-to-500 band there is no Chief AI Officer and this landed on someone as a second job, which is survivable only if the plan is this small.
Reading a vendor’s proposed agenda
The method above doubles as a buying checklist. Most corporate AI workshop proposals arrive as a topic list with a price at the bottom, and six things in one should stop you.
- No named artifacts. If the agenda lists topics rather than the documents people will produce, nobody has designed backward and the day will not survive Monday.
- Under half the clock hands-on. Count the minutes. Blocks labeled “demonstration,” “overview,” and “case studies” are not practice.
- No verification block. A session that teaches speed without checking is a liability in any work that reaches a client, a regulator, or a board.
- Rooms above 20 with one facilitator. Ask directly how many people and how many facilitators. The answer tells you whether the comparison round can exist.
- Follow-through absent from the scope. No day-7 check-in, no library owner, no day-21 readout means you bought an event and will be asked to buy another one in six months.
- Model version numbers all over the deck. A curriculum pinned to specific model releases will be stale within two quarters. Good material teaches the workflow and treats the tool as interchangeable.
Two questions get you most of the way there without reading anything. What will each person walk out having produced, and who normally receives it? A vendor who has designed a real session answers both in one sentence. A vendor who has not will tell you about their framework.
Frequently asked questions
How long should an AI workshop for teams be?
Six working hours is the shortest format that reliably produces a verified artifact per person plus a plan. BCG’s 2025 workplace survey found regular AI use sharply higher among employees who received at least five hours of training with in-person coaching, which makes a full day roughly the floor rather than a generous choice. Four weekly 90-minute sessions produce better retention than one day, because the gaps force real use, but they are harder to schedule and need manager cover to keep attendance.
How many people should be in the room?
Twelve to 16 with one facilitator, and eight if you are running it remotely. The cap comes from the comparison round, where the facilitator reads several real outputs closely and helps the room name the difference between the best and the worst. Above 16 that round degrades into a show of hands. Three sessions of 14 cost more than one session of 40 and produce far more, which is the tradeoff most budget conversations get backward.
Should managers attend with their own teams?
Usually not in the same room. People produce bad first drafts in the cold attempt, and nobody wants to do that in front of the person who writes their review. Run a shorter session for the manager layer first on the same artifacts, so they can sponsor the work and ask for it in one-to-ones without watching the practice. Small peer-like teams are the one case against it, where splitting the group costs more than the hierarchy does.
Which AI tool should we train on?
The one your team already pays for, in the account tier they will have on Monday. ChatGPT, Claude, Gemini, and Microsoft Copilot are all capable of the drafting and analysis work a fluency session covers, and swapping tools after the workshop costs less than people expect once the prompt library exists. Train on the business tier rather than free accounts: the data-handling terms and the available features both differ, and you want the room learning in the environment they will keep.
What should people do before the session?
Four things, all finished before the date is set. A 15-minute task audit naming their five most frequent tasks with time estimates; five real cleared documents per artifact type; one tested login to the business-tier account in the browser they will use; and one or two written sentences from the sponsor on what should change after the day. If any of those is missing, move the date rather than running a diluted version.
How much does a one-day AI workshop cost?
Hands-on AI training for a team runs roughly $2,000 to $15,000 in the current market, driven mostly by how much of the material is built on your own documents, how many facilitators are in the room, how many cohorts you need, and whether follow-through is in the scope. Self-paced catalog seats cost far less per person and do another job, so verify each provider’s current pricing and curriculum directly before comparing. If the quote has no day-7 check-in and no day-21 readout, you are pricing an event rather than a program.
How do we handle the person who thinks this is a waste of time?
Put them to work first and in public. Give them the hardest artifact in the cold attempt and ask for their honest verdict during the comparison round. Skepticism in experienced people is usually a quality worry, not a technology worry, and the verification drill answers it directly by handing them a method for catching the errors they were worried about. What does not work is arguing with them from the front of the room.
How soon can we tell whether it worked?
Three weeks. Re-ask the confidence rating from the task audit, re-measure time on the same artifact, and count how many of the commitments people wrote at the close were actually produced that way. If artifact production is near zero at day 21, the problem is almost always social rather than technical: no manager asked to see the work. Business-level results take a quarter and need a named owner, not a survey.
Can HR or an internal team run this in-house?
Yes, once someone internal has facilitated the loop at least twice. The hard parts are not the content: they are running the comparison round without giving the answer, and holding the hands-on clock when the room wants to be talked to. A workable path is to have a vendor run the first two cohorts with your internal owner co-facilitating, then hand over the prompt library and the run of show. That is a reasonable thing to ask for in the contract, and vendors who refuse are telling you something.
What is the difference between an AI workshop and an AI training course?
A course teaches a syllabus to individuals who complete it on their own schedule; a workshop changes how a specific group produces a specific document, in one room, on their own material. Courses are cheaper per seat and work for keen self-starters. Workshops cost more and are the only format that reliably moves a team whose licenses are already bought and sitting unused, because the practice is on real work with a real recipient.
Key takeaways
- Design backward from a named artifact. Pick the recurring document each person owes somebody next week, write the Monday test, then cut everything that does not shorten the path to it.
- Spend at least half the clock on their own work in their own tool. BCG’s 2025 data puts the practical floor near five hours with in-person coaching, which rules out the 60-minute lunch-and-learn as a capability play.
- The cold attempt and the comparison round are the engine. Let people fail on their real task first, then read the best and worst outputs side by side and let the room name the context gap itself.
- Teach verification as a timed drill, not as advice. Speed without checking is a liability in any work that reaches a client or a board.
- Cap the room at 12 to 16 with one facilitator (eight if remote), group by shared artifact rather than by department, and keep managers out of their own reports’ cohort.
- Cut what happens inside the model, prompt taxonomy, agent building, and tool-stack strategy. Cutting is what raises retention.
- Budget the 14 days after the room empties: library in place on day 1, personal commitment echo on day 2, a “what broke” session on day 7, managers asking to see the work on day 14, three numbers to the sponsor on day 21.
- When reading a proposal, ask two questions. What will each person walk out having produced, and who normally receives it?
Name your three artifacts this week before you talk to anyone about delivery; it takes an hour and it will tell you more about what you need than any vendor agenda will. If you want an outside read on which artifacts to start with and whether a session is even the right first move, book a free AI-readiness call →.
Arthur Wandzel
