The format question for a distributed team is usually framed as online against in-person. The best available evidence says that axis barely matters. In the US Department of Education’s meta-analysis of 50 controlled contrasts, purely online and purely face-to-face instruction came out statistically equivalent, at +0.05 with a p-value of .46. What separated the winners was something else: blended delivery beat face-to-face at +0.35, p < .001, and instructor-directed online work beat independent self-paced work by a wide margin, +0.39 against +0.05 that failed significance.
So the real choice for a fully remote company is not where the room is. It is whether a person runs the session, whether anything follows it, and how many live blocks your time-zone spread forces you to buy. That last one is arithmetic, and no vendor page will do it for you.
This is for whoever authorizes training spend at a US company of 50 to 500 people with no shared building. You hold the licenses, someone upstairs wants evidence they were worth buying, and every page you have read ended with “use a blend of both” and stopped there.
Every page on this question compares the wrong two things
Read who publishes the first screen of results here: a video software blog, five near-identical marketplace pages with the topic word swapped, a vendor arguing that virtual scales and in-person is for leadership, a course affiliate, and learning-platform posts on synchronous against asynchronous delivery. All conclude that you should use some of each. None says how much of each, in what order, or at what cost.
Not one cites a controlled comparison. Evaluation of Evidence-Based Practices in Online Learning, the 2010 Department of Education review by Means and colleagues, is the closest thing to settled evidence here, and it appears in no result in any channel I searched.
The report is careful about its own finding. Blended arms often ran longer and carried extra instructional elements, so the authors warn the advantage “is not necessarily rooted in the media used per se.” Read what that caveat concedes: the gain came from added time and added structure, the first two things a distributed rollout drops because they are hardest to schedule.
Two limits worth stating. The studies are education rather than corporate training, and they predate 2009. The 43 contrasts with older learners, graduate students and working professionals among them, pointed the same way.
Put the independent-learning number in front of your CFO: learners working alone produced +0.05, which did not reach significance. A license handed out with an announcement is that arm of the study. The same point from the retention side is in our comparison of platform and facilitated retention, and the wider case sits in why buying licenses is not the same as building capability.
Start with your overlap window, not your format preference
Before picking a format, count your shared hours. Standard offsets settle it, and the answer dictates most of what follows.
| Team spread | Shared working hours | Live blocks a 3-hour session forces |
|---|---|---|
| US only, Eastern to Pacific | About 5, roughly 12:00 to 17:00 ET | One |
| US plus Western Europe | 2 to 3, roughly 09:00 to 12:00 ET | One with zero slack, or two shorter blocks |
| US plus India | About 1, at the edge of both days | Two blocks, or two cohorts |
| US plus East Asia or Australia | None inside normal hours for both | Two cohorts, always |
A US-only team has one real constraint and it is lunch. Add Europe and a three-hour session fills the window exactly, so one overrun keeps half the group past dinner. Add India and the window is too narrow to hold a working session. Add Sydney and this stops being a scheduling problem: you are buying two deliveries, and the question is whether you knew before signing.
Which is why the first question to a vendor is not “do you deliver virtually.” Everyone does. It is “how do you price the second block,” and the answer says whether they have taught a team shaped like yours.
Six formats, and who each one fails
| Format | Right when | Wastes the budget when |
|---|---|---|
| One live virtual session, whole team | Overlap is 4 hours or more and the group is under about 20 | Spread exceeds 6 hours, or headcount is high enough that nobody gets looked at |
| Time-zone cohorts, same session run twice | Spread over 6 hours, or headcount over 20 | Overlap is wide enough for one block and you pay twice for scheduling comfort |
| Short live blocks over several weeks | The behavior has to survive a return to real work | It becomes six half-attended calls with no artifact |
| Platform license, self-paced | The need is breadth across many topics at low cost per seat | It is bought to change one named behavior in a known group |
| Live core plus asynchronous practice between sessions | You can staff facilitation and someone will read the practice work | Nobody is assigned to read it, so practice becomes homework nobody grades |
| In-person offsite, flown in | Leadership alignment, sensitive material, one annual anchor event | Used for skills that needed four touches, because travel buys one |
Time-zone cohorts are not a worse version of one big session. They are usually better, and the clock is doing you a favor. Work on your own documents needs a facilitator who can look at individual screens, which caps group size far below a webinar’s. Published catalog classes are sized accordingly: Noble Desktop lists its AI for Business Bootcamp at 12 hours and $549 a seat, and live classes of that kind run in the single or low double digits, not a hundred people on mute. Splitting 30 people into two cohorts of 15 is the shape you wanted anyway.
Live core plus asynchronous practice is the format the Department of Education evidence favors and the one most often bought then abandoned. The blended gain came from extra time and structure. Set practice work that nobody reads and you have the structure without the attention, which is the independent arm of the study again. How this interacts with pacing is covered in the sequencing decision between one department first and company-wide.
What a recording carries and what it does not
The default remote move is to run the session for whoever can make it and email the recording to the rest. It is free, it feels fair, and it quietly converts a directed session into independent study for everyone watching later.
A recording transfers what was said. It does not transfer the three things a live block is bought for: a facilitator looking at your document and naming what is wrong with your prompt, the correction a colleague gets that you also needed, and a calendar entry that makes two hours happen.
Treat recordings as reference: useful for someone who attended and wants to repeat a step, or a new hire later. Not a substitute for a cohort. A group that cannot attend live needs its own block, and the pattern of absence is itself a finding.
Set the recording policy first, since hands-on practice means showing real work with the mistakes still in it. State what is captured, who sees it, and how long it is kept.
What a second live block actually costs
Nobody publishes prices for private facilitated programs, including vendors with the most open catalogs. The market range is roughly $2,000 to $15,000 per engagement depending on headcount and session count. Platform seats are the opposite: Coursera for Business lists its Teams tier at $399 per user per year for 5 to 125 users, priced to the cent on a public page. Figures checked October 2026.
Give it a shape. Thirty people across the US and Europe, one three-hour working session, split into two cohorts of 15 because the window and the headcount both demand it.
| Line | Figure |
|---|---|
| Engagement in the facilitated range | $6,000 |
| People reached | 30 |
| Cost per head | $200 |
| Platform seat at list, same 30 people | $11,970 per year |
The facilitated session is not the expensive option here. It looks expensive because it arrives as one quoted number while licenses arrive as a line item someone already approved.
The piece to negotiate is the second block. Preparation is a one-time cost and delivery is not, so a repeat of the same material should not carry the first block’s rate. Ask for the repeat-delivery figure separately, in writing. A vendor quoting two identical full-price deliveries is charging twice for one preparation. One who cannot quote it has not done this before. The rest of that conversation is in how to negotiate an AI enablement engagement.
Which cohort goes first when nobody shares a building
In an office you would start with whoever sits nearest the training room. That proxy is gone.
Start with the time zone holding the most people who do the same work. One cohort of 15 in similar roles beats three cohorts of five grouped by location, because the examples carry. The second cohort gets a session improved by the first one’s questions, so space the blocks a week apart rather than back to back.
Buy one block before committing to a program. A single paid session is the cheapest evidence about whether a vendor can hold a remote room, and the case for it is in running a paid pilot workshop before a company-wide commitment. It also tells you what no reference call will: whether the facilitator can read a room through a grid of thumbnails. If you have to change course partway, replacing a training vendor mid-program covers what to salvage.
On fatigue, which gets overclaimed: Jeremy Bailenson’s 2021 paper in Technology, Mind, and Behavior proposed four mechanisms behind video-call exhaustion, among them close-up eye gaze, cognitive load, constant self-view, and reduced movement. The paper states plainly that these are arguments not yet directly tested. Use them as design guidance (shorter blocks, breaks on the hour, self-view hidden), not as a reason to buy an offsite.
Ten questions to ask before you sign a remote delivery
These separate firms that deliver remotely from firms that deliver in person over video.
- Maximum group size per facilitator for hands-on work? Above roughly 20, nobody’s screen gets looked at.
- Which time zone is the facilitator in? An 06:00 local facilitator is a different purchase.
- How is a second cohort priced? Separately, with preparation counted once.
- Do people work on their own documents or your sample files? Sample files are a demonstration, not practice.
- What happens in the breakouts, and who sits in them? Breakout rooms with nobody in them are a break with extra steps.
- Who owns the recording, how long is it kept, and can we decline? Settle this before the session.
- What follows the session, and is it included? Office hours, a practice task someone reads, a check-in at four weeks.
- What does each participant leave with? A working artifact built on their own material, or a certificate of attendance.
- How do you handle the person who can attend neither block? “Send the recording” is an answer, and now you know what you bought.
- What do you measure, from which source, and when? License usage exported before session one is the one baseline you already own.
Re-check any named provider’s price and curriculum on their own page before signing. These catalogs reprice quarterly.
A decision table you can defend
| Your situation | Buy this | Not this |
|---|---|---|
| Under 20 people, US only, one named behavior to change | One live virtual working session, plus a check-in at four weeks | A platform license for 20 seats |
| 20 to 60 people, spread over 6 hours or more | Time-zone cohorts of 12 to 18, repeat priced separately | One all-hands webinar |
| Broad curiosity, no single target behavior, tight budget | Platform license with a named owner and a review cycle | A one-off session with no follow-up |
| Capability has to survive a quarter of real work | Live core plus practice between blocks, read by someone | Recordings and good intentions |
| Leadership alignment on AI policy and spend | One in-person anchor session if travel is already budgeted | Four virtual leadership calls nobody finishes |
If the answer keeps landing on “build it ourselves,” that is a different decision with its own arithmetic, worked through in the buy-versus-build decision for enablement capability.
Frequently asked questions
What is the best AI training format for a fully remote team?
A live, facilitated virtual working session on the team’s own documents, split into time-zone cohorts of 12 to 18, with one practice task between blocks and a check-in at four weeks. The controlled evidence favors directed and blended delivery over both unaccompanied self-paced study and a single face-to-face event. Keep hands-on cohorts under about 20 per facilitator.
Is virtual AI training as effective as in-person?
On the available evidence, yes. Across 50 controlled contrasts the two were statistically equivalent, at +0.05 with p = .46. What moved results was direction and blending, not location.
How long should a virtual session be?
Blocks of two to three hours with a break on the hour, not a full day. A six-hour remote session usually survives as four hours of attention plus two hours of email, and it does not fit a three-hour overlap window at all.
How do you handle a team spread across more than six hours?
Stop hunting for one slot. Run the same session twice for two time-zone cohorts and negotiate the repeat delivery as a separate line, since preparation is a one-time cost.
Can we just send everyone the recording?
Use recordings as reference for people who attended and for later new hires. A recording carries what was said, not the correction of your own work, and a group that cannot attend live needs its own block.
How much does AI training for a distributed team cost?
Private facilitated programs sit in a market range of about $2,000 to $15,000 per engagement depending on headcount and session count. Platform licenses are published: Coursera for Business lists its Teams tier at $399 per user per year. For 30 people, one facilitated engagement often costs less than a year of seats.
Does a self-paced license ever make more sense?
Yes, when the need is broad coverage of many topics at low cost per seat and nobody has committed to one behavior change. It needs a named owner and a review cycle, because the arm of the research where learners worked alone produced no significant effect.
Should we fly everyone in once a year?
Only if travel is already budgeted and the purpose is alignment or sensitive material rather than skills. An offsite buys one touch at the highest cost per touch available, and the habits you want need several touches.
What should we measure after a remote session?
Export license usage before the first block, then compare at 30 and 60 days alongside the workflows people rebuilt and the artifacts they produced. Hours saved and revenue are not attributable to one session and will not survive scrutiny.
Key takeaways
- Online against in-person is the wrong comparison. In the 50-contrast Department of Education meta-analysis the two were equivalent, while directed and blended delivery pulled ahead and self-paced study alone showed no significant effect.
- Count your shared working hours first. A spread over six hours settles it: you are buying two deliveries, and the price of the second is the term to negotiate.
- Keep hands-on cohorts under about 20 per facilitator. The clock usually forces that split anyway.
- Recordings are reference, not delivery. A group that cannot attend live needs its own block.
- One paid block tells you whether a vendor can hold a remote room. Cheaper evidence than a reference call.
The format you can staff beats the format you prefer, and a clock decides most of that. For a second opinion on the proposal on your desk, or on a schedule you are trying to make work across four time zones, book a free AI-readiness call and we will go through it with you.
Arthur Wandzel
