No. An annual AI workshop refresher is the wrong unit, and it is wrong by about eleven months. The follow-up that protects your first session belongs three to four weeks after it, and the pass that keeps people current belongs each quarter, aimed at what the products changed rather than at what the slides said.
The annual line item exists for reasons that have nothing to do with how people learn. Budgets renew once a year, catalog subscriptions are priced per user per year, and training providers are openly advised to build recurring revenue on refresher sessions and renewals. So the question arrives already shaped: should we do this again next year? That is a renewal question wearing a learning question’s clothes.
The better question is where the capability leaks, how fast, and what stops it most cheaply. The answer is a short second sitting while the skill is warm, a standing place to ask questions, something brief each quarter on what changed in the tools, and AI enablement inside onboarding so new hires are not waiting for next year.
The short answer, by situation
The right cadence depends on when you last ran a session and what has happened since.
| Your situation | What you need | Not this |
|---|---|---|
| Session ran in the last six weeks | A 60 to 90 minute second sitting on the team’s own work, then a standing slot for questions | An annual date in the calendar |
| Session ran three to six months ago, usage patchy | A clinic on the two or three tasks that stalled, plus a quarterly update pass | A full repeat |
| Session ran a year ago, nothing since | A first session again, scoped from where people are now | Refresher pricing, first-session expectations |
| New hires since the session | Fluency inside onboarding, in the first two weeks | Waiting for the next all-hands session |
| Your approved assistant shipped something the team needs | A two-hour pass on what changed | Twelve months of drift |
| A policy, insurer or client wants a training record | A dated session with an attendance record | Treating the record as the capability plan |
Read rows two and three together, because the gap between them is where most budgets land. A team that has gone twelve months without practice is not a refresher case. The skill is largely gone, and pretending otherwise buys a session pitched over the heads of the people in the room.
Twelve months is the worst place to put the only follow-up
Saks and Belcourt put numbers on this twenty years ago and the numbers have aged well. Across 150 organizations, the share of employees still applying what they were trained on ran at 62% right away, 44% at six months, and 34% at one year. Watch the shape rather than the levels. Most of the loss lands in the first half of the year, the stretch an annual cadence leaves unattended, so by the time the refresher runs two thirds of your people have stopped.
The skill-decay literature has a sharper edge for this subject. Arthur, Bennett, Stanush and McNelly pooled 189 data points from 53 studies and found loss running from an effect size near zero right after training to d = -1.4 after more than 365 days of nonuse. Their moderators matter more than that headline: cognitive, artificial and accuracy-based tasks decayed faster than physical, natural and speed-based ones. Writing a good prompt is cognitive, artificial and accuracy-based, which is the worst case in that taxonomy. A year of nonuse does not leave a rusty version of the skill; it leaves someone who remembers the session was useful.
We see the softer version in the room. Ask a group three weeks out what they changed and you get specifics: the weekly report, the first pass at a proposal, inbox triage. Ask at six months with nothing in between and you get two enthusiasts and a lot of polite silence. The enthusiasts kept practicing. Everyone else had a quarter end.
The spacing research puts the second pass three weeks out
If the decay is front-loaded, where does the second exposure belong? That has an answer, and it is more precise than any cadence on the market.
Cepeda, Vul, Rohrer, Wixted and Pashler ran over 1,350 people through a design that varied both the gap between study and review and the delay before the final test. The result is a ridgeline rather than a straight line: performance rises with the gap, peaks, then falls. The optimal gap was about 20% of the test delay at delays of a few weeks, falling to roughly 5% when the delay was one year.
Do the arithmetic as a buyer. To still have the capability in twelve months, the high-value second pass sits at about 5% of twelve months, which is roughly three weeks. Not three months. Not next year. Three weeks.
That short, unglamorous session a month after the first one does more for twelve-month retention than a full repeat in month twelve, at a fraction of the cost. Most buyers already have the money for it, sitting in a line labeled next year. BCG’s 2025 survey of more than 10,600 leaders, managers and frontline employees across 11 countries agrees on dose: only one third said they had been properly trained, and regular use rose sharply at five or more hours of training, with in-person training and coaching strengthening the effect.
A refresher has to chase the product, not the slide deck
Even if your people lost nothing, the tools would have moved. You can check this one yourself in a minute. Microsoft publishes release notes for Microsoft 365 Copilot on a roughly two-week cadence, and the batch covering September 22 to October 6, 2026 lists about fifteen distinct feature entries across Copilot, Excel, PowerPoint, Outlook, SharePoint and the admin center. ChatGPT, Claude and Gemini publish their own changelogs on similar rhythms.
An annual refresher runs about twenty-six release batches behind the product on the day it starts.
That is not an argument for training twenty-six times a year. It is an argument about content. A refresher built from last year’s deck teaches a product that no longer exists in that form, which is how a session earns a reputation as a wasted afternoon. One built from a release-notes diff plus what your team got stuck on is the only kind worth buying.
Notice who benefits from the annual framing. Coursera prices its business team plan per user per year for up to 499 users, and Section sells membership at $62.50 per month billed annually with recordings and materials after each course. Both are reasonable products whose unit of sale is a year of access, because that is how software renews rather than how learning works.
Credit where it is due: one provider has read the room. Mostly Serious publishes a Platform Updates session, quarterly or bi-annual, two hours, with standalone sessions from $3,500 and continuing plans at $10,000 to $30,000 a year. Quarterly and tied to platform change beats annual. What is missing is the measurement that would say whether the quarterly pass earns its line.
Four problems wearing one name
“Do we need an annual refresher” is four questions stacked on each other, running on four different clocks, and the annual session is sold as the one purchase that handles all of them.
Individual skill decay. Weeks to months, front-loaded. Cheapest fix: a short second sitting at three to four weeks on live work, plus a standing slot where people bring the task that is fighting them. Not a workshop. A clinic.
Product change. Two to four weeks per release batch, with a real shift every quarter or two. Cheapest fix: a two-hour quarterly pass built from the release notes for the tools you approved, with your own documents on screen.
Roster change. Continuous. BLS puts median tenure with a current employer at 4.1 years in January 2026, and 3.9 years in the private sector, so an annual session trains whoever is in the room that day and leaves anyone who arrives in month two waiting ten months. Cheapest fix: fluency in the onboarding sequence, which costs little once the material exists and someone internal owns it.
The paper record. Whatever your policy, insurer or largest client says. Cheapest fix: a dated session with a roster and a written set of data rules. This is the one need an annual cadence genuinely meets, so name it as the compliance artifact rather than mistaking it for a capability plan.
Split the four and the budget gets easier, because three of them are cheap and only one looks like a workshop.
What to buy instead of an annual refresher
A schedule that survives contact with a real calendar, for a team that has had one good session:
- A 60 to 90 minute second sitting, three to four weeks out. Agenda set by what people tried and where it broke, not by a deck. Highest return on the list, and the item most often skipped.
- A standing slot for questions. A recurring half hour where whoever is stuck brings the real task. Cheap, internal if you have someone capable, and it catches a dying habit while there is still something to save.
- A two-hour pass each quarter on what changed. Built from release notes plus the stalls logged since the last one. Private hands-on AI training sits in a market range of roughly $2,000 to $15,000 by length, headcount and customization, and a short follow-on is normally quoted well under a first full day.
- Fluency inside onboarding. Every new hire in their first fortnight, using material you already paid to have built. That needs someone to own the material, which is a question to settle before the first session, as we work through in the piece on who keeps the curriculum once the trainer leaves.
- One measurement per quarter. Pick a task, time it before and after, keep the record. Without it you will have this same argument next year with nothing but opinions.
Place the sittings where the hours already exist. For teams that bill their time, that constraint shapes the schedule more than any learning theory, as our piece on scheduling around a team that bills by the hour sets out. The design logic underneath all of it sits in our session design guide.
When an annual session is the right call
Four cases where a yearly date is the correct purchase. A low-change function with a stable toolset, where two prompts cover two tasks and nobody wants more. A documented training obligation, where a contract, an insurer or a policy wants evidence of training on approved use; keep it short, keep it about rules, keep it separate from the capability work. Headcount growth in waves, where one session a year serves the new intake while the existing team runs on a different clock. And a genuine restart, where a year of silence means a first session; price it and brief it as one, because buying a refresher and expecting first-session outcomes is how a budget gets quietly reallocated.
How to tell a refresher from a replay
Four questions separate the real thing from a replay of last year’s deck. Ask them before you sign.
- “What is in this one that was not in the last one, and where did it come from?” Release notes, your own stalled tasks, workflows that did not exist a year ago. A provider who cannot name the source plans to reuse the deck.
- “Which tasks are we re-practicing, and how did you pick them?” The answer should reference what people struggled with. If nobody collected that, collecting it is the first job.
- “Who is in the room, and are new hires on a separate track?” Mixing ten months of practice with none bores one group and loses the other.
- “What do we keep?” Updated prompts your team wrote and tested, a revised page of data rules, before-and-after timing on one task. A video of the session is the weakest artifact available, for reasons we lay out in the piece on recording a session and what it costs you.
Vague answers mean you are being sold a renewal. That is the pattern underneath most AI spending that shows activity without capability, and the case we make at length in our argument against buying licenses and calling it enablement.
Frequently asked questions
Do you need an annual AI refresher workshop?
No, not as your main follow-up. Application of training falls from 62% right after a session to 44% at six months and 34% at a year, so an annual cadence reinforces nothing until most of the loss is done. Book a short second sitting three to four weeks out and a two-hour pass each quarter on what the tools changed.
How often should a team refresh its AI training?
A 60 to 90 minute follow-up at three to four weeks, a standing slot for questions, a two-hour quarterly pass tied to product change, and fluency inside onboarding. Teams with a stable, minimal toolset can run yearly and be fine.
Why three to four weeks and not three months?
Across more than 1,350 participants, the best gap between a session and its review was roughly 5% of the retention interval at a one-year delay. Five percent of a year is about three weeks.
Is a refresher cheaper than the first workshop?
Usually, and it should be. A corporate AI workshop sits in a market range of about $2,000 to $15,000 depending on length, headcount and customization, and a short follow-on is normally quoted well below a first full day because the discovery and the materials exist. If a provider quotes a repeat at the original price, ask what is being rebuilt.
What if nobody has touched the tools in a year?
Then you are buying a first session, not an AI workshop refresher. More than 365 days of nonuse puts skill loss at about 1.4 standard deviations in the pooled literature, and prompting is the kind of task that decays fastest. Scope from where people are now.
Should the refresher cover new features or re-teach the basics?
Both, with the stalls your team logged setting the agenda. Re-practice the two or three tasks that did not stick, then spend the rest on what changed in the products since.
Can we run the refresher internally?
Often yes, and it is the cheapest version that works. It needs one person who owns the material, keeps the data rules current and runs the question slot. Settle that ownership before the first session or the internal option quietly disappears.
How do we know whether the refresher was worth it?
Pick one real task per quarter and time it before and after. Attendance and license dashboards will not tell you, and one timed task written down makes the next budget conversation short.
Do new hires need their own session?
They need the material, not necessarily a session. Median tenure with a current employer is 4.1 years, so a team of any size is always onboarding somebody. Put a short fluency block in the first two weeks using the prompts and rules the original session produced.
Key takeaways
- An annual refresher is the wrong unit. Application of training falls to 44% by six months and 34% by a year, so a yearly cadence reinforces nothing until most of the loss is done.
- The valuable second pass lands at roughly 5% of your retention target, which is about three weeks out for a one-year target. It is also the cheapest session you will buy.
- The product moves faster than the budget cycle. Copilot feature batches publish about every two weeks, so last year’s deck teaches a tool that has changed underneath it.
- Four problems hide behind one request: individual decay, product change, roster change and the paper record. Each has its own clock and cheapest fix, and only one is shaped like a workshop.
- A year of silence means a first session. Price it and brief it that way, and set the cadence before you leave the room.
Set the cadence before you book, not after the first session goes quiet. A free AI-readiness call will tell you where your team sits on the decay curve, which follow-up format fits your calendar, and whether you need a refresher or a restart.
Arthur Wandzel
