90% of your training evaporates: the 2 techniques that fix it

I have delivered AI trainings to 300+ people across 6 countries, and my own industry has a problem it rarely names: the standard one-shot training day is a machine for forgetting. People leave energized, and 3 weeks later they have kept perhaps a tenth of it. That decay is one of the oldest and most replicated findings in psychology, and the fixes are as well documented as the problem.

The forgetting curve replicates.

In 1885, Hermann Ebbinghaus memorized lists of syllables and tested himself at increasing intervals. Retention collapsed fast: most forgetting happened within the first day, then the curve flattened. In 2015, Murre and Dros redid the experiment with modern controls and got the same curve ¹. A hundred and thirty years apart, same result. Teach something on Tuesday and most of it is gone by Friday, unless you interrupt the curve.

Two techniques interrupt it. Both are cheap, and both stay underused because they feel wrong while you use them.

Fix 1: make people retrieve instead of re-reading.

Roediger and Karpicke (2006) had students learn short science texts. One group restudied the text several times. Another studied it once, then took recall tests instead of restudying. Five minutes after learning, the restudy group looked better. One week later, the ranking flipped: the students who took the tests remembered far more than the ones who reread ².

Two details matter for anyone who teaches. First, the effect comes from the act of pulling information out of memory: testing without corrective feedback still beat restudying. Second, the restudy group was more confident about how much they would remember, and more wrong. Re-reading feels fluent, and fluency masquerades as knowledge.

The 2013 monograph by Dunlosky and colleagues reviewed 10 popular learning techniques against hundreds of studies. Only two earned the top utility rating: practice testing and distributed practice. The techniques most people rely on, highlighting, rereading, summarizing, all rated low ³.

Fix 2: space it, with a formula.

Cramming works for tomorrow and fails for next quarter. The question is how far apart to space the reviews. Cepeda and colleagues (2008) ran a study with 1,354 participants across gaps from minutes to months, and found the answer depends on how long you need to remember: the optimal gap between study sessions is roughly 10 to 20% of the retention interval ⁴.

That gives you arithmetic instead of intuition. Want a team to still master a workflow in 60 days? The first review should come about 6 to 12 days after the initial session. Want retention at one year? Review around the 1-to-2-month mark. A review the next morning feels diligent and is mostly wasted; space it out to the point where recall takes effort and you buy months of retention with the same 30 minutes.

The protocol I use in corporate trainings.

These two findings translate into a training design, whether you teach AI tools, onboarding, or compliance:

1. Close each session with retrieval instead of a recap. Last 15 minutes: blank page, "write the exact steps you will use on Monday, from memory". No slides visible. A recap by the trainer is restudy; it produces confidence without retention ².

2. Replace demos with attempts. Watching me prompt an AI model is fluent and forgettable. In my sessions, participants do the task in the tool within minutes, fail a little, then get the correction. The early failure is the part they remember.

3. Schedule the follow-up with the 10-20% rule. Decide the date where the skill must still be alive, take 10 to 20% of that interval, and put a 30-minute review there ⁴. For a typical "must still work in 3 months", that is a review at week 2. Without this single calendar invite, you are buying the forgetting curve at full price ¹.

4. Make the review a quiz. Five real scenarios, answered cold, then corrected. Retrieval plus spacing stack: each act of difficult recall flattens the curve further ²³.

5. Distrust the smile sheet. Post-training satisfaction measures fluency, and fluency is the thing that lies ². Measure instead: can they do the task, cold, 2 weeks later? That number is the training.

The best-feeling training (smooth, complete, impressive demo, glowing feedback) and the best-working training (effortful, spaced, quiz-heavy) sit close to opposites. Choose which one you are optimizing for.

Sources.

1. Murre, J.M.J., Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLOS ONE, 10(7). doi:10.1371/journal.pone.0120644
2. Roediger, H.L., Karpicke, J.D. (2006). Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention. Psychological Science, 17(3). doi:10.1111/j.1467-9280.2006.01693.x
3. Dunlosky, J., et al. (2013). Improving Students' Learning With Effective Learning Techniques. Psychological Science in the Public Interest, 14(1). doi:10.1177/1529100612453266
4. Cepeda, N.J., et al. (2008). Spacing Effects in Learning: A Temporal Ridgeline of Optimal Retention. Psychological Science, 19(11). doi:10.1111/j.1467-9280.2008.02209.x