Forty minutes into a feedback workshop, a trainer senses the room going stale. The planned case study is about a retail chain, and the cohort sells to distributors. So she drops the case, tells a story from her own sales days, takes three questions, and runs the practice ten minutes late with half the pairs watching instead of doing. The feedback forms say she was engaging. Six weeks later, their managers report that feedback conversations still happen the way they always did: badly and rarely.
The session felt alive. It taught nothing new. This is the failure mode nobody audits, because the room enjoyed it.
The trainer who seems able to handle anything is usually described as a natural: charisma, quick wit, years on stage. That description is wrong, and it tells L&D teams to hire for personality instead of building preparation. Improvisation is not a trait. It is what over-preparation looks like when the plan meets a real room. The trainer who can depart from the script safely has memorised the parts that matter so thoroughly that her attention is free to read the room. The trainer who cannot has to spend that same attention remembering what comes next.
Why improvisation is a memory problem
Cognitive load theory starts from two facts about the mind. Working memory, the space where conscious thinking happens, is small: the NSW Centre for Education Statistics and Evaluation’s research summary puts it at roughly four chunks of new information at once, citing Cowan’s 2001 review. Long-term memory, by contrast, has no known capacity limit, and knowledge stored there as practised schemas can be run with almost no conscious effort: automaticity occurs after extensive practice, and automated schemas free working memory for new information.
Count what a trainer holds in working memory during a live session. The content of the next block. The timing. The names and moods of twenty-four people. The question someone just asked that might be a genuine objection or a status play. If the content and the sequence are not automated, they occupy most of those four slots. There is nothing left to read the room with. The script is not a cage for a novice trainer. It is her memory, propped up on paper.
This is why the confident departure goes wrong. The trainer frees attention by dropping the plan but has stored nothing to rebuild it, so she improvises toward whatever is easiest to hold in mind: her own stories, the loudest participant, the slide she likes best.
The expertise researchers have a name for the difference. Hatano and Inagaki, in 1986, distinguished routine expertise, mastering a procedure until it is fast and accurate, from adaptive expertise, understanding why the procedure works well enough to modify it or invent a new one when conditions change. The routine expert delivers the plan beautifully and collapses when it stops fitting; the adaptive expert knows what each piece is for and can swap pieces without losing the point.
The evidence that this is built, not born, comes with a caveat worth stating. Ericsson, Krampe and Tesch-Römer’s 1993 study of violinists found that by age 18 the best group had accumulated an average of 7,410 hours of solitary practice, the good group 5,301, and the future music teachers 3,420. But a 2019 double-blind replication by Macnamara and Maitra did not reproduce that clean correspondence between practice hours and skill level; the effect was real but considerably smaller than the original. The useful part is not the number: the gap between freezing and adapting sits in what was practised until automatic, and that is a design decision.
Four things to know cold
Here is the preparation, in the order the work actually happens. Each item is something the trainer must be able to produce without notes, because in the moment of improvisation there is no time to look anything up.
One: the behaviour and its evidence. One sentence: by the end of this session, participants can do X, and I will see it when they do Y. For a feedback workshop: participants can open a corrective conversation with an observed behaviour rather than a verdict, and I will see it in the paired practice when the first sentence names something the other person actually did. If she cannot say it from memory, she cannot protect it when she departs from the plan; every swap gets tested against it: does this still produce the evidence?
Two: the spine, with the load-bearing parts marked. A session has three to five waypoints that carry the learning, and a handful of blocks that are delivery vehicles. The trainer marks each block as load-bearing or replaceable, and knows the minimum viable path: the shortest sequence that still produces the behaviour. Improvisation is then a bounded operation. Replaceable blocks can be swapped, compressed or dropped. Load-bearing blocks cannot, whatever the room’s mood. Most bad improvisation is a load-bearing exercise being dropped because it felt slow, and slow is exactly what practice is supposed to feel like.
Three: two examples per core concept, one from the cohort’s world. Cases are the parts you swap; the concept stays. A trainer who knows only the textbook case is welded to it, because dropping it means dropping the concept too. A trainer who can teach the same point with a distributor credit-block case or a retail complaint case can change vehicles in ninety seconds without touching the learning. She prepares the second example from the cohort’s industry, not from her own past.
Four: three recovery moves, decided in advance. The derailments are predictable: an exercise dies early, a senior participant takes the room, the technology fails. For each, one pre-decided move, rehearsed once. If the exercise dies, run it live with the same checklist. If the senior participant hijacks, give him the observer role with a written task. If the projector fails, the case goes on the flipchart because the spine was drawn on paper anyway. Deciding in advance means the decision costs no working memory in the moment; it is already a schema.
This is the head-side complement to the handover: the run sheet, failure scripts and checklist in the handover document a train-the-trainer programme actually uses put these decisions on paper; the four items are what must stay in the trainer’s head when the paper is face down.
The blank-page test
Preparation claims are cheap. This test costs ten minutes and cannot be faked.
The day before delivery, the trainer redraws the session from memory on a blank sheet: the behaviour sentence, the waypoints with timings, the load-bearing marks, and the second example for each core concept. No deck, no guide, no phone. Then, for every block, she says one sentence out loud about what it is for.
Two failure signals. If a waypoint is missing from the drawing, that part of the session is not yet hers, and she is not allowed to improvise around it. If she cannot say what a block is for, the block is decoration: cut it, or learn its job. A session that passes the test can survive a dead projector, a hostile room and a fire alarm. A session that fails it should be delivered exactly as written, because the script is the only thing holding it together.
A worked example
The numbers below are illustrative, built to show the decision, not a client result.
A trainer is preparing a four-hour feedback workshop for twenty-four pharmaceutical sales representatives. Her spine, drawn from memory the night before:
| Time | Block | Minutes | Load-bearing |
|---|---|---|---|
| 09:00 | Opening and behaviour contract | 20 | Yes |
| 09:20 | Modelled conversation, done right | 25 | Yes |
| 09:45 | Paired practice, round one | 40 | Yes |
| 10:25 | Break | 15 | No |
| 10:40 | Case study: retail complaint | 35 | No |
| 11:15 | Paired practice, round two, observer checklist | 45 | Yes |
| 12:00 | Objection clinic | 30 | No |
| 12:30 | Commitment and manager copy | 30 | Yes |
At 10:40 the retail case lands flat. The reps sell through distributor stockists, and the room cannot see themselves in a shop-floor complaint. This trainer runs her pre-decided move: the case dies, so it goes live. She takes a real distributor credit-block dispute from the room’s last month and assigns the same observer checklist the block was always going to use. The swap takes ninety seconds. Round two starts at 11:15 with the checklist intact, which means the behaviour evidence still gets produced.
What changed: the vehicle. What did not change: the concept, the checklist, the two practice rounds, the timing of everything load-bearing. That is the whole skill. The room experienced it as responsiveness. It was preparation, cashed in at the moment it was needed.
Where this advice stops
Three honest limits.
First, preparation cannot rescue a broken design. If the session’s objective is wrong, or the practice never resembles the job, improvising around it decorates the failure. The blank-page test tells you whether the trainer knows the plan. It does not tell you whether the plan is worth knowing. That question belongs to the design review, before anyone is certified to deliver.
Second, some content is fixed by rule. Compliance modules, certification syllabi, safety-critical procedures: the mandated content does not get improvised, ever. The freedom described here applies to method and examples, never to what a regulator or a standard requires to be taught verbatim. A trainer who improvises the content of a safety briefing is not adaptive. She is a liability.
Third, the research base is direction, not prescription. The practice-hours ladder was only partly replicated, and the four-item checklist above is practitioner judgement built on the cognitive load and adaptive expertise findings, not a validated instrument.
Monday morning
Take the session you are delivering next week. Run the ten-minute blank-page test: behaviour sentence, waypoints, timings, load-bearing marks, second examples. Whatever you cannot redraw is frozen: deliver it exactly as written, and practise it until it is yours. Whatever you can redraw, mark as improvable, and write the three recovery moves on the back of the run sheet.
Make this the gate in every certification dry run in your train-the-trainer programme: the blank page, not the feedback form, is the evidence of readiness. Where the plan itself must bend to shift patterns and noise, the same discipline applies to the brief, as the plant-floor trainer brief sets out. Building the standard across a whole trainer pool is a design job, the kind our training solutions team scopes with L&D heads before a workshop is booked.
Sources
- https://graphics8.nytimes.com/images/blogs/freakonomics/pdf/DeliberatePractice(PsychologicalReview).pdf
- https://hhs.purdue.edu/skill-learning-and-performance-lab/wp-content/uploads/sites/43/2024/08/macnamara-maitra-2019-the-role-of-deliberate-practice-in-expert-performance-revisiting-ericsson-krampe-tesch-romer.pdf
- https://education.nsw.gov.au/content/dam/main-education/about-us/educational-data/cese/2017-cognitive-load-theory.pdf
- https://en.wikipedia.org/wiki/Adaptive_expertise
