# Five phases. One behaviour. No completion theatre.

### Phase 1 — Discovery

Name the behaviour in break-room language. Example: “Someone pastes tenant data into a chatbot and hopes for the best.” Not “AI governance awareness.”

Agree:

- Who needs to still be able to do this, and by when (retention horizon)
- What “good” looks like on the job
- What you will **not** measure (completion alone is out)
- Privacy basis, who sees individual data, and what managers see (team patterns, not a dossier)

Gate: Is the outcome observable and worth changing?

### Phase 2 — Convert existing training

Take the day, the six-week course, or the in-house materials. Cut to the decisions that change what someone does.

Build short practice items that require:

- **Retrieval** — answer before you see the answer  
- **Feedback** — why that choice was right or wrong  
- **Application** — a slightly different scenario next time  
- **Interleaving** — mix related decision types so people practise picking the right rule  

Where Study Hub themes apply: space the revisits; self-test rather than re-read; after a gap, summarise from memory before opening notes.

Gate: Are items accurate, role-relevant, and usable on the devices people actually have?

### Phase 3 — Spaced retrieval practice

Practice runs across weeks, not as a second dump of content.

Default scheduling idea for MVP (hypothesis, not a universal law): revisit after gaps that grow when recall is effortful and correct, and shorten after errors. A candidate pattern such as days 2, 7, 21, 45, 90 is a starting hypothesis to calibrate with you — not “the science says this exact sequence.”

Sessions are short. Quiet hours and a weekly time cap are agreed up front. Channels sit beside what you already use (email, Teams, printed cards, or web) — not instead of your LMS.

Gate: Can people complete practice without wrecking the working day?

### Phase 4 — Manager observation

Managers use a small rubric for the same behaviour the practice is about. They record what they see at work. They do not need item-by-item learner scores.

This is coaching support, not surveillance. Purpose limits are written down before launch.

Gate: Can managers act on the output without drowning in data?

### Phase 5 — Delayed check and readout

After the agreed gap — not the afternoon of the last session — we check:

- Delayed retention (can they still retrieve without the prompt scaffolding?)
- Transfer (a new scenario, not the memorised one)
- Observed behaviour (manager rubric)
- Business outcome only if you already have a real operational measure

Then a plain readout: what improved, what did not, for whom, under what conditions, and whether to stop, revise, or scale.

Gate: Did active practice beat the comparator you agreed — without harming trust or time?

---