If you are weighing up an interactive case platform — or wondering whether to build something yourself — the question underneath all the others is what a single learner actually costs to serve. It is a hard question to answer from the outside, because the pricing pages in this category tend to quote a per-student figure and stop there.
So here is the inside view, taken from instrumenting a full case experience and reading the bill afterwards.
One simulation, measured
The unit is a six-turn role-play: a learner talks to an AI character who holds a position and does not simply agree with them, followed by an examiner that reads the whole transcript and writes a debrief judged against the case's own criteria.
| Component | Volume | Cost |
|---|---|---|
| Language model | 7 calls · 33,352 in / 4,471 out | $0.167 |
| Speech | 7 clips · 2,235 characters | $0.112 |
| One simulation + debrief | $0.279 |
A whole case experience is several of these plus the connective tissue — an introduction, a narrated film, the guide that sits beside the learner — and lands around $1.20 to $1.40 per learner for the interactive parts.
The surprise: speech costs more than thinking
Most people assume the language model is the expensive part. It is not. Across a full run, voice was roughly 60% of the bill, and the model about 40%.
That inverts the usual intuition, and it matters because the two behave completely differently. Model calls are inherently unrepeatable — a live conversation has to be generated fresh every time, because the learner says something new. Narration is the opposite: a scripted film says exactly the same words to every learner, forever.
The number that made this obvious. Our worked case opens with an animated film — 81 spoken lines, 8,696 characters. Re-synthesised on every play, that film cost $0.435 every single time someone watched it. For a film that never changes.
The fix is boring and total
Record the narration once, publish the audio as static files, and look it up at play time. A replay then costs nothing — not less, nothing. We verified it the only way worth trusting: play the whole film through the real player, confirm every line came from cache, and check the spend record did not move by a cent.
That single change took the per-learner figure from about $1.75 to roughly $1.25, and it gets better with every replay rather than worse.
What this means if you are buying
- A per-seat price in the £15–20 range is mostly margin, not cost. That is not automatically wrong — someone has to fund the faculty co-authoring and the production — but it is worth knowing what you are paying for.
- Ask what a replay costs. If a platform charges per attempt, ask why. Letting a learner run a conversation again is where a lot of the learning happens, and it should not be rationed for billing reasons.
- Ask whether narration is generated live. If it is, someone is paying for the same audio repeatedly, and eventually that someone is you.
- Building it yourself is cheaper than it looks, and harder than it looks. The API bill is genuinely small. The work is in everything around it: scoring that survives a determined student, a debrief worth reading, and not falling over in week three of term.
Why publish this
Partly because the alternative is a category where everyone quotes a price and nobody shows their working, which does not help anyone decide anything. And partly because the numbers are not a secret — they are public API rates and arithmetic. Anyone can reach them with an afternoon and a spreadsheet.
What is hard is not the cost. It is making an hour of it worth a learner's time.
See where the money goes
Averholt Medical is a complete worked case — a film, four simulations in different formats, a guide, and an examiner's debrief at the end. It is free to play, and it is the exact case these figures were measured on.
Play the worked case →