Use case 1
30 stakeholder interviews into a synthesis that holds up
Upload the transcripts. You get themes with attributed quotes — and, separately, the places two models read the same interviews differently. That second list is where the real finding usually is: one model heard consensus, another heard three people saying the same words about different problems.
Try this prompt
Attached are [N] stakeholder interview transcripts from a [type] engagement at [company type]. Produce: (1) five dominant themes, each with two verbatim quotes and the speaker's role; (2) three contradictions between what different levels of the organisation said; (3) anything said once, by one person, that would change the recommendation if true; (4) the question I should have asked and did not. Do not smooth over disagreement between interviewees — surface it. Cite the transcript and line for every quote.
Use case 2
Size a market without the number falling over in the room
Ask for both methods, not one. Top-down from industry reports and bottom-up from customer counts almost never agree, and the gap is the conversation. One model sizes it each way, a third reports where they diverged, and the verification pass tells you which figures have a source behind them and which do not.
Try this prompt
Size the [market] for [geography] for a client deck. Do it twice: top-down from published industry figures, and bottom-up from an estimate of addressable customers × realistic spend. Show both numbers and the assumptions behind each. Where they disagree, say which assumption is doing the work. Mark every figure that you cannot attach a current source to — do not estimate silently.
Use case 3
Have the recommendation attacked before the steering committee does
Debate mode, with two models assigned opposite sides. What comes back is not general pushback but the specific condition your recommendation depends on and nobody stated — the churn rate, the segment, the assumption that the pilot generalises.
Try this prompt
Here is my recommendation and the reasoning behind it: [paste]. Argue the strongest case against it as the client CFO who has seen three consulting engagements fail. Then, separately: what would have to be true for this recommendation to be correct? List those conditions explicitly and say which of them we have evidence for and which we have assumed.
Use case 4
Executive summary that survives a 90-second read
The version a partner will not have to rewrite. Then a second model reads it cold and tells you which sentence a client will stop on — usually a claim you inherited from the analysis three weeks ago and never re-checked.
Try this prompt
Write a one-page executive summary of the attached report. A CEO reads it in 90 seconds: recommendation first, then the three reasons, then the two risks, then the specific ask. No throat-clearing, no 'in today's rapidly evolving landscape'. Then list every factual claim in your summary and mark whether the source document actually supports it.
Use case 5
Turn the findings into the deck, not into chat text
Presentation Studio builds the deck from the analysis you have already done in the project, so the numbers in the slides are the numbers that were checked — rather than re-typed from a chat window at 11pm.
Try this prompt
From the analysis in this project, build a 12-slide findings deck for [audience]: verb-led action titles, one message per slide, the evidence for each finding in the speaker notes rather than on the slide, and a final page listing every assumption the recommendation depends on.
