Use when you have a pile of win and loss call notes and need the pattern rather than the anecdotes.
win-loss-interview-synthesis.md
You are a research analyst reading win and loss interviews. You are counting patterns, not collecting the quotes you like.
Interview notes or transcripts: {{INTERVIEW_NOTES}}
Outcome and value of each deal: {{DEAL_OUTCOMES}}
What we currently tell buyers: {{CURRENT_PITCH}}
Who we lost to, where known: {{COMPETITORS_NAMED}}
Produce:
1. The decision drivers raised unprompted, ranked by how many interviews mention them. Give the count for each and mark any driver appearing in fewer than three interviews as THIN.
2. A table with columns: Driver | Mentions in wins | Mentions in losses | Verbatim quote | What it implies we change.
3. The moment each lost deal turned and the stage it happened at. Group them where the same moment repeats.
4. Every place {{CURRENT_PITCH}} is contradicted by what buyers said, quoted side by side.
5. Losses to {{COMPETITORS_NAMED}} against losses to no decision, and what separates the two groups.
6. The three changes with the most evidence behind them, each with its supporting interview count.
Constraints: rank by frequency across {{INTERVIEW_NOTES}}, not by how strongly one buyer put it. Do not merge a win reason and a loss reason into a single theme. Where {{DEAL_OUTCOMES}} does not say whether a deal closed, leave it out of the counts and list it separately. No em dashes.
Replace each placeholder with your own detail. The more specific you are, the less the model invents.
When should I use this rather than reading the call notes myself?
Once there are enough interviews that memory starts picking favourites. Reading them yourself surfaces the vivid quotes. This counts drivers across the whole set and ranks by how many interviews mention each, marking anything under three interviews as THIN, which produces a different list from the one you would recall.
What do I need alongside the interview notes?
The notes or transcripts, outcomes saying whether each deal closed and its value, what you currently tell buyers, and who you lost to where known. Include the losses you think you already understand, since those notes hold the surprises. Deals with no recorded outcome are listed separately rather than counted.
What comes back, and what should I act on first?
Ranked unprompted drivers with counts, a wins against losses table with verbatim quotes, the moment each lost deal turned grouped where it repeats, every place buyers contradict your current pitch quoted side by side, competitor losses against no decision, and three evidenced changes. The contradiction section is the fastest thing to act on.
What is the mistake that ruins the counts?
Promoting one eloquent quote to a finding and quietly dropping the THIN marker. Frequency is the whole method. The other is merging a win reason and a loss reason into one theme, which the prompt forbids, because they usually turn out to be different buyers rather than a single factor.