Verified against ChatGPT · 2026-08-11
Synthesize win-loss interview notes into patterns a sales team will actually act on
Pulls recurring themes out of a batch of win-loss interview transcripts or notes, separating what actually drove the decision from surface-level comments, so the output becomes a real input to messaging and product priorities.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are synthesizing a batch of win-loss interview notes into patterns the sales and product teams can act on. Your job is to find what actually drove each decision, not just collect quotable lines.
WIN-LOSS INTERVIEW NOTES
Interview 1 (loss to Competitor X): 'Pricing was close, but their implementation team had a dedicated onboarding specialist and ours would've been self-serve.' Interview 2 (win): 'We picked you because the demo showed the reporting feature our finance team specifically needed.'
DEAL OUTCOMES (win or loss, and to whom if a loss)
Interview 1: Loss, to Competitor X. Interview 2: Win. Interview 3: Loss, to internal build.
TIME WINDOW AND SEGMENT
Last two quarters, mid-market deals over $50K ACV.
HYPOTHESIS YOU WANT TESTED, IF ANY
We suspect we're losing mid-market deals on implementation support, not price.
PHASE 1 — EXTRACT
For each interview, separate what the buyer said explicitly caused the decision from what they said as polite context around it. Buyers often soften a real reason ('pricing was a factor, but honestly the team just felt more comfortable with the other vendor's onboarding') — extract the second half as the real driver, not the first half as the official reason.
PHASE 2 — GROUP
Group extracted drivers into recurring themes across the batch, and for each theme state how many of the interviews support it and whether it appears more in wins or losses. Do not force a theme that only shows up once into a 'pattern' — a single data point is an anecdote, not a pattern, and should be reported separately as one.
PHASE 3 — TEST THE HYPOTHESIS
If a hypothesis was given, state plainly whether this batch of interviews supports it, contradicts it, or is inconclusive, and why — do not let a stated hypothesis bias which quotes get emphasized in your synthesis.
OUTPUT FORMAT
1. Top three recurring themes, each with supporting interview count and win/loss skew.
2. Any single-interview anecdotes worth flagging separately, labeled as such.
3. Hypothesis verdict, if one was given.
4. One recommendation each for sales messaging and for product, each tied directly to a named theme above.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The two-half extraction rule in Phase 1 targets a specific, well-documented behavior in win-loss interviews: buyers routinely lead with a socially comfortable reason (price, timing) and only reveal the real driver as a secondary aside, because naming the real reason directly can feel like a criticism of a relationship with the vendor they're declining. A model asked to just 'summarize the reason for the loss' without this instruction will typically extract the first, more prominent sentence, because that's structurally where a stated reason usually lives in a sentence — explicitly directing it to weigh the softer aside as the real signal corrects for this positional bias. Requiring an explicit count and win/loss skew per theme, and refusing to promote a single mention into a 'pattern,' addresses the tendency of language models to smooth interview data into confident-sounding generalizations regardless of how many data points actually support them — without this constraint, one vivid quote can end up driving an entire recommendation, which is exactly how bad product and messaging decisions get made off thin qualitative data. The hypothesis-testing phase is deliberately placed after the theme extraction rather than before, and the instruction to not let it bias quote emphasis exists because stating a hypothesis upfront is a well-known way to induce confirmation bias in an LLM's synthesis — it will otherwise selectively surface supporting quotes once it 'knows' what answer is expected, which defeats the actual purpose of testing a hypothesis against evidence rather than confirming it.
What you get back
Top theme: implementation support depth, supported by 4 of 6 interviews, skewing heavily toward losses (3 losses, 1 win mentioning it positively). Anecdote (single mention, not a pattern): one buyer cited a specific integration with a niche tool as decisive — worth tracking, not yet actionable. Hypothesis verdict: supported — the batch backs the suspicion that implementation support, not price, is the recurring loss driver in mid-market deals. Recommendation for sales: lead mid-market pitches with the onboarding model explicitly, before pricing comes up.
Verified against
ChatGPT GPT-5.1 · 2026-08-11
Changelog
- 2026-08-11 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — Google Ads management, built and maintained for you — that's Scult's day job.
EXPLORE GOOGLE ADS MANAGEMENT
