Customer Research Synthesizer

What it does

Uncovers what customers actually think, feel, say and struggle with — so positioning, product and copy rest on reality rather than assumption. It runs in two modes: Mode 1 extracts signal from existing assets (interview/sales-call transcripts, surveys, support tickets, win/loss notes, NPS verbatims); Mode 2 gathers unfiltered customer language from online "digital watering holes" (role-specific subreddits, G2/Capterra, Hacker News, Indie Hackers, 1-3 star app store reviews, SparkToro).

From each asset it extracts Jobs-to-be-Done (functional + emotional + social), pain points (prioritizing those mentioned unprompted and with emotional language), trigger events, desired outcomes (verbatim, not paraphrased), customer vocabulary and alternatives considered. It then clusters themes and scores them by frequency × intensity (how many sources × how strong the emotional language), segments by customer profile, selects 5-10 "money quotes" per theme and flags contradictions.

When to use it

When you need to sharpen messaging/positioning, build evidence-backed personas, understand churn/conversion drivers, find product gaps, surface competitor-vs-you perception, or assemble a VOC quote bank. It fits both analyzing research you already hold and gathering fresh online research. It produces the foundational input for downstream work like copywriting, CRO or churn strategy.

Method / frameworks

• Jobs-to-be-Done (JTBD) — functional/emotional/social job split; what the customer "hires" the product to do.
• Frequency × Intensity scoring — ranks themes by prevalence × emotional strength, not raw count.
• Confidence-level labeling — every insight stamped High / Medium / Low.
• Sample-bias checks — reviewers skew to power users, tickets to problems, Reddit to technical/skeptical; conclusions corrected accordingly.
• Recency window — last 12 months weighted more heavily.
• Minimum viable sample — no persona/messaging conclusion from fewer than 5 independent data points per segment.
• Persona anti-patterns — no averaging across segments, no invented detail, blanks left blank.