Skip to content
← Selected workCASE STUDY / 04

VOICE OF CUSTOMER

Hear the buyer.
Then write the page.

Turn recorded conversations into evidence your team can use. Keep the buyer’s words, separate interpretation, and bring their priorities into your messaging and qualification rules.

Discipline
Buyer research
Context
US executive search firm
Explore the build ↓
THE SYSTEM IN ACTIONIllustrative demonstration

ILLUSTRATIVE BUYER LANGUAGE

Push
    Pull
      Anxiety
        Habit
          638 excerpts in. 367 buyer quotes kept and coded. 101 seller-inferred, kept separately. 56 removed, each with a reason.

          From a sales conversation to a traceable piece of evidence

          The situation

          The calls were recorded. Nothing was done with them.

          Sales calls were recorded, and nothing was done with them. Messaging on the site and in outreach was built on what the team believed buyers cared about, and the team's belief was mostly right and entirely unverified. The calls held the buyers' own words for what was broken, what they wanted, what they were afraid of, and what had convinced them, and nobody had gone in to get them.

          The task: extract what buyers said, in their words, from every recorded call. Keep only what a buyer actually said, code it so sales can find the right evidence at the right moment, and use it to rebuild the ICP on evidence instead of assumption.

          What was built

          Six steps from a recording to a fit weight.

          Extraction, two tiers, removal with a reason for every row, coding on five frameworks, a playbook for sales, and an ICP the scoring engine runs on. Not one quote appears on this page.

          1

          Extraction

          A pipeline that reads call transcripts and pulls candidate quotes.

          callsrecorded, transcribedexcerpts638 candidate quotes outtiers109 curated, 529 filtered
          2

          Two-tier filtering

          A human-curated set treated as trusted, and a larger AI-filtered set that gets re-filtered rather than trusted. The two overlapped almost completely, so duplicates were collapsed with the curated version winning.

          Tier A109, trustedTier B529, re-filteredoverlap109 collapsed. A wins.
          3

          Removal with reasons

          Anything that wasn't a buyer speaking came out, and every removal got a disposition: third-person write-ups, vendor marketing language, filler and logistics, ambiguous speaker. Seller inferences, things the firm's own people said about buyers, went into a separate bucket rather than the bin, because they're useful and they're not the same thing as a buyer's words.

          third-person8 outvendor copy14 outfiller, logistics30 outambiguous speaker4 outseller-inferred101, kept apart
          View the implementation12 lines +

          The audit, as it was written down. Every row in is accounted for.

          # The source audit, 21 April 2026. Every row in is accounted for.
          Total rows in (both files)                 638
            tier_a_manual, human-curated              109
            tier_b_ai, AI-filtered                    529
          Cross-file duplicates (manual reappears in AI)   109
          Within-AI-file duplicates                     5
          Removed, third-person write-ups               8
          Removed, vendor marketing copy               14
          Removed, filler and logistics                30
          Removed, ambiguous speaker                    4
          Seller-inferred, kept separately            101
          Unique retained buyer-voice quotes          367
          4

          Coding

          Five established frameworks, applied in sequence: hierarchical voice-of-customer coding into primary and secondary themes, six-phase thematic analysis, the jobs-to-be-done forces of progress (push, pull, anxiety, habit), category entry points for triggers, and evaluation criteria in the customer's own words. Each quote carries its force, its funnel moment, its content type, its themes, its emotional register, a confidence grade, and a quotability score.

          Pushwhat's brokenPullthe progress they wantAnxietyfears, past failuresHabitwhat keeps them put
          View the implementation13 lines +

          What every quote carries. The quotes themselves are not on this page and never will be.

          # One row per quote in the bank. Every quote carries all of these.
          quote_id             stable id
          verbatim_quote       the buyer's words, unedited. Never leaves the firm.
          source_file          tier_a_manual or tier_b_ai
          evidence_confidence  how sure we are it is a buyer speaking
          forces_tags          PUSH, PULL, ANXIETY, HABIT, in any combination
          funnel_moment        where in the decision it was said
          content_type         CRITERION, TRIGGER, ATTRIBUTION, COMPARISON, EMOTION
          primary_theme        one of 23
          secondary_theme      one of 153, under its primary
          emotional_register   how it was said
          likely_moments       where sales could use it
          quotability_score    how well it stands on its own
          5

          The playbook

          The bank organised so sales can locate evidence by moment: what's broken, what they want, what they fear, what keeps them where they are, what triggered the search, how they evaluate, and what they say worked afterwards. Plus a collaborative section for sales to log what landed and what fell flat.

          5.1Push, what's broken5.2Pull, the progress they want5.3Anxiety, fears and past failures6Triggers, by category entry point9For sales: what landed, what fell flat
          6

          The ICP

          Rebuilt from the bank as an evidence-graded classification. Those definitions are the title, revenue, industry and geography weights in the scoring engine, which is how one piece of work ends up running inside another.

          bank367 quotes, 23 themesICPevidence-graded tiersenginefit weights, out of 30
          What it produced

          Every row in, accounted for.

          Snapshot: 21 April 2026. These are corpus sizes, not business volumes, so the counts stay.

          Buyer-voice quotes kept and coded367
          Seller-inferred, kept separately101
          Duplicates collapsed114
          Removed, each with a reason56

          Of 638 rows in. 23 primary themes, 153 secondary. Five named coding frameworks applied in sequence.

          The handoff

          The playbook to sales. The ICP into the engine.

          Sales gets the bank organised by moment and a section to write back into. The ICP definitions become the title, revenue, industry and geography weights in the scoring engine. Stack: Fathom, Claude Code, Excel, Markdown. Built with Amr: the coding and the bank were the two of us; the extraction pipeline, the ICP and everything downstream is mine.

          Any firm that records its calls and hasn't read them has this sitting there. Getting it out is a pipeline. Deciding what it means is the work.

          YOUR WEBSITE / A FRESH LOOK

          You’ve seen the work.
          Let’s look at your website.

          Get a practical review of the journey your buyers take—and the useful context your team could be missing.

          Get my free teardown ↗See exactly what’s included
          A FOCUSED REPORT, WITHIN TWO DAYS
          01

          What’s getting in the waySpecific examples from your public website.

          02

          What it means for the buyerWhy each issue is worth your attention.

          03

          What I would fix firstA clear order of priorities for you or your team.

          Free. No account access. No call required.