Pain-Qualified Segments from Calls

The play in six steps

1. 01

   Step 1: Gather the calls and the customers

   The best segments come from what was true for real buyers at the moment they decided to talk or buy.

   • Pull sales call transcripts from Fireflies. Calls with the buyers themselves outrank internal kickoffs, which only capture what the team believes.
   • Export closed-won and closed-lost deals from HubSpot with close dates and reasons. This is the backtest set.
   • Tag every line of context with where it came from. Anything unsourced is an assumption.
2. 02

   Step 2: Mine the buying situations

   Each call is read for the situation, not the industry, and for the words the buyer used to describe it.

   • Claude extracts, per call: the situation, why now, what they already tried, and verbatim pain lines with the speaker.
   • Record the outside-in fingerprint: what a stranger could have seen beforehand, such as a job post, a filing, a tool or a hire.
   • Batch large call sets across parallel runs with one shared schema.
   • Count every situation. Came up in 12 of 100 calls is a finding; came up a lot is not.
   • Build a language bank of the exact phrases buyers used for the pain.
3. 03

   Step 3: Write each segment as a situation and a tension

   One signal is noise that every vendor fires on. Two facts pulling against each other is the pain.

   • Situation: one sentence about something that happened to them, not something they are.
   • Tension: the thing they are doing, pulling against the thing that makes it hurt.
     • Example: adding sales reps, with nobody running the system they will work in.
   • Amplifier: the deadline, target or season that makes it hurt now.
   • Existential data point: the number that, past a threshold, costs someone the quarter. No such number means the segment is incremental, and the doc says so.
   • Kill segments that are vague, ungrounded, a single signal, or a pain the offer does not fix.
4. 04

   Step 4: Write the membership test field by field

   A segment is only real if the next person can rebuild it from the spec alone.

   • Every condition gets a source, a field, an operator and a threshold.
   • Mark each one public, proxy or not public.
     • Public: the field states it. Use it to filter and in copy.
     • Proxy: the field suggests it. Use it to score and to ask, never state it as fact.
     • Not public: drop it from the test and turn it into a question for the call.
   • A segment whose core tension rests on a not-public condition is a discovery question, not a segment.
5. 05

   Step 5: Measure before building

   Three checks turn a hypothesis into a measured segment.

   • Count it with free or near-free counts, such as an AI Ark preview, and record the query and the date.
   • Open five rows the test says are in and check each one is really in the situation. If two of five fail, fix the test before counting further.
   • Backtest: what share of recent closed-won accounts were in the situation in the 90 days before they bought, against a random sample of the TAM.
     • The ratio is the lift. Low lift means a firmographic in disguise.
   • Label each segment measured, partly measured or hypothesis.
6. 06

   Step 6: Build the list and the opener

   The measured segments become a refreshable list and a message written in the buyers' own words.

   • Discover rows on one filter, then score on the rest. Stacking discovery filters collapses the list.
   • Build in Clay with AI Ark for companies and people and Exa for evidence that needs the web. Allowed answers include unknown, and unknown never counts as a pass.
   • Set a refresh cadence from how long a row stays true.
   • Open on the situation, using vocabulary from the language bank, say what you saw and where, and ask if the read is right.
   • Send by email through Instantly and on LinkedIn through lemlist. Never quote one buyer to another.