Sales: discovery summariser that keeps the human

A sales VP we worked with had a problem with discovery summaries. His reps wrote them. They were inconsistent — sometimes a paragraph, sometimes three pages, sometimes nothing. The downstream effects were predictable: account executives walked into demos cold, sales engineers showed up unbriefed, leadership reviewed pipeline with stale context.

The discovery summariser AI employee solves the consistency problem without losing what makes a discovery call useful — the texture, the human stuff, the things between the lines that experienced reps read by ear.

The shape of the role

Title. Sales Operations AI — Discovery Specialist.

Mission. Convert every discovery-call recording or transcript into a structured summary in the team's chosen framework (MEDDIC, BANT, custom), within 90 seconds of call end.

Outcomes. Summary completion rate, summary accuracy (vs. rep memory), AE/SE prep-time, deal-velocity for accounts with full summaries vs. without.

Reports to. Head of Sales Operations.

Tools. Call transcription, CRM read/write, sales framework templates, deal-history database.

Boundaries. Summarises and posts to CRM. Doesn't make qualification calls. Doesn't author follow-ups.

What "doesn't flatten" means

The temptation with AI summaries is to extract a checklist of facts. Discovery calls aren't checklists. The real value is in:

What the prospect emphasised (signal of priority).
What the prospect avoided (signal of fear or political constraint).
The exact phrases the prospect used (for use in follow-up writing).
The unspoken concerns that experienced reps notice (timing, decision-maker absence, competing initiatives).

A working discovery summariser captures these. The summary has structured fields (the framework's checklist) and a free-form section that quotes specific prospect statements with timestamps.

The free-form section is the difference between a summary an AE can use to prep and a summary that's just data entry.

The MEDDIC pattern

For teams using MEDDIC, the agent's output:

Metrics. What the prospect said about their numbers — current, target, what they'd pay to move from one to the other. Quoted.
Economic Buyer. Who has signing authority. If the rep didn't ask, the agent flags it.
Decision Criteria. What the prospect explicitly listed as criteria. Quoted.
Decision Process. Steps and timeline as discussed. If not discussed, flagged.
Identify Pain. What hurts. Quoted directly.
Champion. Who in the prospect's org is helping the deal move. If unclear, flagged.

Plus a free-form "tone and texture" section: how the call went, what was emphasised, what was avoided, what to follow up on.

What this saves

A typical AE walks into a demo with a 15-minute prep window. Half of that is reading whatever notes the SDR wrote. With the discovery summariser:

The AE reads a complete summary in 3 minutes.
The remaining 12 minutes go to actual demo prep.
The demo is more relevant, hits the prospect's stated pain, uses their own language.
Demo-to-next-step conversion improves measurably.

The compounding gain shows up in deal velocity. Better-prepared demos lead to faster deals. Faster deals lead to more closed business per quarter.

CRM hygiene as a side-effect

The agent writes to CRM. Always. Every discovery call ends with a CRM record that includes the structured fields, the free-form section, and timestamps to the recording.

Before the agent: CRM completeness was 40-60% on key fields. Reps did the typing eventually, or didn't.

After: CRM completeness is 95%+ within 90 seconds of call end. This sounds like a small thing. It compounds. Pipeline reviews improve. Forecast accuracy improves. Marketing's targeting based on CRM data improves.

The reviewer loop

The rep reviews the summary before the system marks it final. Edits feed the eval set. Common edits the team will see:

"The prospect said X, but the context was Y" — the agent missed nuance.
"The framework field was misread" — the agent classified incorrectly.
"This was a key quote we'll use in follow-up" — the agent didn't surface it.

Each pattern improves the agent over a quarter. After two quarters, edits drop sharply.

What we won't ship

Auto-sending follow-ups. That's the rep's voice and the rep's relationship.

Auto-qualifying leads. Qualification is the rep's call.

Sharing the summary outside the deal team. Discovery content is sensitive; access controls matter.

Recording without consent. State and country laws vary; the agent's deployment respects them.

How to start

One sales team, one product line. Run the agent on every discovery call for one month. Compare summaries to what reps would have written. Tune the eval set. Then expand.

Most teams roll this out across the entire org day one. The ones that ship pick one team, dial it in, then expand.

Close

The discovery summariser is an AI employee whose job is to make every other person on the deal team better-prepared. Not by replacing the rep's ear — by capturing what the rep heard, in a form everyone else can use. The texture of the call survives. The data entry doesn't.