CRM/Sales

Sales AI & Deal Scoring

How interaction analysis, MEDDPICC qualification and the explainable deal score work end to end.

What it is

Three layers work together on every sales opportunity: interaction analysis turns each customer email into a structured, evidence-backed record; methodology skills supply the sales expertise (MEDDPICC, SPIN, Rackham’s Advance, and more) that analysis is read through; and deal scoring combines everything into one explainable, persisted number per opportunity.

8

Weighted scoring components

0–100

Deal score range

±8

Max AI nudge on the score

How interaction analysis works

Each interaction gets exactly one structured, schema-enforced model call at temperature 0, so re-analyzing the same text later can’t silently return a different answer. A single pass extracts sentiment (a five-level label plus a continuous score and a confidence), engagement, typed objections and buying signals, commitments made by either side, SPIN-classified questions, how clear the agreed next step is, competitors mentioned, stakeholders, risk flags and free-form topics.

Guard rails that protect trust

A quoted verbatim must be an exact match for something in the source message or it is dropped before storage — a fabricated quote attached to a score is worse than no quote at all. Analysis failure never blocks storing the underlying email, and a message the model can’t analyze is left honestly blank rather than guessed at with a keyword fallback.

Methodology as skills

Sales expertise is authored as Markdown, not code — each methodology is one directory of skill files, selected contextually so irrelevant methodology never gets loaded: an early discovery call pulls in Advance, MEDDPICC, SPIN and Challenger; a negotiation with active objections swaps in Sandler Objections alongside Advance, MEDDPICC and Challenger; drafting an email leans on Cialdini and Challenger; a pure analysis pass uses Advance and MEDDPICC.

Deal scoring

A new score is computed whenever one of a few things happens:

  1. 1

    A new interaction is analyzed

    The most common trigger — every analyzed email or logged interaction can move the score.
  2. 2

    The opportunity changes stage

    Stage velocity is one of the components, so a stage change recomputes it.
  3. 3

    Someone presses Analyze

    A manual trigger, recorded as such in the score’s history.
  4. 4

    A nightly recompute or a backfill runs

    Keeps scores current even on quiet opportunities, and lets historical data be scored after the fact.

The score itself blends eight weighted components: recency of contact, reciprocity (the balance and pace of back-and-forth — a rep sending eighteen messages to two replies scores very differently from a live ten-and-ten exchange), sentiment trend over time, next-step clarity, MEDDPICC qualification, stage velocity against a benchmark, content engagement, and objection health.

  • A component with no data behind it is skipped, and the remaining weights are renormalized — a dimension the system can’t see is never scored as zero.
  • The model never sets the score directly. It may nudge the deterministic result by at most ±8 points, and only when it gives a reason and cites the specific interactions as evidence — otherwise the nudge is discarded.
  • Every computation is kept, with the exact weights used at the time, so a score from months ago stays explainable even after the weights are retuned.

The current score and its strong / healthy / at_risk / critical band are kept directly on the opportunity so "show me at-risk deals" is a plain, sortable filter rather than something recomputed on every page load.

MEDDPICC qualification

Qualification is assessed pillar by pillar under the MEDDPICC framework and stored as a single structured record per opportunity, with a qualification score promoted alongside it — one of the eight inputs the overall deal score weighs in.

Screenshot pending — Interaction analysis evidence showing an objection with severity and the buyer's exact words
Evidence behind every extracted signal.

What you can see today

Note

In the product today, this shows up as the typed evidence on each interaction (objections, buying signals, commitments, next step) and the opportunity-level AI Analysis card (win probability, sentiment trend, engagement, next best action, suggested follow-up). The underlying explainable deal score and MEDDPICC breakdown are computed and stored on every analyzed opportunity and are reachable through the API for reporting and pipeline review, ahead of a dedicated on-screen breakdown.

Related

New to this? Start with the Sales AI & Deal Scoring guide.