MEDDIC auto-extraction

The job is not to complete MEDDIC. The job is to propose values someone actually said on a call or in email, each tied to a quote, and to leave every other field empty.

A filled panel that nobody can trace to a sentence is worse than a blank one. It tells forecast meetings the deal is qualified when the only event was a model completing a form. Empty stays empty. The quality output is suggested Metrics, Economic buyer, Decision criteria, Decision process, Identified pain, and Champion, each with a cite, or a visible gap.

Conversation intelligence in the Gong, Chorus, and Aviso class already holds the transcript. Salesforce already holds the fields. Extraction sits between them. It does not replace the narrative note and next steps from post-call summary and CRM auto-update. This use case is narrower: the six qualification fields, quoted, reviewable, and owned by the AE.

If managers do not already ask "what is the metric" and "who is the champion" with the CRM open, filling the fields still creates the look of a methodology. Score call coverage with a methodology scorecard if you need a coaching grade.

Run extraction after the recording and related thread are available, on the opportunity the AE already created. Read this call plus emails in the same thread. Do not treat the contact's title in Salesforce as evidence. Do not scrape an org chart and call that a champion.

For each letter, write three things or write nothing: a proposed value in the shape of the CRM field, a verbatim quote with speaker and timestamp or email date, and a status of suggested, AE-confirmed, or empty.

Metrics. Keep quantities the buyer attached to their own problem or success condition. "We're losing about two days a week to rebuilds" is a candidate. "This could be huge" is not. Do not convert a vague pain into a dollar impact they never stated.

Economic buyer. Record the person the account named as having authority to say yes, with the sentence that named them. If the person on the call says "I own the budget," capture that as a claim, not as confirmation they are the economic buyer. If they named someone who was not on the call, keep that person as suggested and unmet.

Decision criteria. Quote the tests they said they will use: must land in the CRM they already run, needs a security review, price has to fit a named range. Do not paste your playbook into their criteria because the call felt aligned.

Decision process. Quote the steps, owners, and any date they volunteered. "Ops recommends, finance reviews, the VP signs" is a process. "We'll know after the next meeting" is not enough. Leave the field empty and show the gap.

Identify pain. Use the problem they described in their words, not persona pain from a battle card. If they mentioned three issues and explored one, extract the explored one. Send dropped threads to discovery gap analyzer.

Champion. Fill this only when someone inside the account is selling on your behalf in language you can quote: they will run the internal meeting, they will take the argument to procurement, they asked for ammunition for their leadership. Friendliness is not a champion. A director who liked the demo and holds a helpful title is still empty. Inventing a champion so the panel looks complete is the fastest way to lie in a forecast call. For who else sits on the account, use buying-committee mapping. That map is not a substitute for a quoted champion.

If the quote is ambiguous, the field stays empty. The AE can type a value. The model should not guess to be helpful.

Take a first discovery with a director of operations. They describe losing "about two days a week" to a manual forecast rebuild, and they say the rebuild is why they took the meeting. They mention that "our VP of Finance has to sign anything that touches reporting." They list one hard criterion: it has to land in the CRM they already run. They sketch a path: operations recommends, finance reviews, the VP signs. They are warm and offer to introduce "a couple of people." They never say they will sell it internally, never ask for a one-pager for leadership, and never name themselves as the person who will carry the deal.

A correct extraction:

• Metrics: about two days per week lost to forecast rebuild. Quote: "we're losing about two days a week to rebuilds."
• Economic buyer: VP of Finance, named, not met. Quote: "our VP of Finance has to sign anything that touches reporting." Status: suggested, unconfirmed.
• Decision criteria: must work in the existing CRM. Quote attached.
• Decision process: ops recommends, finance reviews, VP signs. Quote attached. No dates, so no close-date inference.
• Identified pain: manual forecast rebuild. Quote attached.
• Champion: empty.

The failure is filling Champion with the director because they were friendly and the title looks senior, or filling Economic buyer with that same director because they were the only person on the call. Both writes make the opportunity look later-stage than the conversation. The empty champion is the coaching object: the next meeting needs someone who will sell it, or the AE should stop treating the director as coverage.

This is a pattern, not a measured result. Do not invent a company-wide impact, a close date, or a champion name to complete the form.

Show suggestions on the opportunity in Salesforce, next to the six fields, with the quote visible without hunting the recording. The AE accepts, edits, or rejects each one. Until they confirm, the CRM field does not change.

Never overwrite a value the AE already corrected. If they set Champion to empty after rejecting a bad suggestion, the next transcript of the same friendly director must not put that name back. Lock AE-owned fields, or write new suggestions to a holding field with a diff. Silent write-back trains the team to stop trusting the CRM.

Conversation platforms in the Gong, Chorus, and Aviso class can propose the structured payload. Salesforce remains the system of record. Do not grant the extractor unrestricted update rights on qualification fields. A review queue and a one-click confirm are the control.

If the AE disagrees with the quote, keep the disagreement. "Not a champion; they offered an intro" is more useful than a corrected blank with no history. That trail is what you inspect when a deal that looked MEDDIC-complete dies in committee.

Do not grade reps on gap count. People scored on empty fields will stop recording, or they will coach the buyer into saying the words the form wants. You lose the input, then you lose the truth.

A complete MEDDIC panel is not a qualified opportunity. It is a claim that six sentences existed in the call or email. The buyer can be wrong about who signs. The champion can vanish. The metric can be a wish. Treat extracted fields as evidence for a human qualification decision, including a no.

When the panel is full of gaps, that is a qualify-stage result, not a data-quality incident. Put those gaps on the next call plan. If the same letters stay empty after two conversations, that is input to a disqualification recommender, not a reason to have the model try harder.

Refuse these automations even after AEs trust the quotes:

• Inventing a champion from title, seniority, or tone.
• Writing Economic buyer from "I own this" without treating it as a claim.
• Overwriting an AE correction on the next transcript pass.
• Advancing stage, raising forecast category, or clearing a stage gate because extraction filled the form.
• Scoring the deal qualified because six fields are non-blank.

The quality you are buying is a CRM whose qualification fields are either quoted and AE-confirmed, or honestly empty.