A MEDDPICC gap analysis is only as good as the evidence standard behind it. This guide teaches you to judge what actually counts as evidence for each element, to separate what your notes never mention from what your notes disprove, and to catch the moments when an AI assessment sounds certain but isn't. The scorecard is the easy part — the judgment is the work.
When this approach is appropriate
Use it after any discovery or qualification call where you captured real notes — typed, transcribed, or dictated. It works with any capable AI assistant, and it works best inside a project or workspace that already knows your MEDDPICC definitions (set that up once — see the Claude setup guide). It is not a substitute for discovery. If your notes are three lines long, the analysis will tell you that, and that is the correct output.
What counts as evidence, element by element
The most common failure in AI-assisted qualification is grading mentions as evidence. A prospect saying the word "budget" is a mention. Evidence has a name, a number, a date, or a quote attached. Hold each element to this bar:
Metrics. Evidence is a quantified business outcome the prospect stated — a number, a target, a cost they are carrying. "They want to improve efficiency" is not evidence; "they lose roughly six hours per rep per week to manual research" is.
Economic buyer. Evidence is a named person plus a reason to believe they control the spend — title, budget ownership, or the prospect saying so. An org chart guess is not evidence.
Decision criteria. Evidence is criteria the prospect articulated, ideally ranked. Criteria you pitched and they nodded at belong in a lower tier — flag them as "seller-introduced" and validate later.
Decision process. Evidence is steps, owners, and dates. "They usually run a committee" is hearsay until someone on the account says it about this purchase.
Paper process. Evidence is named steps — security review, legal, procurement tooling — with owners or timelines. This is the element reps most often score on assumption.
Identify pain. Evidence is pain in the prospect's own words, tied to a consequence. Pain you inferred from their industry is a hypothesis.
Champion. Evidence is observed behavior — they brought colleagues, shared internal context, moved a meeting for you. Enthusiasm on the call is the weakest form of champion evidence there is.
Competition. Evidence is a named alternative, including "do nothing" and "build internally," in the prospect's words.
Absence of evidence versus evidence of absence
These are different findings and your analysis must keep them apart. If the notes never mention the paper process, that is absence of evidence — an open question for the next call. If the prospect said "I honestly don't know who signs this," that is evidence of absence — a confirmed gap, and a more serious one, because someone inside the account has told you the path is unmapped. When you ask an AI to grade your notes, require it to label every gap as one or the other. A gap map that treats the two the same will send you into your next call asking questions you already have answers to, or worse, assuming answers you never got.
Weighing contradictory notes
Real discovery notes contradict themselves. The VP says budget is approved; twenty minutes later the director says finance hasn't seen it. Do not let an assessment average the contradiction away into a middling score. The rules that serve you better: recency beats earlier statements, the person closer to the money beats the person further from it, and specific beats general. When two credible statements still conflict, the correct grade for that element is "contradicted — resolve next call," not a midpoint. A contradiction surfaced is one of the most valuable outputs a gap analysis can produce, because it hands you your sharpest next question.
Recognizing false confidence in an AI assessment
Language models complete patterns, and a qualification scorecard is a strong pattern. Left unchecked, an assistant will fill empty elements with plausible inference and grade the deal healthier than your notes support. Watch for the tells: an element scored without a quote or paraphrase from your notes to back it; hedged language ("likely," "appears to," "typically") presented inside a confident score; and criteria or process steps that appear in the output but nowhere in what you pasted. The correction is structural — require the assessment to cite the note it relied on for every score, and to mark any element without a citation as ungraded. One instruction like the following, added to your analysis prompt, does most of the work:
Illustrative instruction — add to your gap-analysis prompt
"For each MEDDPICC element: quote the exact line from my notes that supports your score. If no line supports it, score the element NO EVIDENCE and state whether the notes are silent on it or actively disprove it. Never infer evidence I did not provide. List every contradiction between speakers separately with both quotes."
This is one instruction, not the whole method — a complete, reusable gap-analysis prompt lives in the library, and the step-by-step version belongs to the workflow that produces the scorecard itself.
Deciding what to validate on the next call
A gap map with six open items is not a call plan. Choose what to validate using two questions: which gap, if it stays open, most threatens the deal's close date — usually paper process and economic buyer — and which gap can this specific attendee actually close. Do not spend a champion-building conversation interrogating procurement details the person cannot know. Pick at most two validation targets per call, phrase them as questions a buyer would find natural, and leave the rest mapped for the stakeholders who can answer them.
A worked example
An AE pastes notes from a second discovery call with a mid-market prospect. The raw AI pass grades the deal 6 of 8 — healthy. The evidence-standard pass tells a different story: Metrics holds (the ops lead quoted a rework cost), Pain holds, Champion is downgraded because the only evidence is enthusiasm, Decision criteria are flagged seller-introduced, Economic buyer is evidence of absence ("that's above my pay grade" is in the notes), and Paper process is absence of evidence — never discussed. The deal isn't 6 of 8. It is 2 confirmed, 2 weak, 1 confirmed gap, and 1 unknown — and the next call has an obvious agenda: get introduced upward, and ask who has bought software this size before and what that took.
Common failure modes and corrections
Grading mentions as evidence — apply the name-number-date-quote bar. Letting the model fill silence with industry assumptions — require citations, mark uncited elements ungraded. Averaging contradictions — grade them "contradicted" and take the question to the call. Re-validating what you already know — separate the two kinds of gap. Treating the scorecard as the deliverable — the deliverable is the two questions you will ask next, and the scorecard exists to produce them.
Troubleshooting
If every element comes back NO EVIDENCE, your notes are too thin for analysis — capture quotes, not summaries, on the next call. If the assessment keeps inferring despite the instruction, paste smaller sections and grade one element at a time. If scores swing wildly between runs, the notes are ambiguous — that volatility is itself a finding about the deal.