Write a neutral win-loss interview guide a colleague can run without steering the buyer

Start with the decision and an honest opening

Write down the one decision this round of interviews should inform, the product, segment and buyer role, the outcome you are studying (won, lost, or no decision), and any hypotheses, labelled as such. Sample, cadence and who asks are covered in win-loss analysis without a research team; a study of anything other than a completed buying decision belongs with planning a customer interview round.

A copied opening script that promises anonymity or aggregate-only use creates neither; your arrangements do. Fill these fields with what will happen, then read the result aloud before a buyer hears it:

• Who is asking, and that they did not sell this deal and will not sell during the call.
• Which decision the interviews inform.
• Who will see the notes, and in what form: raw notes, a summary, or both.
• Whether quotations will carry the buyer's name or company, and by what rule they are removed.
• Whether the conversation is recorded, transcribed or processed by an AI tool, and where that material is stored.
• How long the material is kept.
• How the buyer can decline a question, stop the interview or withdraw afterwards, and anything offered in return.

Tell the buyer the purpose and the topics when you invite them. Withholding every question is an editorial choice, and some people need the questions in advance to prepare or to take part at all. Treat a prepared account as a recollection to probe; it is not proof of what happened.

Choose a short core and optional probes

Six core topics are a workable starting point, and the number is an editorial choice: the trigger that started the search, the timeline, the alternatives considered including staying put, the criteria and the people involved, the concrete comparisons or turning points, and anything you did not ask about. The programme lesson already supplies five questions that cover most of these; reuse them, and add wording for the trigger and for a no-decision outcome, where you ask how the deferral happened rather than assuming a competitor won.

Give each topic one open question about the buyer's experience, and keep the follow-ups in a separate column so the interviewer can see the core at a glance. Ask for the buyer's remembered decision date as its own item; if it disagrees with the CRM, compare the two records later and correct neither on the call. Drop any topic that only satisfies curiosity; rehearsal, below, tells you how many fit the time.

Klue advises keeping a core list while staying agile, and the Product Marketing Alliance recommends keeping the standard set consistent so answers can be compared. The two fit together as a stable core with a changing set of follow-ups.

Have Claude flag leading wording

Paste the core questions into a chat, or into the Claude Project you keep for this product if you will repeat the study. Ask for each question that contains a built-in desired answer, an unverified premise, an answer category introduced before the buyer raised it, jargon, two questions in one, or a counterfactual presented as fact, with a rewrite that keeps the original learning objective and a one-line reason. A single chat is enough for a one-off guide; Anthropic's Projects page says free accounts can create up to five Projects (checked October 12, 2026).

Read the rewrites with the same test yourself, because a rewrite can be so neutral it no longer concerns the deal. Two examples, invented for this lesson:

• "Was price why you rejected us?" names a reason before the buyer does. Ask what shaped the final comparison, and probe price when it comes up or as a clearly marked later topic.
• "Did you need Salesforce integration, and how did the other vendors handle it?" is two questions. Ask which systems the product had to connect to, then how each option handled that.

A flag is a reason to think again, and a clean pass is not proof of neutrality.

Probe the buyer's account without overruling it

When the buyer gives an evaluation, ask for the incident or evidence behind it, then what it meant for them. "Their onboarding seemed stronger" becomes a question about when they saw that and what happened. Record the evaluation either way; an opinion without an example is still data, marked as such.

The stop rule covers the other outcomes. If the buyer did not understand, explain the question in plain words rather than repeating it. If they cannot remember, write "cannot recall" and move on. If they decline, skip the question or end the interview. A short answer may be complete, a pause is not evidence of dishonesty, and a hypothetical "yes, that would have changed it" records what the buyer believes now, not what would have happened. Sairu suggests that reasons offered after the first answer may be closer to the real one (in Japanese); that is a reason to keep listening, and no test proves the first answer was only a symptom.

Do not sell or defend the product. Someone who was not on the deal runs the interview, and an objection to your product is recorded, not answered. Some buyers will still soften their account for a vendor's researcher; note where you suspect it and let the analysis weigh it. MaRS warns that a rigid, checklist-style reading of the guide gets in the way of follow-up.

Rehearse the guide and record its version

Have a colleague answer the guide from a scenario invented for the purpose: a no-decision outcome, vague recollection, a misunderstood question, a refusal. Reading questions aloud exposes awkward wording and a roundabout order that looks fine on the page, as Macromill's interview preparation guidance (in Japanese) notes. Claude can play the buyer for a labelled rehearsal; nothing said in a rehearsal is buyer evidence, and none of it reaches the note sheet.

Fix a leading or unusable question as soon as you find it, in rehearsal or after the third interview. Give each question a stable identifier, and give the guide a version number, a date and a short change note, so later comparisons know which wording each answer followed. Clozd advises letting the guide evolve as priorities shift; the version log keeps that evolution comparable. MangoApps' win-loss interview SOP already combines a consent statement, neutral questions, follow-up prompts and note checks; this lesson adds the wording audit, the rehearsal cases and a traceable note sheet.

Keep notes another researcher can verify

Fix the note sheet before the first interview, and fill the same fields every time:

• Interview ID and date, guide version, and interviewer.
• Buyer role, segment, and outcome: won, lost, or no decision.
• The decision date as the buyer remembers it.
• Question ID beside every recorded answer.
• Verbatim passage with a timestamp or note-line locator; paraphrase in a separate field.
• The episode or evidence behind each evaluation, or 'no example given'.
• The buyer's own uncertainty, in their words.
• Not asked, declined, or cannot recall, marked per question.
• Interviewer inference, in its own field and labelled as inference.
• Outstanding follow-up, its owner, and anything promised to the buyer.

When the buyer named several interdependent factors, record them as given, the same day, and do not force them into one primary reason and two secondary ones. If Claude sorts raw notes or a permitted transcript into these fields, keep the original file, check every quotation and assigned field against it, and leave a field marked missing instead of letting it be filled. Claude has only the text you gave it: tone, a pause and anything said off the record are absent unless you wrote them down.

Store the reviewed guide, the opening fields and the blank sheet together, with the current version in Project knowledge or pasted into the next chat; Anthropic's Project management page says material uploaded to a Project is used across that Project's chats. The sheet feeds analysing win-loss interviews, which counts across interviews. This lesson stops at the record.

When a recurring survey fits better

Writing the guide needs no software. What lapses is collection period after period: scheduling, a named interviewer and same-day write-ups. The DIY remedy is the guide, opening fields and note sheet kept together, one owner and a simple tracker.

Steve Win/Loss covers a narrower job. You request generated survey questions, review and edit them, and save the survey; the buyer follows a link into an asynchronous AI chat, which asks follow-up questions to clarify a response. A thin answer is rephrased and asked again, up to three attempts, which differs from the participant-choice rule above. Responses are kept beside the buyer's own words in the analysis, and you can click Generate Summary when new responses arrive; each summary is built from the responses available at that moment. Two tradeoffs apply, as of October 2026. There is no live human interviewer in that flow. Check what the survey says about consent and data use, and how it describes confidentiality, against your own opening fields and how responses are kept and shown. If ongoing collection is the step you cannot sustain, see Steve Win/Loss.