You have a folder of closed-lost deals and a CRM field that says "price" on most of them. You do not believe it, and you are right not to.
Win-loss analysis is the cheapest research a sales org can run and the most commonly skipped, because it needs somebody to read fifteen deals properly rather than argue about one. The prompts here do the reading, working from the buyer's own words rather than the dropdown a rep picked while closing the opportunity. For the mindset that makes a loss useful, how to win more deals by losing better covers that; this page is the analysis.
The stated loss reason is almost never the real one
The field is filled in at the worst moment: after the deal is gone, by the person with the most reason to pick a cause outside their own control. "Price" is the safest answer in every CRM in the world.
Here are the notes and final email thread for one closed-lost opportunity. State the loss reason the record gives, then every alternative explanation the evidence supports, quoting the line behind each. Mark which I have direct buyer evidence for and which are inference.
Run that on one deal before fifteen. The field either survives the pass or collapses in it, and both outcomes change what you do next.
Prompts that separate what the buyer said from what the rep concluded
Most loss narratives are a rep's summary of a buyer's position, two retellings from anything the buyer said out loud. The separation has to be forced; nobody does it voluntarily about their own deal.
From these call notes, build two columns: statements the buyer made in their own words, and conclusions my team drew from them. Name the statement each conclusion rests on, and flag any conclusion with nothing underneath it.
The flagged rows are the output. A loss attributed to budget, resting on one remark about "timing", is an unknown you have treated as a fact. The transcript craft sits in mining call transcripts with AI.
Running the win-loss interview: prompts for questions that get a real answer
A buyer who chose somebody else will talk to you, and will be polite about why. Politeness is the problem: the questions have to make honesty easier than flattery.
I am interviewing a buyer who chose a competitor. Write ten questions that make it easy to tell me something unflattering, ordered so the hardest comes last. Avoid anything answerable with "price" or inviting them to reassure me.
Here is the transcript. List what the buyer told me that I did not already believe, separately from what merely confirmed my view. Then name the one thing they said that I am most likely to dismiss.
That last instruction exists because the dismissed answer is usually the real one. Technique for the live conversation is in Meeting Prep & Discovery.
Finding the pattern across fifteen losses rather than relitigating one
One loss is an anecdote with a strong opinion attached. The pattern appears only at volume, and rarely matches the story the team tells itself.
Here are fifteen closed-lost deals. Group them by the explanation the evidence supports, not the stated loss reason. Report how many sit in each group, which groups rest on thin evidence, and what the stated reasons got wrong across the set.
Across these losses, find the earliest point in each deal where the outcome was already likely, and say what signal was present then. Rank the signals by how often they appeared.
The second prompt is the one that pays. A signal present in eleven of fifteen losses by the second meeting is a qualification change, not a coaching note, and belongs in your pipeline inspection from that day on. The early-warning read is the 7 signs your deal is dead; the reporting layer is RevOps, Data & Analytics Support.
Prompts for the wins, which nobody analyses and should
Every win-loss programme reads the losses. The wins get a celebration and no autopsy, so teams repeat something that worked for a reason nobody wrote down.
Here are ten closed-won deals. Identify what was true here and not in my losses, using only evidence present in both sets. Separate things I did from things about the buyer I got lucky with.
That split is the value: a repeatable move and a favourable buying committee feel identical from inside a win, and only one is a playbook.
Competitor-specific loss patterns, and the battlecard claims they break
Here are the losses a named competitor won. Summarise what the buyer said they did better, in the buyer's language. Then compare that against our battlecard and list every claim the evidence does not support.
For the two most common competitor strengths here, draft the honest discovery question that surfaces the comparison early rather than at the end.
A battlecard built from product marketing's view of a competitor ages badly; one corrected against fifteen buyers who chose them does not. Rebuilding it is covered in how to build a competitive battlecard, and the closing set in Negotiation, Procurement & Closing.
For managers: what to change versus what to accept
A win-loss pass produces a list of causes, most not yours to fix. Sorting them keeps the exercise from becoming a complaint.
Here is the pattern analysis across our last fifteen losses. Sort the causes into three groups: what a rep can change next week, what needs a change in product, pricing or process, and what we accept as the cost of competing here. Name the specific behaviour for every item in the first group.
Third-group items need saying out loud: a team that believes every loss was preventable inspects deals worse, not better. The coaching layer is in Leadership, Coaching & People Management, the deal-level view in Deal Strategy & Stakeholder Management, and the enablement set in AI prompts for sales enablement.
Take one prompt free, with no account, at /free-prompt. Pro is $12 per month or $89 per year, and lifetime access is $149 — See pricing.
Why a model reading your CRM loss-reason field learns nothing
Point a model at the loss-reason column and it returns what the column says, with a confident summary attached. That field is the least reliable data your company holds about why deals are lost, and summarising it faithfully reproduces the error at speed.
It also cannot tell you what was never recorded: the stakeholder who went quiet and was never asked why, the demo that landed badly and generated no note. Missing evidence reads to a model as no cause.
What it does well is read fifteen messy deals without a stake in the answer, hold the buyer's words apart from your reading of them, and name the inconsistency. That is the job a person with a quota cannot do about their own losses.
The rest of the catalogue, by role and by stage, is mapped in AI prompts for sales. Browse the library → or read the full taxonomy.
Frequently asked questions
What are good ChatGPT prompts for win-loss analysis?
The ones that refuse to take the CRM at face value. Four earn their place: a prompt listing every explanation the evidence supports beside the stated loss reason, a two-column prompt splitting the buyer's words from your conclusions, a pattern prompt across fifteen deals grouped by real cause, and a competitor prompt checking battlecard claims against what buyers said.
How do I find the real reason a deal was lost?
Work from the buyer's own language, not the loss-reason field. Pull the call notes and final email thread, have a model list each explanation the evidence supports with the line behind it, and mark which rest on direct buyer evidence. Anything resting on inference is an open question, not a cause.
Can AI run a win-loss interview?
It can write the questions and read the transcript; it should not sit in the chair. Ask it for ten questions ordered so the hardest lands after trust is built, with nothing answerable by "price". Afterwards, ask it to separate what you learned from what you merely had confirmed, and to name the answer you are most likely to dismiss.
How many lost deals do I need before a pattern is real?
Ten to fifteen in the same segment and motion is enough to see something; fewer than five is an anecdote with arithmetic attached. Consistency of context matters more than the count — fifteen losses across three products and two buyer types produce a pattern describing none of them.
What should I never conclude from AI win-loss analysis?
That a cause is absent because no note mentions it. Silent stakeholders, a demo that landed badly, a competitor named once in passing — none reaches the record, and a model reads missing evidence as no evidence. Treat the output as ranked questions to verify, not a verdict.