Win loss analysis for b2b deals, asked properly

Price is the plurality reason in almost every CRM and it is an artefact of a dropdown. The measured effects of question format are large enough to flip your conclusion.

Most win loss analysis for b2b deals fails in the same place, and it is not the part anyone worries about. Teams worry about getting buyers to agree to the call. The failure is earlier and quieter: the questions asked on that call are written in a way that guarantees a useless answer, and nobody can tell afterwards, because a confident answer and an accurate answer read identically in a transcript.

This is not a soft problem. Survey methodologists have measured it for decades, and the effects are large enough to flip a conclusion. Borrow their rules rather than inventing your own.

The reason your CRM loss codes are not a dataset

Open your CRM and count how many closed lost deals carry the reason "price". In most pipelines it is the plurality, sometimes the majority. That number is not a finding. It is an artefact of a dropdown.

Two things produce it. The rep picks the reason that requires least explanation to their manager, and "price" implies nobody made a mistake. And the buyer reaches for the most socially comfortable answer available, which is also price. Price is the polite no: it costs the buyer nothing and it ends the conversation.

So the first decision in a win loss programme is to treat the CRM field as a log of what reps typed, not as evidence about why deals were lost. Those are different objects. Keep the field, because operationally you need it. Do not analyse it.

What survey methodology says about asking

The Pew Research Center publishes its questionnaire design method openly, and three of its findings bear directly on a win loss call.

The first is the gap between open and closed questions. Pew describes this as one of the most significant decisions that can affect how people answer questions, and gives a measured example: in 2008, 58% of respondents chose the economy when it was offered as an option, against 35% who volunteered it unprompted. The same subject, the same population, a 23 point difference produced entirely by whether the answer was on the list.

Read that back as a win loss problem. If you ask "was it price, timing, features or implementation risk", you will get a distribution across those four, and it will look like data. If you ask "walk me through how you decided", you get a smaller and messier set of reasons, and it is the real one.

The second is acquiescence. Pew notes that in agree-disagree formats, less educated and less informed respondents have a greater tendency to agree with such statements, and that the effect is stronger when an interviewer is present. A win loss call is an interviewer, present, usually on video. Every question shaped as "would you say the pricing felt high" is collecting agreement, not information. Pew's recommended fix is direct: offer a choice between alternative statements instead.

The third is order. Pew warns that earlier questions can unintentionally provide context for the questions that follow, and documents a case where asking about one subject first moved support for a related position from 37% to 45%. It also notes that open-ended questions placed after related closed-ended ones tend to pull in whatever the closed questions raised. That single sentence invalidates the standard interview script, which runs through a rating grid and then asks "anything else?" at the end. By then you have taught the buyer your vocabulary and they will give it back to you.

A script that survives those three problems

The order matters more than the wording.

• Open, before anything else. "Take me back to when you first decided to look at this. What was happening?" No options, no grid, nothing that names a category. Let it run long.
• Chronology, not causes. Ask what happened next, and next. People reconstruct sequences more reliably than they explain motives. The decision usually reveals itself as a point in the sequence rather than a reason they can state.
• The alternative, in their words. "What did the other option let you do that ours did not?" This is a comparison question that does not supply the dimension of comparison.
• Who else was in it. Names and roles. Most losses that get recorded as product gaps are internal decisions you never saw.
• Only now, the closed questions. If you need ratings for comparability across deals, ask them last, after the open answer is already recorded and cannot be contaminated.

One structural rule is worth more than the whole script: the person who owned the deal does not run the interview. Not because they will lie, but because the buyer will be kind to them. A buyer will tell a neutral third party that the demo was confusing and will tell the rep who gave it that everything was great and the timing was wrong.

Why ten deals is the honest threshold

A single loss interview is a story. It helps the rep who lost it and almost nobody else, because you cannot tell which parts are the market and which parts are that one buyer's budget cycle.

Patterns start appearing around ten, and the reason is mechanical rather than statistical. Ten interviews is roughly where the same sentence starts arriving from unrelated buyers in different words. That repetition is the signal. One buyer saying implementation worried them is a data point. Four buyers independently describing a gap between what the demo showed and what the first month would involve is a finding, and it is actionable in a way no individual interview is.

Be honest about what this is not. Ten is not a sample size in any inferential sense, and you cannot calculate anything from it. Nobody selling win loss services will tell you the thresholds they quote are from a methodology study, because they are not. Treat ten as the point where you stop drawing conclusions from anecdote, not the point where you start having statistics.

Where the interviews have to live

The practical thing that kills most win loss programmes after two quarters is not effort. It is that the output is eleven documents in a shared drive, and the pattern across them exists only in the head of whoever did the interviews.

What makes it compound is structure: one row per deal, the open answer stored verbatim and kept separate from any coding you apply, so that when your categories turn out wrong you can recode without having destroyed the original. The coding will be wrong at first, and that is only recoverable if the raw text survived.

If that currently lives in a spreadsheet that three people edit, the failure mode is the one described in rivl's note on moving from spreadsheets to a database without losing the history, and win loss data is a particularly bad thing to lose history on, because the whole value is in comparing this quarter's answers to last year's.

What to do with a finding once you have one

Findings from win loss fall into three buckets and only one of them is yours to fix.

Some are qualification problems, which means the deal should never have reached late stage. Those are the cheapest to fix and the most common, and they belong in your ideal customer profile tests rather than in the product backlog. If three lost deals share a characteristic you could have seen on the first call, that is a disqualification rule, and the cost of not having it is the whole pipeline's worth of time. The case for acting early is in when to disqualify a prospect.

Some are process problems: the proposal arrived after the decision, the wrong people were in the room, the follow up stopped. These are the ones that respond fastest, and they usually trace to something concrete in the cadence after first contact or to how the proposal was structured.

And some are genuine product or price gaps, where the buyer was right to go elsewhere. Those are worth knowing and mostly not worth acting on immediately, because one or two buyers wanting something is not a roadmap. Record them and wait for the fourth.

The discipline is resisting the third bucket's gravity. Product gaps are the most satisfying finding, because they absolve the sales process entirely. They are also the rarest of the three.

The honest limits

We cannot tell you what proportion of your losses are really about price. The figures circulating on that question come from vendors who sell win loss interviews, measured on their own client base, and several of the most repeated numbers trace back to a single dataset being cited as though it were a literature. The direction of the bias is well established. The magnitude in your pipeline is not knowable from anyone else's data.

We also cannot promise that buyers will tell you the truth in an interview, even a well constructed one. Pew's own point about social desirability is that people have a natural tendency to want to be accepted and liked, and that tendency does not switch off because you asked a better question. Good method reduces the distortion; it does not remove it.

What is solid enough to act on: the question format changes the answer by a measurable and large amount, interviewer presence pushes respondents toward agreement, earlier questions contaminate later ones, and your CRM loss field is a record of rep behaviour. Those four are enough to redesign a win loss programme around, and all four are documented rather than asserted.