How to Run: Win-Loss Analysis Survey

Introduction

A win-loss analysis survey asks buyers on recently decided deals why they chose you, a competitor, or no vendor at all. Run it by coding three outcomes, analyzing deals instead of respondents, pairing a stated reason with a comparative rating, and reporting every share with an interval.

Last updated October 2026.

How to Run a Win-Loss Analysis Survey Step by Step

A win-loss survey is only as good as four decisions made before the first send:

Code three outcomes. Every decided opportunity is won, lost to a named alternative, or lost to no decision. "No decision" is its own outcome, because folded into "lost" it hides what’s often the largest loss cell.
Analyze deals, not people. A buying group of five on one opportunity shouldn’t outvote five single-contact deals.
Pair a stated reason with a comparative rating. Ask for one primary reason, then ask everyone to rate your offer against the chosen alternative on the same factors. The stated reason tells you what buyers say, and the won-versus-lost comparison tells you where outcomes actually differ.
Report every share with an interval. At 20 to 30 deal-level responses, a 50 percent reason share typically carries a 95 percent interval running from about 30 to 70 percent.

Then work the seven steps in order:

Define the three outcome codes and a deal-size threshold.
Pull every buying-group contact from each decided opportunity.
Send a short survey from your own domain on a fixed cadence.
Ask one primary reason and a comparative factor rating.
Collapse responses to one record per deal.
Report each reason share with a Wilson interval.
Check respondents against nonrespondents on customer relationship management (CRM) fields.

What a Win-Loss Analysis Survey Is

A win-loss analysis survey is a structured questionnaire sent to buyers after a sales opportunity closes, asking what drove the decision. It’s the quantitative layer of win-loss analysis, the broader practice of learning from decided deals.

The survey measures incidence: how many buyers named each reason, in which outcome, against which competitor. Interviews measure the story behind a single deal. Both belong to the same program, and a common design runs the survey across every decided deal and interviews a subset.

The Three Outcomes a Win-Loss Survey Codes

A win-loss survey typically codes every decided opportunity into one of three outcomes, and the coding decides every number that follows:

Won
Lost to a named alternative
Lost to no decision

A deal lost to a named alternative includes a competitor, an in-house build, or an existing tool the buyer kept. A deal lost to no decision is one where the buyer ended the evaluation without choosing anything, commonly after a priority shift or a failed internal consensus.

Deals still open are excluded, and so are deals the buyer reports as still undecided even when the CRM shows them closed. That rule keeps the denominator honest, because a CRM stage change is often a sales convention rather than a buyer decision.

How Win-Loss Differs from Its Neighbors

Win-loss explains outcomes in your own pipeline. Churn research explains why customers leave after buying, and a brand tracking survey describes your position in the whole market, including deals you never saw.

This guide owns the pipeline and hands the other three off at the points where a win-loss survey stops being the right tool.

When to Run a Win-Loss Analysis Survey

The clearest case is a sales motion where the CRM "closed-lost reason" field is the only explanation anyone has for losses. That field is typically typed by the rep who lost the deal, and the next sections explain why it’s the account you can least afford to rely on alone.

A win-loss survey fits when deal volume is high enough to count. A team closing a few dozen opportunities a quarter can pool windows and report patterns, while a team closing three enterprise deals a year is better served by interviews alone.

It fits after a competitive shift. A new entrant, a pricing change or a repositioning creates a before-and-after that a fixed instrument can measure.

It fits when product, marketing and sales disagree about why deals are lost. A buyer-reported count of reasons, by outcome and competitor, gives product, marketing and sales one account to dispute instead of three internal ones.

And it fits as the sampling frame for interviews. The survey’s follow-up permission question produces a consenting, outcome-stratified list of buyers willing to talk.

A win-loss survey doesn’t fit as a one-off. The value comes from a fixed instrument read across windows, and a single window at small n rarely supports a decision on its own.

When Not to Run a Win-Loss Survey

Seven situations, each naming the instrument or fix that’s actually right. The first two are the ones most likely to change what you do.

You need the story of the deal. A survey returns incidence, not sequence. It can’t reconstruct the internal politics, the moment the deal turned, or the reason behind the reason a buyer gives.

The better instrument is a buyer interview run by someone other than the account’s rep. Interviews, whether human-moderated or AI-moderated, are qualitative instruments that answer a different question from fewer people, and they sit outside the survey category.

You need to know whether fixing a reason would change the outcome. A won-versus-lost gap on a factor is an association inside a self-selected sample, not a cause.

The better instrument is an experiment on the specific lever: Gabor-Granger analysis or a conjoint analysis for price, a message testing survey for positioning, and a product concept test for a product gap.

You need your position in deals you never saw. Win-loss sees only your pipeline, so buyers who never shortlisted you’re absent by construction. The better instrument is a brand tracking survey or a category-wide study.

You need the size of the market you lose. A win-loss survey counts reasons inside deals that reached your pipeline. It can’t say how many buyers in the category reject your product before they ever talk to sales, or why. The better instrument is a category study fielded to a panel sample, which reaches buyers who never became opportunities.

You need precise shares from a single quarter. This one is a limit of sample, not of the instrument. At 20 to 30 responses, a 50 percent share carries an interval of roughly 17 to 20 points either side, and smaller shares are lopsided.

The fix is pooling windows, covered in the sample-size section. The alternative for an urgent single-quarter read is a handful of interviews that explain instead of count.

You need to know why customers leave after buying. Post-sale defection has different causes and a different respondent. The better instrument is churn and cancellation research, and a voice of customer survey covers how customers experience you after the sale.

You need to evaluate individual reps. Per-rep sample sizes are too small to separate a rep from their territory and deal mix. A rep with four responding losses in a year has an interval on any reason share that spans most of the scale. The better instrument is deal review and call recordings, used for coaching. Ranking reps on win-loss scores can also give reps a reason to discourage responses.

The first limitation deserves one more sentence, because it shapes the whole program. Treat interviews as the second half of the method, drawn from survey respondents who agreed to talk, so the program counts across every deal and explains a few.

The second limitation is where readouts most commonly go wrong. A gap on a factor invites an action plan, and an action plan built on an association in 17 deals can spend a quarter of engineering time on the wrong fix.

Hold the first two limitations hardest when results are presented. They’re the two questions stakeholders most want a win-loss survey to answer, because "why" and "what to fix" are the questions that make a slide.

Designing the Study and the Five Forks

Every win-loss survey takes five design decisions. Listing question categories without settling these five is how a program ends up with a dashboard nobody trusts.

The forks are the instrument, the unit of analysis, the outcome coding, how reasons are measured, and whose account you collect. The committed default comes first, and the evidence follows.

Survey every decided opportunity above a size threshold you set, on a fixed cadence, and interview a stratified subset of respondents who agree to a follow-up. Code three outcomes: won, lost to a named alternative, and lost to no decision.

Invite every known buying-group contact but analyze at the deal level. Ask for one primary reason and a comparative rating of your offer against the chosen alternative on the same factor list for everyone. Keep the rep’s CRM reason as a column and report how often it matches the buyer’s.

Fork 1: Survey, Interviews, or Both

Run both, with the survey as the frame. The survey reaches every decided deal, removes interviewer effects, and produces counts that can be compared across windows.

Interviews typically recover what the survey can’t: sequence, politics and the reason behind the stated reason. Draw them from survey respondents who consented, stratified by outcome and competitor, instead of interviewing whoever says yes first. That design gives the interview program a known relationship to the whole pipeline, which a stand-alone interview program generally lacks.

Fork 2: The Rep's Account or the Buyer's

Collect the buyer’s account, and keep the rep’s as a comparison column. The reason is a large and well-documented attribution bias.

Mezulis, Abramson, Hyde and Hankin, in a 2004 meta-analysis in Psychological Bulletin, pooled 266 studies and 503 effect sizes on the self-serving attributional bias. They found "a large bias," averaging d = 0.96, present "in nearly all samples," though its size varied across groups.

People credit success to themselves and failure to circumstance. A closed-lost reason typed by the rep who lost the deal is exactly that attribution, and "price" and "missing feature" are the circumstances most often available.

That finding comes from general samples, not sales teams specifically, but the mechanism it describes is the one a closed-lost field asks for. The design rule follows: never report the CRM field as the reason a deal was lost.

Fork 3: Two Outcomes or Three

Code three outcomes: won, lost to a named alternative, and lost to no decision. A deal lost to no decision has no chosen vendor to rate you against, needs its own question branch, and usually has different causes.

The size of that cell is commonly larger than teams expect. Dixon and McKenna, in their 2022 book The JOLT Effect, attribute "40-60% of lost deals to indecision," from machine-learning analysis of 2.5 million recorded sales conversations.

That’s proprietary data published in a trade book, not a peer-reviewed study, so treat it as an indication and not a benchmark. It’s still enough to justify the rule, because a pipeline where half the losses have no competitor can’t be understood with a two-outcome code.

Exclude deals still open. A buyer who reports "still deciding" routes out of the survey, and a mismatch with the CRM outcome is itself worth recording.

Fork 4: Stated Reasons, Derived Reasons, or Both

Collect both, because a stated primary reason is what the buyer says drove the decision, and it’s a plausible account rather than a measured cause.

Nisbett and Wilson, in Psychological Review in 1977, reviewed evidence that "there may be little or no direct introspective access to higher order cognitive processes," and proposed that people reporting on their reasons "do not do so on the basis of any true introspection."

So pair the stated reason with a derived one. Ask every buyer to rate your offer and the chosen alternative on the same six to eight decision factors, then compare the relative rating between won and lost deals.

Where the two disagree, report both. Buyers who name price while the won-versus-lost gap sits on implementation confidence are telling you something neither measure shows alone.

Keep the derived method simple. Bacon, in the International Journal of Market Research in 2003, compared importance-performance approaches across 15 datasets and found "the traditional 2x2 grid approach can be misleading" and that "indirect (e.g. multiple regression) methods for determining importances may also be misleading." At win-loss sample sizes, report the gap per factor with an interval, and don’t regress the outcome on a dozen ratings.

Fork 5: The Unit of Analysis

Analyze deals, not respondents. Invite every known contact on an opportunity, because different roles see different parts of the decision, then collapse to one record per deal before computing any share.

Use the decision-maker’s answer where one is identified, and a documented rule otherwise, such as the most senior respondent. This rule is practitioner judgment that follows from the unit-of-analysis problem, not a cited finding.

Counted per respondent, one large deal with five respondents outweighs four small deals with one each. The result would describe your biggest buying committees and not your pipeline.

The Design Tree

Work down the questions in order until a branch fires.

Do you close enough deals to count? If yes, run the survey on every decided deal above your threshold. If you close a handful a year, run interviews alone and skip the rest of this guide.
Can you reach buyers through a channel you control? If yes, send from your own email domain or a shared link. In-product surveys can’t reach lost buyers, who never became users, and panels can’t reach a specific buyer.
Can your CRM trigger a send when a deal closes? If yes, send automatically at close. If no, export decided opportunities on a fixed cadence and send one batch per window.
Did the deal end with no decision? If yes, route to the no-decision branch and rate your offer against keeping things as they are. If no, rate your offer against the chosen alternative.
Does the opportunity have more than one respondent? If yes, collapse to one record using your written rule before any analysis. If no, the respondent is the deal.
Will any reported cell fall under the base your decision needs? If yes, pool windows before reporting, and read the sample-size tables first.

In Sprig: Email distribution sends from your own domain and takes a comma-separated values (CSV) contact list per email run. No CRM trigger is documented, and the Zapier integration only pushes responses out, so a win-loss program on Sprig runs as a batch export from the CRM on a fixed cadence. Email is available to Enterprise teams on link surveys only.

Writing the Instrument

A win-loss instrument has a fixed order, and the order protects the answers. Confirm the decision first, ask the primary reason before any list of factors, and ask about the process last.

The Instrument Spec

This is the instrument in fielding order, ready to lift. Replace the competitor list and the factor list, and keep the order.

Confirm the decision. "Which option did your organization choose?" Choices: your company, each named competitor, built in-house, decided to do nothing for now, still deciding. "Still deciding" routes out.
Primary reason. "What was the single most important reason for that decision?" Forced single choice, options randomized, with an "other" text box.
The reason in their words. "In a sentence or two, what made that reason decisive?" Open text.
Comparative factor rating. Rate your company and the chosen alternative on the same six to eight decision factors, on the same scale. The no-decision branch rates your company against "keeping things as they are."
Evaluation process. Who was involved, how many options were seriously considered, and when the decision was made.
The near-loss question, for won deals. "What almost made you choose differently?"
Reconsideration, for lost deals. "What would need to be true for you to consider us in future?"
Follow-up permission. "Would you be willing to speak with a researcher for 20 minutes about this decision?"

Four rules sit under the list:

Use the same factor list for every outcome so won and lost deals compare
Never frame questions as "why did we lose," which invites a polite answer
Never show the respondent the rep’s recorded reason for the outcome
Keep the respondent’s total time under 10 minutes including open text

The fourth rule is a working convention. Revilla and Ochoa, in the International Journal of Market Research in 2017, found a median ideal web survey length of 10 minutes among panelists in one country, and busy buyers owe you less patience than panelists do.

Choosing the Factor List

Build the factor list from your last round of interviews or deal reviews, not from a generic template. Six to eight factors is a common working range, but that range is a convention, not a sourced rule.

A typical business-to-business (B2B) software list includes:

Product fit
Implementation confidence
Integration coverage
Total cost
Sales experience
Vendor viability
Security and compliance
Time to value

Freeze the list once fielding starts. A factor added mid-program has no history, and a factor reworded mid-program is a different factor.

Why the Near-Loss Question Matters

Many programs analyze losses first and treat won deals as a control group. A buyer who chose you can often say what nearly stopped them.

That answer is frequently the earliest warning of a loss pattern, because it describes a weakness buyers were still willing to overlook. Track it across windows the same way as primary loss reasons.

In Sprig: The Design Agent can draft this instrument from a written brief, including the outcome branch and randomized reason order. It won’t know your factor list, your competitor names or your outcome codes unless the brief states them, so check all three before fielding. Display logic, which the outcome branch depends on, is available on Enterprise plans.

Question Formats and Scoring

Three format decisions shape the data more than any wording choice: how the primary reason is asked, how the comparative rating is built, and how open text is handled.

The Primary Reason as a Forced Single Choice

Ask for one reason, not "select all that apply," which commonly flattens the answers. A multi-select list typically produces a long tail of ticked boxes that all look equally important.

Randomize the order of reasons so the first option doesn’t collect answers by position. Keep the list to the same codes you’ll report, plus "other," so the stated reason and the CRM reason can be mapped to one code list later.

The Comparative Rating

Rate both vendors on the same five-point scale, question by question, instead of asking "how did we compare." The analysis then uses the difference: your rating minus the chosen alternative’s rating, which runs from minus four to plus four.

Read the difference, not the absolute scores. Absolute scores often drift with respondent mood and politeness, and the difference cancels much of that drift because both ratings come from the same person.

For the no-decision branch, the comparison is your offer against the status quo. That comparison answers a different question, and its gap should be reported separately from the competitive gap.

Open Text

Keep open text to two questions: the reason in their words and the near-loss or reconsideration question. Each additional open box adds time that the 10-minute limit has to absorb, and a third box rarely finds a theme the first two missed.

Code open text against the same reason codes as the forced choice first, and add new codes only when a cluster clearly doesn’t fit. That keeps the open text useful as a check on the forced choice and not a separate, uncomparable dataset.

Sample Size and Precision

Size a win-loss program from the precision each reported cell needs, not from a single quarter’s volume. Win-loss samples are small by nature, and the arithmetic below is what small looks like. The same interval logic applies to any survey, and the guide to analyzing survey data with AI covers the wider rigor checks.

No methodology source publishes a win-loss sample-size rule. Published vendor guidance generally gives either no figure or a count of interviews, and a count of interviews answers a different question from a count of survey responses.

Two Questions That Get Confused

Discovery asks whether a reason that affects some share of deals will show up at least once. The chance is one minus (one minus p) to the power n, where p is the reason’s true share and n is the number of responses.

Estimation asks what share of deals a reason affects, and how precisely. The answer is a confidence interval on a proportion.

Discovery is cheap and estimation is expensive. Growth Velocity, a win-loss consultancy, publishes a floor of about 20 interviews, derived from the discovery formula credited to usability researchers James Lewis and Robert Virzi. That derivation is sound for discovery. It says nothing about estimating a share.

Use the Wilson Interval at Small N

Use the Wilson score interval and not the textbook interval of 1.96 times the standard error. Brown, Cai and DasGupta, in Statistical Science in 2001, showed that the textbook interval’s coverage is erratic at small n and recommended the Wilson or Jeffreys interval for small samples.

The textbook interval can also run below zero. At two mentions in 20 responses it gives minus 3 to 23 percent, which isn’t a possible range for a share.

The Derived Tables

The tables below are computed, not copied. Wilson intervals were computed by a hand-coded formula and checked against a statistics library, and detectable gaps were computed two ways, by the normal approximation and by the arcsine method.

They assume 95 percent two-sided confidence, 80 percent power for detection, and deal-level records.

Table A. 95 percent Wilson interval for a reason share, by number of deals

| Deals | 50% observed | 30% observed | 10% observed | |:-----:|:------------:|:------------:|:------------:| | 10 | 24 to 76% | 11 to 60% | 2 to 40% | | 20 | 30 to 70% | 15 to 52% | 3 to 30% | | 30 | 33 to 67% | 17 to 48% | 3 to 26% | | 50 | 37 to 63% | 19 to 44% | 4 to 21% | | 100 | 40 to 60% | 22 to 40% | 6 to 17% |

Table B. Chance a reason affecting p percent of deals appears at least once

| Reason's true share | 5 deals | 10 | 20 | 30 | 50 | |:-------------------:|:-------:|:---:|:--------:|:--------:|:--------:| | 5% | 23% | 40% | 64% | 79% | 92% | | 10% | 41% | 65% | 88% | 96% | 99% | | 20% | 67% | 89% | 99% | over 99% | over 99% | | 30% | 83% | 97% | over 99% | over 99% | over 99% |

Read the other way, a reason affecting 10 percent of deals needs 16 deals for an 80 percent chance of appearing at least once, and 29 for a 95 percent chance. A reason affecting 5 percent needs 32 and 59, a reason affecting 20 percent needs eight and 14, and a reason affecting 30 percent needs five and nine.

Table C. Smallest won-versus-lost gap on a 50 percent measure detectable at 80 percent power

| Deals per group | Normal approximation | Arcsine method | |:---------------:|:--------------------:|:--------------:| | 20 | 40 points | 39 points | | 30 | 33 | 33 | | 50 | 28 | 27 | | 100 | 20 | 19 | | 200 | 14 | 14 |

The two methods agree within about one point at every size shown. The lesson is in the column itself: at 20 or 30 deals per group, only gaps of 33 to 40 points are detectable, which is why a single-window gap test rarely settles anything.

| Completes needed | 5% response | 10% | 20% | 33% | |:----------------:|:-----------:|:---:|:---:|:---:| | 20 | 400 | 200 | 100 | 61 | | 30 | 600 | 300 | 150 | 91 | | 50 | 1,000 | 500 | 250 | 152 |

What the Tables Mean in Practice

A quarter’s 25 lost-deal responses will reliably surface any reason affecting a fifth of losses. The same 25 responses can’t tell a 40 percent reason from a 60 percent one, and can’t detect a won-versus-lost gap smaller than roughly 35 to 40 points.

So the practical unit of reporting is usually a rolling window, not a quarter. Pool two or three quarters for estimation, and report each new quarter for discovery only.

Zero Mentions and Small Populations

If a reason never appears in n responses, the one-sided 95 percent upper bound on its share is one minus 0.05 to the power one over n. That’s close to the rule of three, 3 divided by n, described by Hanley and Lippman-Hand in JAMA in 1983.

Thirty responses with zero mentions of a reason still allow a true share of up to about 10 percent. Absence in a small sample is weak evidence of absence.

When the population of decided deals is small, a finite population correction narrows the interval. Thirty responses from 80 lost deals gives about plus or minus 14 points at 50 percent, against about 18 without the correction.

That correction assumes respondents are a random subset of deals, which nonresponse makes doubtful. Treat it as an upper bound on your precision, not as the estimate.

No Minimum Floor

This guide publishes no minimum sample. Pick the interval width in Table A that your decision can tolerate, then pool windows until each reported cell reaches it.

Audience, Targeting and Who to Invite

Invite every known buying-group contact on each decided opportunity above your size threshold. Different roles often see different parts of a decision, and the deal-level collapse handles the weighting problem later.

Set the size threshold in writing before the first send. Including every small deal typically floods the data with fast transactional decisions that have different causes from the deals the program exists to explain.

Pull contacts from the opportunity’s contact roles, not from the account, which often holds a broader list. An account-level list commonly includes people who never saw the evaluation.

Record each respondent’s role: economic buyer, technical evaluator, end user, procurement. Role is typically the first cut most programs want, and it can’t be reconstructed afterward.

Screen out nobody on the survey side. The confirmation question does the screening, because a respondent who says "still deciding" or names an option inconsistent with the CRM record flags the data instead of disqualifying the deal.

Check your consent basis and suppression lists before the first send. An email to someone who declined to buy can count as marketing contact in some jurisdictions, and the rules depend on where the buyer is located.

In Sprig: The email contact list requires Email, First Name, Last Name and User ID columns. Additional columns are documented only as email personalization variables, and whether they reach the response export isn’t documented.

Put the CRM opportunity or contact ID in the User ID column, then join outcome, competitor, amount and the rep’s reason after export. The platform enforces a limit of one email per person per 24 hours across contact lists.

Fielding and Delivery

Send the invitation from a person or a research function the buyer recognizes, from your own domain. Cycyota and Harrison, in a 2006 meta-analysis of executive surveys in Organizational Research Methods, found that "topical salience and sponsorship by an organization or person in the executive’s social networks" raised response, while follow-ups and personalization showed no significant effect.

Use a short note that states the purpose plainly, not a sales-branded email: you want to learn from the decision, the rep won’t see individual answers, and the survey takes a few minutes. Sprig’s email deployment supports a custom sending domain for exactly this reason.

Promise confidentiality and keep it. Identify responses to the research owner so they can be joined to the CRM, and never show individual answers to the rep who owned the deal.

Send one reminder at most. No study tests reminders for win-loss, and the executive-survey literature doesn’t settle the question, so treat one reminder as a working convention and not a lever.

Keep the invitation short and the survey shorter.

An Invitation You Can Adapt

The invitation typically works best at under 100 words, signed by a named person in research or product marketing and not sales. A version to adapt:

"You recently evaluated [PRODUCT CATEGORY] options and made a decision, and we would like to learn from it, whichever way it went. The survey takes about [PILOT MEDIAN] minutes. Your answers go to our research team, not to the account team you worked with, and they’re used only to improve how we build and sell. Thank you for considering it. [NAME], [TITLE]"

Keep the same wording across windows. A changed invitation is a changed fielding condition, and response rates across windows generally stop being comparable when it moves.

The Pre-Launch Checklist

Work through this list in order before the first send. Most items can’t be fixed once responses start arriving.

Write the three outcome codes and the size threshold down, and confirm the CRM export uses them.
Write the deal-level collapse rule down before any response arrives.
Confirm the contact list pulls from opportunity contact roles, not the account.
Put the CRM ID in the join column and test the join on a sample export.
Confirm the outcome branch routes no-decision buyers to the status quo comparison.
Confirm reason order randomizes and the factor list is identical across branches.
Confirm the invitation comes from your own domain and names a research sender.
Confirm no report, dashboard or CRM field shows individual answers to the deal’s rep.
Check your consent basis and suppression lists for every region in the send.
Pilot with a few colleagues and record the median completion time.

Items 2, 4 and 8 can’t be fixed after the first readout. A missing collapse rule, a broken join key or a leaked answer typically surfaces only then, when the fix means rebuilding trust with the sales team or fielding again.

In Sprig: AI Follow-Ups from the Field Agent can add a short probe after the primary reason, asking the buyer to expand on it in their own words. A follow-up is still self-report and not an interview, and it adds verbatims to code rather than replacing the coding step.

Timing and Cadence

Send as soon as the outcome is final and the buyer has been told. Memory of a decision’s reasons generally fades and gets rewritten with time, and a fixed interval after close keeps windows comparable.

No study publishes a send window for win-loss. Klue, a competitive intelligence vendor, recommends sending within 30 days, based on practitioner experience, which is a reasonable working window, not a rule.

Send on a fixed cadence if your CRM can’t trigger sends at close. A monthly batch caps the gap between close and invitation at one cycle plus the time it takes to build the contact list.

Report on a slower cadence than you send. A monthly send with quarterly discovery reporting and a rolling two-to-three-quarter estimation window generally matches the sample arithmetic. No source publishes a reporting cadence for win-loss, and this guide doesn’t invent one, so match the reporting window to the precision in the sample-size tables.

Avoid sending during the buyer’s implementation kickoff for won deals. A new customer in the middle of onboarding commonly answers about onboarding and not about the decision, which blurs win-loss with early customer experience.

Hold the send interval constant when volume changes. A quarter with twice the usual closes is still sent on the same schedule, so the extra deals add sample and don’t change the timing.

Quality Control and Data Hygiene

Three checks matter most, and the first two are specific to win-loss.

Outcome mismatches. Compare the buyer’s confirmed decision with the CRM outcome. A buyer who reports choosing a competitor on a deal coded "no decision," or who reports still deciding, points to a CRM coding problem that will distort every share.

Multi-respondent disagreement. When two contacts on one deal name different primary reasons, keep both in the respondent file and apply the collapse rule for the deal file. A high disagreement rate is often a sign the decision had several owners.

Speeders and straight-liners. Flag respondents who rate both vendors identically on every factor, or finish in a fraction of the pilot median time. Review before deleting, because a buyer who genuinely saw no difference between two vendors is a finding.

The Nonresponse Check

Compare respondents against nonrespondents on every CRM field you hold: outcome, deal size band, segment, competitor and rep. Win-loss is one of the rare surveys where the frame is rich enough to do this for every nonrespondent.

If lost deals respond at a materially different rate from won deals, or large deals respond more than small ones, report shares within outcome and size cells and not pooled. Report the response rate by outcome with every readout.

Report exclusions with the results, stating the starting count, the number removed, the rule used, and the final count by outcome.

The Core Calculation

A win-loss survey produces five numbers that no other study in this series produces together: deal-level response rates by outcome, reason shares with intervals, the won-versus-lost factor gap, the rep-buyer agreement rate, and the nonresponse comparison.

A worked example with illustrative numbers shows all five. A mid-market human resources software company closes 140 opportunities over two quarters: 52 won, 48 lost to a competitor, and 40 lost to no decision.

Deal-Level Response Rates

Invitations go to 310 contacts across the 140 deals, and at least one contact responds on 54 deals, 39 percent of the pipeline.

By outcome, 26 of 52 won deals responded, 50 percent. Lost-to-competitor deals responded at 17 of 48, about 35 percent, and no-decision deals at 11 of 40, about 28 percent.

That gradient is typical of the problem the critique section describes. The deals you most need to understand are the ones least likely to answer, so every share below is reported within its outcome cell, never pooled.

Reason Shares with Intervals

Among the 17 lost-to-competitor deals, buyers name implementation confidence seven times, price four times, a missing integration three times, the sales experience twice, and another reason once.

| Primary reason, lost to competitor | Deals | Share | 95% Wilson interval | |:----------------------------------:|:-----:|:-----:|:-------------------:| | Implementation confidence | 7 | 41% | 22 to 64% | | Price | 4 | 24% | 10 to 47% | | Missing integration | 3 | 18% | 6 to 41% | | Sales experience | 2 | 12% | 3 to 34% | | Other | 1 | 6% | 1 to 27% |

The bottom two intervals are too wide to rank anything. Implementation confidence leads, but its interval overlaps price’s. The honest reading is that implementation confidence is a leading reason worth investigating, not that it’s proven to outrank price.

Among the 11 no-decision deals, five cite an internal priority shift, a share of 45 percent with an interval of 21 to 72 percent. At 11 deals that cell supports discovery, not estimation.

The Won-Versus-Lost Factor Gap

Take the relative rating on implementation confidence: your rating minus the chosen alternative’s, on a scale from minus four to plus four. Won deals average plus 1.4 and lost deals average minus 1.1.

The gap is 2.5 points. With standard deviations of 1.6 and 1.8 on bases of 26 and 17, the 95 percent interval on the gap runs from about 1.4 to 3.6 points, computed with a Welch t-interval.

That interval excludes zero, so implementation confidence separates won from lost deals in this sample. Combined with its lead among stated reasons, it becomes the program’s first hypothesis to test.

Rep-Buyer Agreement

For the same 17 lost deals, reps recorded price 10 times, a missing feature five times, and another reason twice.

After mapping both to one code list, rep and buyer agree on five of 17 deals, 29 percent. Cohen’s kappa, which discounts the agreement expected by chance, comes out near 0.1 with these counts. The two code lists differ, because reps rarely record a code like implementation confidence, so part of the disagreement is structural.

Reps named price on 10 deals where buyers named it on 4. Don’t read that as reps being careless. Part of the gap may be the self-serving attribution the design section describes, and part is structural, since price may also be the polite answer buyers give. Track the agreement rate as a program-health metric.

The Nonresponse Comparison

Compare the 54 responding deals with the 86 nonresponding deals on deal size band, segment and competitor.

If responding lost deals skew small, or skew toward one competitor, report within those cells. In this example the check is what licenses reporting lost-to-competitor shares at all.

Cutting any of these numbers by segment belongs to cross-tab analysis, and coding open text into themes belongs to the theming workflow.

Benchmarks and What a Good Result Looks Like

There’s no published, sourced benchmark for win-loss survey results.

No methodology source stands behind a cross-industry norm for loss-reason shares, win rates or win-loss survey response rates, and the figures in circulation come from vendors who don’t publish their method.

Figures in Circulation and Why This Guide Doesn't Use Them

Clozd, a win-loss interview vendor, reports that reps were wrong about the real reason for the outcome 85 percent of the time. It gives no sample size, time period or method.

Clozd also reports interview participation above 30 percent and survey participation below 5 percent, again with no source.

A survey software competitor’s win-loss article states that a strong program "can boost win rates by up to 50%," with no source.

Each of these may reflect real experience, but none is checkable. Quoting them would put an unsourced number at the center of a readout.

The One Reference Point That Transfers

Academic surveys of executives return about a third of invitations. Cycyota and Harrison reported an overall response rate of 32 percent across 231 mailed surveys of executives from 1992 to 2003.

Those are academic surveys of top managers, not win-loss surveys. A win-loss survey to lost buyers should expect to do no better and should plan for worse.

Build Your Own Benchmark

Your benchmark is your own previous windows on the same instrument, the same outcome codes and the same size threshold. That’s the logic of a brand tracking survey, where the instrument stays fixed so movement means something.

Track three program-health numbers across windows: response rate by outcome, the respondent-versus-nonrespondent comparison, and the rep-buyer agreement rate.

A rising agreement rate over several windows generally means the sales team is absorbing the buyer’s account. That’s often the most useful trend a win-loss program produces, and it needs no external norm.

The Program-Health Scorecard

Report these five lines at the top of every readout, before any reason share:

The warning column describes patterns, not thresholds. No source publishes a cutoff for any of these five, so the scorecard is read against your own previous windows.

Opening with the scorecard puts the evidence’s limits in front of stakeholders before the reasons, so an overlapping interval arrives with its context already on the page.

Interpreting and Acting on the Result

Read the output in a fixed order, because the order is what stops a win-loss readout from becoming proof of whatever each team already believed.

Start with response by outcome. If lost deals responded at a fraction of won deals' rate, every comparison between the two carries that caveat on its first line.

Then no-decision reasons. This cell is often the largest loss cell and the one sales teams discuss least. Its causes, such as priority shifts, failed consensus and budget freezes, generally call for different fixes from competitive losses.

Then competitive loss reasons, by competitor. A reason that dominates losses to one competitor and never appears against another is a positioning finding, not a product finding.

Then the factor gaps. Read the gaps with their intervals. A factor where the gap is large and the interval excludes zero is often a candidate cause, and a factor buyers name often but with no gap is usually a socially acceptable reason and not a decisive one.

Then rep-buyer agreement. Report it without naming reps. The point is to calibrate the CRM field, not to grade individuals, and in the worked example a kappa near 0.1 says the field and the buyers barely agree beyond chance.

Then the near-loss answers from won deals. Set them beside the loss reasons. A weakness that appears in both is a finding two independent questions support.

Here’s how one finding typically travels through that order, using the worked example. Implementation confidence leads the stated loss reasons against competitors, and its factor gap excludes zero.

Suppose nine of the 26 won buyers also named implementation in their near-loss answers, 35 percent with an interval of 19 to 54 percent. Reps, meanwhile, recorded price on 10 of the 17 competitive losses where buyers named it on 4.

That chain generally supports a specific hypothesis: buyers doubt the implementation plan, and the sales team reads the doubt as price resistance. The next step isn’t cutting price. It’s interviewing a few lost buyers about implementation and testing a revised implementation offer.

What the Result Licenses

A win-loss survey licenses hypotheses and priorities: which reason to investigate first, which competitor to prepare for, which part of the sales process to examine.

It doesn’t license a claim that fixing a factor will raise the win rate. That claim needs an experiment, such as a message testing survey for positioning or a conjoint analysis for price.

In Sprig: AI open-text theming from the Synthesize Agent codes the reason and near-loss verbatims into themes with response-level traceability. At 20 to 40 verbatims per window, read every response yourself as well.

Themes regenerate and no reproducibility claim is published, so theme counts aren’t comparable across windows, and each window should be coded against the fixed reason list. Researchers remain responsible for the final coding.

Running the Analysis with Claude or ChatGPT

The win-loss calculations are simple arithmetic on two joined files, where a model helps with speed and can fail quietly on joins and counts. Every interval and gap in this guide is computed outside the survey platform either way, because Sprig publishes no significance testing.

This chapter covers only what’s specific to win-loss. Theme creation belongs to the advanced cross-tab guide, segment cuts belong to the simple cross-tab guide, and connector setup belongs to the documentation.

What the Analysis Produces

The analysis produces five outputs: a joined deal-level file, response rates by outcome with the nonresponse comparison, reason shares with Wilson intervals per outcome, factor gaps with intervals, and the rep-buyer agreement rate.

Save each output as its own named file. Separate files make each number checkable on its own, where one long summary doesn’t.

Prompt One: Join, Collapse and Count

You’re analyzing a win-loss survey joined to a CRM export.

WHAT THIS PROMPT DOES: joins survey responses to CRM opportunities, collapses
multi-respondent deals to one record, computes response rates by outcome with a
nonresponse comparison, and computes primary-reason shares with intervals.
WHAT IT RETURNS: a deal-level file, a response table and a reason table.

Survey file: [PATH TO SURVEY EXPORT]
CRM file: [PATH TO CRM EXPORT OF ALL DECIDED OPPORTUNITIES]
Join key: [SURVEY COLUMN HOLDING THE CRM ID, default userId]
CRM outcome column: [COLUMN, values WON, LOST_COMPETITOR, LOST_NO_DECISION]
Primary reason question: [QUESTION COLUMN, default Q2_Response]
Role column: [ROLE COLUMN OR ATTRIBUTE, default Attributes_1]
Collapse rule: [RULE, default use the economic buyer, else the most senior role]
Minimum base: [MINIMUM DEALS PER CELL, no default, set from the interval width in Table A that your decision tolerates]

Use code to calculate this, not estimation. Don’t compute any value by
reasoning about it in prose.

Rules:
1. Before any analysis, report survey rows with no CRM match and CRM
   opportunities with no survey row, as two counts.
2. Exclude a response from reason counts if the primary reason is blank, null
   or a non-response code. Blank isn’t "other."
3. Exclude any deal where the respondent reports still deciding.
4. Collapse to one record per opportunity using the collapse rule. Report how
   many opportunities had more than one respondent and how many of those had
   respondents who disagreed on the primary reason.
5. Report response rate by outcome as responding deals over decided deals.
6. Compare responding and nonresponding deals on outcome, amount band,
   segment and competitor, as counts and shares.
7. Report each primary-reason share within each outcome with a 95 percent
   Wilson score interval.
8. Any cell with fewer deals than the minimum base, including zero, is
   reported as below threshold and not interpreted.

Recount deals per outcome and per reason a second way, with an explicit loop
over the deal file in place of a grouped count, and report both counts.
Recompute every interval a second way with the exact Clopper-Pearson method.
The two methods differ by design, so report both and flag any share where
they disagree on whether the interval overlaps another reason’s interval.

If a count, share or interval doesn’t reconcile between the two passes, say
that you can’t reconcile it and show me the inputs. Don’t correct a mismatch
by choosing the answer that looks more reasonable.

Save as winloss_deals.csv, winloss_response.csv and winloss_reasons.csv.

Prompt Two: Factor Gaps and Rep-Buyer Agreement

WHAT THIS PROMPT DOES: computes the won-versus-lost gap on each decision
factor and the agreement between rep-recorded and buyer-stated reasons.
WHAT IT RETURNS: a factor-gap table and an agreement table.

File: winloss_deals.csv
Factor rating columns for our company: [OUR RATING COLUMNS, default Q4_Response to Q11_Response]
Factor rating columns for the chosen alternative: [ALTERNATIVE RATING COLUMNS, default Q12_Response to Q19_Response]
Rep reason column: [REP REASON COLUMN, default Closed_Lost_Reason]
Code map: [MAPPING OF REP AND BUYER REASONS TO ONE CODE LIST, default the buyer reason list with "other" for anything unmapped]
Minimum base: [MINIMUM DEALS PER GROUP, no default, set from Table C for the gap your decision needs to see]

Use code to calculate this, not estimation.

Rules:
1. For each factor, compute the relative rating as our rating minus the
   chosen alternative’s rating. Exclude a deal from a factor if either rating
   is blank. Blank isn’t zero.
2. Compare won deals with lost-to-competitor deals only. Report no-decision
   deals separately against the status quo, never pooled with competitive
   losses.
3. Report the mean relative rating per group, the gap, and a 95 percent Welch
   t-interval on the gap.
4. Any group with fewer deals than the minimum base, including zero, is
   reported as below threshold and not interpreted.
5. Map rep and buyer reasons to the code list and report percent agreement
   and Cohen’s kappa. Don’t name or rank individual reps.

Recompute every factor gap interval by bootstrapping [ITERATIONS, default
5000] resamples of deals within each group. Compute kappa once from a
hand-built confusion matrix and once with a library function, and report both.

If the two passes disagree on any value, or the bootstrap and Welch intervals
disagree on whether the gap excludes zero, say that you can’t reconcile them
and show me the inputs. Don’t pick the answer that looks more reasonable.

Don’t describe any factor as a driver or cause of winning. Describe it only
as separating won from lost deals in this sample.

Save as winloss_factor_gaps.csv and winloss_agreement.csv.

Setting Up and Importing Your Data

Export the survey side as CSV, one row per respondent, with the join column intact.

Export the CRM side as every decided opportunity in the window, including those with no response, because the nonresponse comparison needs the full denominator.

The join key is whatever ID you placed in the invitation list when you sent the survey. If that column held a contact ID, bring a contact-to-opportunity map from the CRM as a third file.

Put the data above the instructions in the prompt, and keep the two files separate instead of pre-joining them by hand.

In Sprig: Export the responses as CSV or pull them through the Model Context Protocol (MCP) connector. The documented export columns include surveyId, userId, createdAt, completedAt, Q#_Question_Text, Q#_Response, Themes and Attributes_#, and the join key is the User ID column from the email contact list.

A win-loss window typically fits in one MCP call, which returns at most 1,000 responses. Whether custom contact-list columns reach the export isn’t documented, which is why the CRM fields are joined afterward.

Reading the Output

Read the join report first. Unmatched rows on either side generally mean a broken ID in the contact list, and every downstream share inherits the error.

Read reason shares as intervals. Two reasons whose intervals overlap aren’t ranked, however far apart their point shares look.

Read factor gaps from the interval, not the point, because a gap whose interval crosses zero has no finding in this window.

What to Verify Before Reporting

Check five things before any number leaves the analysis:

Confirm the deal count after collapsing matches the CRM count of responding opportunities.
Confirm no cell below the minimum base is interpreted anywhere in the readout.
Confirm the Wilson and Clopper-Pearson passes agree on which reasons overlap.
Confirm no-decision deals are reported separately from competitive losses.
Recompute one reason share and one interval by hand from the counts.

If the hand calculation disagrees with the model, stop and rerun instead of reporting either number.

Researchers remain responsible for the code map, the collapse rule and every number that leaves the analysis. The model does the arithmetic, and it doesn’t decide what the arithmetic means.

Pitfalls

Six documented failure modes apply to any survey analysis run in a language model, and two more apply to win-loss files specifically.

Discovering themes is harder than applying them. Hill and colleagues, in PLOS Digital Health in April 2026, found models matched human analysts on coding against an existing codebook, 93.5 percent against 92.7, and were materially worse at inductive discovery. Code win-loss text against your fixed reason list.

Rare themes get over-predicted. Ashwin, Chhabra and Rao, in Sociological Methods and Research in May 2025, found non-random bias in 10 of 19 codes, with sparse codes systematically over-predicted. At win-loss sample sizes, almost every code is sparse.

Outputs vary between runs. Thinking Machines Lab reported in September 2025 that 1,000 completions at temperature zero produced 80 unique outputs. Neither chat client exposes a seed, so produce any reportable number twice.

The middle of a long input gets lost. Liu and colleagues, in the Transactions of the Association for Computational Linguistics in 2024, documented it. Put data above instructions.

Models tend to agree with the user. Sharma and colleagues documented sycophancy at the International Conference on Learning Representations in 2024, and OpenAI withdrew a model update in April 2025 for being overly agreeable. Never ask a model to confirm that price is the reason you lose.

Verbatims contain personal information nobody asked for. Win-loss text often names competitors' reps, internal colleagues and budgets. Whether that data trains a model depends on the vendor’s terms and your plan tier.

Ranking reasons whose intervals overlap. Asked for "the top reason," a model will name one. Specify that overlapping reasons are reported as tied.

Counting respondents as deals. Unless the prompt says otherwise, a model will count rows. Collapse to deals before any share is computed.

The Critique You Should Know

Three objections apply to win-loss surveys specifically. The defense is real, and it’s narrower than the method’s advocates usually claim.

Stated Reasons Are Constructed After the Fact

A buyer’s stated reason is an account, not a readout. Nisbett and Wilson’s review found that people describing the causes of their own judgments often report what seems like a plausible cause and not what actually moved them.

In win-loss, the plausible causes are frequently the ones buyers can say politely. "Price" and "missing feature" are easy to give a vendor, while "we didn’t trust your implementation team" or "our CFO preferred the other vendor’s CEO" aren’t.

The Decision Reshapes the Ratings

After choosing, people typically rate the chosen option higher and the rejected one lower. Brehm documented this spreading of alternatives in the Journal of Abnormal and Social Psychology in 1956.

Applied to win-loss, winners inflate their ratings of you and losers deflate them. A won-versus-lost gap on a factor therefore overstates the real difference between how the two groups experienced your offer.

The Buyers You Most Need Are the Least Likely to Answer

Lost buyers owe you nothing, and executives are typically hard to survey at the best of times. Academic surveys of top managers average about a third responding, and a win-loss survey to lost buyers generally does worse.

If willingness to respond is related to why a buyer said no, the respondents aren’t a fair picture of the losses. A buyer who felt badly treated by the sales process may be the least likely to spend a few minutes telling you.

The Three Objections Compound

The three objections typically stack and don’t cancel. The buyers who answer are a selected group, their stated reasons are the socially available ones, and their ratings are bent by the decision they already made.

A readout that reports "41 percent of losses were about implementation" without saying any of this is reporting a number that’s weaker than it looks on three separate counts, which is why the core calculation always travels with its interval, its response rate and its nonresponse check.

The Counter-Critique

Three responses, and each one is supported and not merely asserted.

Low response alone doesn’t mean bias, and win-loss can check. Groves and Peytcheva, in Public Opinion Quarterly in 2008, analyzed 59 methodological studies on when nonresponse rates relate to nonresponse bias, and the field generally reads the rate as a weak predictor of bias.

Bias depends on whether response propensity relates to the thing measured. Win-loss is unusual because you know every nonrespondent’s outcome, deal size, segment, competitor and rep, so you can test that relationship directly instead of assuming it.

Part of the post-decision spread is a measurement artifact. Chen and Risen, in the Journal of Personality and Social Psychology in 2010, showed that spreading "occurs even when preferences remain stable, because ratings imperfectly measure preferences while choices partially reflect them." Coppin and colleagues, in the Quarterly Journal of Experimental Psychology in 2014, found a genuine choice effect for rejected options only.

Some of the gap is real preference, some is inflation, and rating both vendors on the same scale reads the gap relative to that shared inflation.

The alternative is worse. Without a buyer survey, the organization runs on the rep’s CRM reason, which is the kind of attribution the self-serving bias literature describes. Mezulis and colleagues report an average d = 0.96 in general samples, and nobody has measured it in CRM fields. A buyer account you can partly check against CRM fields beats a rep account whose bias can’t be checked at all.

Common Mistakes

Trusting the CRM closed-lost field. The rep’s reason carries a self-serving attribution. Keep it as a comparison column, never as the finding.

Folding no-decision into lost. A deal with no chosen vendor can’t be rated against one, and its causes commonly differ. Code it separately and give it its own question branch.

Counting respondents instead of deals. Large buying committees otherwise outvote small deals, which often distorts every share. Collapse to one record per opportunity first.

Asking "select all that apply" for the main reason. Multi-select produces a flat list of ticks. Force a single primary reason and use the factor rating for the rest.

Ranking reasons whose intervals overlap. At 17 deals, 41 percent and 24 percent can be statistically indistinguishable. Report overlapping reasons as tied.

Reporting one quarter as a trend. A quarter supports discovery. Pool windows before estimating shares or comparing periods.

Letting the rep send the survey. A request from the person who lost the deal often invites a polite answer. Send from a research function and promise the rep won’t see individual responses.

Ignoring won deals. The near-loss question often surfaces a weakness buyers were still willing to overlook, which is the earliest warning a program gets.

Using win-loss scores to grade reps. Per-rep samples are too small, and grading teaches reps to discourage responses.

Rewriting the factor list every quarter. A reworded factor is a new factor with no history. Freeze the list and add to it only between reporting windows.

Mistakes in Fielding and Analysis

Surveying the account instead of the opportunity. Account-level contacts frequently never saw the evaluation. Pull contact roles from the opportunity.

Skipping the nonresponse comparison. The CRM frame makes it free, and it’s the only defense against the strongest critique of the method.

Sending three reminders. One reminder is a reasonable courtesy. More is associated with lower executive response across published studies, and it spends goodwill with buyers you may sell to again.

Reporting the pooled response rate only. A 39 percent overall rate can hide lost deals responding at far lower rates than won ones. Report response by outcome, every window.

Mixing no-decision buyers into the competitive factor gap. A no-decision buyer rated you against the status quo, not a competitor. Pooling the two gaps compares answers to different questions.

Letting the CRM outcome override the buyer. When a buyer reports choosing a competitor on a deal coded "no decision," the CRM record is often the one that’s wrong. Recode it and log the correction instead of discarding the response.

Quoting a verbatim that identifies the buyer. A memorable quote with a company name or a job title attached can let the rep identify the buyer. Strip identifying details before any verbatim leaves the research team.

Treating the first window as the baseline for everything. The first window often has the roughest code map and may have a lower response rate. Treat it as a pilot, and start trend lines from the second.

Comparing theme counts across windows. Regenerated themes are a different instrument each time. Code every window against the fixed reason list, as the advanced cross-tab guide describes.

Synthetic Respondents

Synthetic respondents aren’t usable for win-loss analysis at any sample size.

The evidence a win-loss survey collects is a specific buyer’s account of a specific decision. No language model has that information, so a synthetic answer to "why did your organization choose the other vendor" measures nothing the question defines.

The general evidence points the same way. Bisbee and colleagues, in Political Analysis in 2024, compared 7,530 real respondents against 3.6 million synthetic responses and found averages often close. But 48 percent of regression coefficients differed significantly from the human data, and the sign flipped 32 percent of the time, in ChatGPT-persona political survey data.

Won-versus-lost gaps are relationships between variables, and the guide infers, without having tested it, that they face similar risk.

Pre-testing wording is the one legitimate use. Running a draft instrument past a model can surface an ambiguous factor name or a missing answer option before any buyer sees it. Treat those outputs as editorial notes, not data.

How Sprig Supports Win-Loss Research

The platform’s fit for win-loss is the survey layer of a hybrid program: sending from your own domain, probing the stated reason, and handing responses to an AI client for the arithmetic.

Email delivery sends from a custom domain, which puts the invitation under your research function’s name and not a vendor’s. A second email run within the same survey can carry the reminder, built from a CSV of nonresponding contacts and spaced to respect the 24-hour limit on emails per person.

The Field Agent can probe the primary reason in the buyer’s own words through AI Follow-Ups. A follow-up is still self-report and not an interview, and follow-ups can be disabled per study.

The Synthesize Agent themes open-text answers, and a Zapier integration can send each response to Salesforce, HubSpot or Slack through a Zap you build. Route it to a research-only field or channel, because the confidentiality rule means the deal’s rep never sees an individual answer.

The MCP connector gives an AI client governed access to responses. A win-loss window generally fits in a single call under the 1,000-response cap, and agents can’t launch or modify a live study.

Attribution matters here. Theming, AI Study Reports and AI Follow-Ups are Sprig agents. The deal-level collapse, Wilson intervals, factor gaps, agreement rate and nonresponse comparison are run by the connected AI client, not by the platform.

What Happens Outside the Platform

No CRM trigger is documented, so sends run as batches exported from the CRM and don’t fire at close. Zapier is outbound only and can’t start a send.

Email is available to Enterprise teams on link surveys only. In-product surveys and panels don’t reach lost buyers, by construction.

Whether contact-list columns beyond the four required reach the response export isn’t documented, so the CRM join happens after export. No weighting, significance testing or sample-size guidance is published.

The email documentation doesn’t describe anonymity handling, so confidentiality from the rep is enforced by process and permissions. Hosting is on Amazon Web Services in the United States with no published regional options, which matters for buyers under data-residency rules.

The platform doesn’t offer synthetic respondents and doesn’t replace a researcher’s judgment on the code map, the collapse rule or the readout. Researchers remain responsible for all three.

Alternatives and Adjacent Methods

Buyer interviews fit when the question is what happened inside a deal. They’re qualitative instruments, outside the survey category, and they generally work best drawn from survey respondents who agreed to talk.

Deal reviews and call recordings fit when the question is how an individual rep sold, since they support coaching in a way win-loss sample sizes generally can’t.

Churn research fits when the question is why customers leave after buying, because the respondent and the causes are typically different. A voice of customer survey covers how customers experience you after the sale.

A brand tracking survey fits when the question is your position across the whole market, including buyers who never shortlisted you.

Gabor-Granger analysis or a conjoint analysis fits when win-loss points at price and you need to know what a change would do.

A message testing survey or a product concept test fits when win-loss points at positioning or a product gap. Both manipulate the stimulus, which a win-loss survey never does, so they typically answer the causal question directly.

A usage and attitudes study fits when you need a category view of how buyers choose before they reach your pipeline.

Frequently Asked Questions

What Is a Win-Loss Analysis Survey?

A win-loss analysis survey is a short questionnaire sent to buyers after a sales opportunity closes, asking why they chose your company, a competitor, or no vendor at all. It measures how often each reason appears across decided deals, and it typically pairs with interviews that explain individual deals in depth.

How Do I Run a Win-Loss Analysis Survey?

Run a win-loss analysis survey in seven steps. Code decided deals as won, lost to a named alternative, or lost to no decision, and set a deal-size threshold. Invite every buying-group contact from your own domain, ask one primary reason and a comparative factor rating, collapse responses to one record per deal, report each share with an interval, and compare respondents against nonrespondents using customer relationship management (CRM) fields.

What Questions Should a Win-Loss Survey Ask?

A win-loss survey should confirm the buyer’s decision, ask the single most important reason, ask for that reason in the buyer’s words, rate your company and the chosen alternative on the same decision factors, and ask about the evaluation process. Won deals add "What almost made you choose differently?" and every respondent is asked about a follow-up conversation.

Is a Win-Loss Survey or a Win-Loss Interview Better?

Use the survey to count and the interview to explain. A survey covers every decided deal and produces comparable counts, while an interview recovers the sequence and politics behind one decision. A common design runs the survey across the pipeline and interviews a subset of consenting respondents.

How Many Responses Does a Win-Loss Survey Need?

A win-loss survey needs enough deal-level responses for the precision each reported cell requires, not a fixed total. At 20 to 30 deals, a 50 percent reason share carries a 95 percent interval of roughly 30 to 70 percent, so most programs pool two or three quarters before estimating shares.

When Should a Win-Loss Survey Be Sent?

Send a win-loss survey as soon as the outcome is final and the buyer has been told, at a fixed interval after close so windows are comparable. No study publishes a send window, and practitioner guidance commonly suggests within 30 days, and a monthly batch export is a common working pattern when the CRM can’t trigger sends.

Who Should Receive a Win-Loss Survey?

Every known buying-group contact on each decided opportunity above your size threshold should receive the win-loss survey. Pull contacts from the opportunity’s contact roles, record each respondent’s role, and collapse responses to one record per deal before analysis.

Why Shouldn't I Rely on the CRM Closed-Lost Reason?

The CRM closed-lost reason is typed by the rep who lost the deal, and people generally attribute their own failures to circumstance. A 2004 meta-analysis of general samples found an average self-serving bias of d = 0.96, a large effect, so keep the CRM reason as a comparison column and not the finding.

How Should No-Decision Deals Be Handled in Win-Loss Analysis?

Code no-decision deals as their own outcome and don’t fold them into losses. A no-decision buyer has no chosen vendor to compare against, so their survey branch rates your offer against keeping things as they are, and their reasons are reported separately from competitive losses.

How Do I Get Lost Buyers to Respond?

Lost buyers respond more often to a short invitation from a recognized person or research function on your own domain, with a clear purpose and a promise that the rep won’t see individual answers. Research on executive surveys found topical relevance and sponsorship through the executive’s own network raised response, while follow-ups and personalization showed no significant effect.

Should We Offer an Incentive for Completing a Win-Loss Survey?

No study of win-loss surveys tests incentives. In executive surveys generally, Cycyota and Harrison found that topical relevance and sponsorship through the executive’s own network raised response, and incentives were rarely tested in the studies they reviewed. Treat an incentive as a courtesy and not a lever, and check buyer gift policies, since many employers restrict gifts from vendors.

Should Win-Loss Be Run In-House or by a Third Party?

A win-loss survey can generally be run in-house by a research or product marketing function, provided individual answers never reach the deal’s rep. Interviews benefit more from a neutral party, because a buyer speaking live is more guarded with the vendor. Many programs run the survey internally and use an outside interviewer for the subset.

Should Win-Loss Responses Be Anonymous?

Win-loss responses should usually be confidential and not anonymous. Identify each response to the research owner so it can be joined to CRM fields, and never show individual answers to the rep who owned the deal.

Is There a Benchmark for Win-Loss Survey Results?

There’s no published, sourced benchmark for win-loss survey results. Your benchmark is your own previous windows on the same instrument, outcome codes and size threshold, tracked alongside response rate by outcome and the rep-buyer agreement rate, as a brand tracking survey tracks its own waves.

What Is the Difference Between Win-Loss Analysis and Churn Analysis?

Win-loss analysis explains why buyers chose or rejected you during a sales evaluation. Churn analysis explains why customers leave after buying, so it surveys a different population and generally finds different causes.

Can a Win-Loss Survey Prove Why We Lose Deals?

A win-loss survey can’t prove why deals are lost, because won-versus-lost gaps are associations inside a self-selected sample. It identifies which reasons recur and which factors separate won from lost deals, and an experiment on the specific lever tests whether fixing it changes the outcome.

How Do I Analyze Win-Loss Survey Data?

Analyze win-loss survey data in a fixed order. Join survey responses to the CRM export, collapse to one record per deal, and report response rate by outcome. Then report each reason share with a Wilson interval, the won-versus-lost gap on each decision factor, and the rep-buyer agreement rate. Compare respondents with nonrespondents before reading any of it. The two prompts in this guide run each step.

What Response Rate Should I Expect from a Win-Loss Survey?

No source publishes a win-loss survey response rate. Academic surveys of executives reported 32 percent across 231 mailed studies, and lost buyers owe you nothing, so plan for less. Table D shows the invitations needed at 5, 10, 20 and 33 percent, and your own first window becomes the benchmark for every later one.

Does a Win-Loss Survey Work for Self-Serve or High-Volume Deals?

A win-loss survey fits sales-led deals where an opportunity record names the buying contacts. A self-serve signup has no opportunity or contact role to join, so a survey inside the product at cancellation or abandonment fits better. The deal-level logic in this guide applies only to the sales-led segment.

How Long Should a Win-Loss Survey Be?

A win-loss survey should typically take under 10 minutes, including two open-text questions. The instrument in this guide has eight questions, and the comparative factor rating is usually the longest block, so keep the factor list to six to eight items.

How Do I Connect Win-Loss Survey Responses to CRM Data?

Connect win-loss survey responses to CRM data by placing the CRM opportunity or contact ID in the survey’s contact list, then joining the survey export to a CRM export of every decided opportunity on that ID. Include nonresponding opportunities in the CRM export, because the nonresponse comparison needs the full denominator.