What is a win-loss analysis, and why start in the CRM?
Win-loss analysis is a review of closed deals, won and lost, that looks for the patterns behind the outcomes. It asks where deals stall, which accounts convert, and why buyers chose you, a competitor or nobody. Its purpose is pattern-finding across many deals, not a post-mortem on one.
Most guides start with buyer interviews, run by product marketing. That’s the right end point. But with three years of deals in HubSpot or Attio, you can get most of the way from data you already own. That takes days, not months, and comes before you call a single buyer. It’s also the first thing the GTM engineering work of mapping your market depends on. If you don’t know who you win, every list you build after that is a guess. The TAM, SAM and SOM guide shows how those winners become a sized market.
This guide covers the “why and where” questions. For turning the winners into a target definition, see the companion guide on building your ICP from closed-won data.
Step 1: Export every closed deal, wins included
Pull all deals that reached closed-won or closed-lost in the last two to three years. Including the wins matters: Clozd notesSource 1 that most CRMs hold little data on why you win, so the wins are where the missing half of the story sits.
Fields to export per deal:
| Field | Why you need it | Where it usually lives (HubSpot names) |
|---|---|---|
| Create date, close date | Cycle length; which era of your pitch the deal belongs to | Create date, Close date, Days to close |
| Stage reached before closing | Where lost deals died | Deal stage history |
| Amount (ZAR) | Win rate by value band | Amount |
| Source | Inbound, outbound, referral, partner | Original source or a custom field |
| Owner | Rep-level patterns (handle with care) | Deal owner |
| Company industry, employee band, region | Segment cuts | Associated company properties |
| Closed lost reason and notes | The starting point for step 4 | Closed lost reason, notes, call logs |
| SPICED fields, if you capture them | Whether a pain, impact or critical event was ever found | Custom properties |
Expect gaps. Salesforce’s 2024 State of Sales surveySource 2 of 5,500 sales professionals found that only 35% completely trust the accuracy of their organisation’s data (industry data). The respondents were in North America, Latin America, Asia-Pacific and Europe. Record what share of each field is blank, because a blank field is a finding too.
Step 2: Find where deals die (win rate by stage)
Start with three numbers: the overall win rate (won ÷ all closed), the conversion between each stage, and median cycle length for wins versus losses. Measure only deals that are closed; open deals flatter the early stages.
Winning by Design’s Bowtie modelSource 3 (glossary) splits the buying journey into stages (Awareness, Education, Selection, Commit, then Onboard, Impact and Expansion after the sale). Win-loss work sits on the left side of the Bowtie: the question is which stage transition leaks most. A team that loses most deals before a proposal has a qualification or discovery problem. A team that loses most after the proposal has a value, pricing or decision-process problem. Those need different fixes.
For context, the Ebsta x Pavilion 2025 GTM BenchmarksSource 4, drawn from 655,000 opportunities, put the new-logo win rate at 19% (report overview, page 5). That’s industry data from a vendor dataset, with no South African breakdown. Use it as a sense check, not a target: your definition of an opportunity changes the number more than your sales skill does.
Fictional company, for illustration only. Quorvel Software sells rostering software to South African businesses. Between October 2023 and September 2026 it closed 412 opportunities and won 88: a 21% win rate. Of the 412, 170 reached the proposal stage, and 88 of those were won (52%). So 242 of the 324 losses (75%) died before a proposal. Median cycle: 74 days for wins, 118 days for losses. The leak is early, which points at who enters the pipeline and how discovery runs, not at pricing.
Of Quorvel's 412 closed deals, 170 reached a proposal and 88 were won (21% overall, 52% of proposals). 242 of the 324 losses (75%) died before a proposal, so the leak is early. Example data: Quorvel Software is fictional. About this data: Quorvel Software (fictional), for illustration only.
Step 3: Cut win rate by segment, source and rep, with error bars
Now split the win rate by industry, company size, source, value band and rep. This is where most CRM win-loss reports mislead, because small groups produce big-looking numbers.
Add a range to every rate. The NIST Engineering Statistics HandbookSource 5 favours the Wilson interval for proportions over the simple “plus or minus” formula, which can produce impossible values when samples are small. With wins in column A and closed deals in column B, these two spreadsheet formulas give the lower and upper end of a 95% range:
Lower: =(A2/B2 + 1.96^2/(2*B2) - 1.96*SQRT(A2/B2*(1-A2/B2)/B2 + 1.96^2/(4*B2^2))) / (1 + 1.96^2/B2)
Upper: =(A2/B2 + 1.96^2/(2*B2) + 1.96*SQRT(A2/B2*(1-A2/B2)/B2 + 1.96^2/(4*B2^2))) / (1 + 1.96^2/B2)The rule for reading it is simple: if a segment’s range overlaps your overall win rate, you haven’t proven it’s different.
Quorvel Software, win rate by industry (95% Wilson interval).
| Segment | Closed deals | Won | Win rate | Plausible range |
|---|---|---|---|---|
| Logistics and transport | 64 | 22 | 34% | 24%–47% |
| Retail | 58 | 9 | 16% | 8%–27% |
| Financial services | 18 | 7 | 39% | 20%–61% |
| Everything else | 272 | 50 | 18% | 14%–23% |
| All deals | 412 | 88 | 21% | 18%–26% |
Reading it: logistics is a real strength (its range sits above 21%). Financial services looks like the best segment, but 18 deals can’t show it; the range includes 21%, so it goes on a watch list. Retail is weaker, but its range also overlaps the overall rate, so it’s a question for step 5, not a decision.
Logistics (34%, range 24%–47%) is the only segment whose range sits clear of the overall 21% (18%–26%). Financial services looks best at 39%, but 18 deals give a range of 20%–61%, so it goes on a watch list. Example data: Quorvel Software is fictional. About this data: Quorvel Software (fictional), for illustration only.
Two cautions on the other cuts. Rep-level win rates often reflect the mix of deals each rep was given as much as skill. Ebsta’s report shows a related problem: in its example seller chart, a rep with a high win rate in one segment receives the fewest of that segment’s deals. And source cuts need the same discipline: 10 referral deals at 60% is a hint, not a channel strategy.
The segment cut is also where win-loss hands over to targeting. Which attributes the winners share, and how often those attributes appear in your wider market, is the ICP question. The closed-won ICP guide covers that method, including which patterns are noise.
Timing may be a bigger gap than targeting. Take the 3-minute signal gap scorecard to see where your team stands.
Step 4: Recode the loss reasons before you count them
The closed-lost dropdown is the weakest data in most CRMs. In Clozd’s comparison of 1,000 closed-lost dealsSource 1 against buyer interviews, the CRM reason matched the buyer’s account only 15% of the time. And 44% of logged reasons were outcomes rather than reasons (vendor data from a win-loss interview company, so read it as directional). “Went dark”, “timing” and “no response” describe what happened, not why.
A second pattern to look for is the deal lost to nobody. Research by Matthew Dixon and Ted McKennaSource 6 (published in The JOLT Effect, 2022), based on 2.5 million recorded sales conversations, found that 40–60% of lost deals ended in no decision rather than a competitor win. If your dropdown has no “no decision / stayed with status quo” option, those deals are hiding under “price” and “timing”.
How to recode, in order:
- Split outcomes from reasons. Move “went dark”, “timing” and “no response” into an outcome field. They don’t count as reasons.
- Read the notes on the most recent 50–60 losses. Emails, call notes and the last activity usually show the real reason, or show that nobody knows.
- Map each deal to a short list of reasons: no decision, lost to a competitor (named), price or budget, not a fit (should have been disqualified), product gap (named), unknown. For competitor losses, record why the buyer chose them, not just the name. “Unknown” is an honest answer; it stops guesses being counted as facts.
- Check SPICED coverage. For each deal, was a Pain, an Impact and a Critical Event ever recorded? SPICED is Winning by Design’s discovery frameworkSource 7 (Situation, Pain, Impact, Critical Event, Decision). Ebsta’s 2025 reportSource 4 found that 76% of deals run by lower-performing sellers lacked critical events (industry data). If lost deals rarely had one, the problem is discovery, not the close.
Quorvel Software’s 324 losses, as logged: price 41%, went dark 17%, timing 12%, chose a competitor 11%, blank or “other” 19%. So 29% of “reasons” were outcomes. The notes on the 60 most recent losses tell a different story. Recoded, those 60 split: no decision 40%, price or budget 18%, lost to a competitor 17%, not a fit 15%, unknown 10%. Price fell from the top reason to the second. A critical event was recorded on 125 of the 412 deals. Those won 43% of the time (54 of 125), against 12% for the 287 deals without one (34 of 287). The ranges, 35%–52% and 9%–16%, don’t overlap, so the gap is real.
As logged, price led at 41% of 324 losses, and 29% of the "reasons" were outcomes (went dark 17%, timing 12%). Recoded from the notes on the 60 most recent losses, no decision leads at 40% and price or budget falls to 18%. Example data: Quorvel Software is fictional. About this data: Quorvel Software (fictional), for illustration only.
Step 5: Ask buyers, then make three decisions
Data tells you where and how often. Buyers tell you why. Call five to ten recent buyers, won and lost. Ask what they were trying to fix, what else they considered, what nearly stopped the deal and what made the final call. Keep it to 20 minutes and don’t sell.
Who asks matters. Clozd’s 2025 State of Win-Loss reportSource 8 surveyed more than 1,000 professionals (vendor research). It found that 70% of companies using a third-party provider were satisfied with their win-loss feedback, against 34% of companies running programmes internally. Clozd sells this service, so weigh the gap accordingly. If you run the calls yourself, have someone other than the deal owner make them.
Then write a one-page readout that ends in decisions, each with an owner and a date:
| If the analysis shows… | Decision | Usual owner |
|---|---|---|
| Most losses die before the proposal | Tighten entry criteria and discovery; require a recorded pain and critical event before a proposal | Head of Sales |
| One segment’s win rate is clearly above the rest | Shift outbound and list-building effort towards it; define it properly as an ICP | Head of Sales with RevOps |
| “No decision” is the top real reason | Coach on decision risk and the cost of doing nothing; qualify the critical event harder | Sales manager |
| Loss reasons are mostly blank or outcomes | Make the reason a required dropdown at closed-lost, with “no decision” and “unknown” as options | RevOps or CRM admin |
| A named competitor wins a specific deal type | Build one battlecard for that situation, not a generic one | Product marketing or Head of Sales |
In HubSpot, “Closed lost reason” is a default deal propertySource 9 with options you can customise. The pipeline settingsSource 10 let you mark a property as required when a deal enters a given stage (conditional stage properties; you need permission to edit property settings). If you use another CRM, look for the equivalent stage-level required field.
This is the Evidence layer of the HeyThere GTM System: the step that establishes who you actually win before anyone builds a market list. If you’d rather have it done on your data, that’s what a GTM Data Audit covers.
When a CRM win-loss analysis won’t tell you much
- Under about a year of deal history, or a few dozen closed deals. The ranges will be too wide to act on. Interview buyers instead.
- Founder-led sales. If one person closed most deals, you’re analysing that person’s style, not a repeatable motion.
- A recent change in product, price or market. Deals from before the change describe a different company. Split the data at the change.
- No agreed stage definitions. If “proposal” means different things to different reps, stage conversion rates are noise. Fix the definitions first and measure from that date.