Deals are won by the number of people in the room, not by the talk-to-listen ratio everyone coaches. That is the short version of what came out of 133,068 closed deals and the 2.4 million sales calls behind them. Below are eleven findings strong enough to publish, three we are still working on, and four popular claims our own data refuses to support.
Most sales research you read is a survey of a few hundred people, or a vendor blog that starts “we analyzed 10,000 sales calls.” This one is different in one boring, important way: it is outcome data. Every deal in it is closed, so we know whether it was won or lost, and we can trace it back to what happened on the calls.
What is in the dataset?
The analysis ran across 116 companies selling B2B, from seed-stage teams to enterprise sales organizations. It covers the full record: calls, extracted next steps, scorecard evaluations, CRM loss reasons, and the questions sales teams typed into an AI about their own deals.
| The record | Volume |
|---|---|
| Companies | 116 |
| Calls analyzed | 2,442,525 |
| Closed deals | 278,233 (68,305 won, 209,928 lost) |
| Closed deals with a recorded call | 133,068 (the analyzable set) |
| Extracted next steps | 4,237,764, from 1,398,461 calls |
| Scorecard evaluations | 1,231,692, on 751,117 calls |
| Labeled loss reasons | 253,008, plus 268,661 in free text |
| Questions asked of the AI | 951,746, across 613,543 threads |
| Baseline win rate | 28.7% |
The honest byline is 133,068 closed deals and the 2.4 million calls behind them. For the question-level work, which needs full transcripts, it is a matched sample of 18,752 first calls, drawn evenly from won and lost deals.
How were the numbers guarded?
Two rules applied throughout, and they are the reason the eleven findings below are worth defending in public.
Nothing rests on one big customer. Every contrast is reported three ways: pooled across all deals, as the median of per-company rates, and as a count of how many companies individually reproduce the direction. Two very high-volume accounts with a consumer-shaped sales motion are excluded throughout, along with our own internal account, because their deal counts would have dominated every pooled figure.
Where “more activity” could explain a result, activity is held fixed. Deals that progress naturally accumulate calls and contacts, so a finding like “deals with more meetings win more” can be true and useless. Each headline contrast was re-run on the subset of deals with exactly one recorded call. The findings that survived that test are the ones we publish.
1. Single-threading is the funnel’s biggest self-inflicted wound
58.6% of closed deals never involved more than one person on the buyer’s side. Those deals won 24.3% of the time. Deals with five or more buyer-side contacts won 46.4%, which is 2.7x the odds.
| Buyer-side contacts | Win rate |
|---|---|
| 1 contact | 24.3% |
| 2 contacts | 28.7% |
| 3 to 4 contacts | 35.4% |
| 5 or more contacts | 46.4% |
101 of 111 companies show the same direction. The median per-company win rate is 23.7% single-threaded against 39.0% multi-threaded. Restricted to deals with exactly one recorded call, the ladder survives: 21.8% at one contact, 28.3% at two, 33.3% at three or more. So this is not simply “good deals have more meetings.” It is the widest, most consistent gap in the whole dataset, and it is the one most within a rep’s control. It is also why your champion has to be able to sell the deal without you in the room.
2. The next step that predicts a win has the buyer’s name on it, not a date
Deals where at least one agreed next step was owned by someone on the buyer’s side won 33.7% of the time. Where every action sat with the seller, 18.7%.
This is the most consistent result in the dataset: 38 of 39 qualifying companies agree, across 65,140 deals. Median per company is 30.2% against 16.5%. Restricted to deals whose next steps all came from a single call, it is 28.4% against 19.1%, and 33 of 39 companies still agree.
Due dates, the thing everyone coaches, give 31.6% against 21.0% pooled. But that gap collapses to 26.1% against 21.3% once call count is held fixed. A date helps. A buyer-side owner helps about twice as much.
For color: of 4.2 million extracted next steps, only 38.6% carry a due date and 33.1% are owned by the buyer. The most common opening word is “send” (64,237 times), then “follow” (51,167), then “review” and “schedule.” The median agreed next step in B2B sales is a rep promising to email something. That is a large part of why good deals stall quietly.
3. Talk-to-listen ratio predicts nothing, measured two independent ways
Reps who talked under 40% of a first call won 31.1%. Reps who talked 70% to 79% won 30.4%. Only the reps above 80% were meaningfully worse, at 26.2%.
| Seller share of talk, first call | Deals | Win rate |
|---|---|---|
| Under 40% | 15,289 | 31.1% |
| 40% to 49% | 13,865 | 28.8% |
| 50% to 59% | 21,459 | 28.8% |
| 60% to 69% | 24,527 | 29.6% |
| 70% to 79% | 19,960 | 30.4% |
| 80% or more | 36,436 | 26.2% |
We reproduced this with a different metric on a different sample. Word counts from 18,752 transcripts show buyer share of words is flat: 49.3% win rate under 20% of words, 51.1% at 20% to 34%, 50.3% at 35% to 49%, 50.0% above half. For scale, contact count spans 22 points of win rate. Talk ratio spans about five, and not monotonically.
One nuance does survive. The longest uninterrupted stretch of seller speech has a sweet spot. Under 150 words, 48.7%. At 150 to 299 words, 55.9%. At 600 to 999 words, 46.3%. Reps who cannot hold the floor for ninety seconds do about as badly as reps who will not stop.
4. It is not how much the rep asks, it is how much the buyer asks
Seller question count barely moves the needle. Buyer question count moves it monotonically. Buyers who asked 10 or more questions on the first call closed at 52.4%. Buyers who asked none or one closed at 44.0%.
| Buyer questions, first call | Win rate |
|---|---|
| 0 to 1 | 44.0% |
| 2 to 4 | 46.8% |
| 5 to 9 | 49.5% |
| 10 or more | 52.4% |
This sample of 18,752 first calls across 30 companies was drawn evenly from won and lost deals, so 50% is the neutral line, not 28.7%. Seller questions show a weak inverted U: 46.8% at 0 to 4 questions, 52.8% at 10 to 14, 50.2% at 30 or more. That peaks near the familiar “ask 11 to 14 questions” coaching advice, but it is worth about six points at most.
Conversational tempo points the same way. First calls with 35 or more speaker switches per ten minutes won 53.5%, against 44.9% at 10 to 19 switches. The median first call has 16 seller questions and 11 buyer questions. Your job on call one is not to get through your list. It is to get them asking.
5. Habits, not tenure, explain the gap between colleagues
Inside the same company, selling the same product at the same price, the top-quartile rep wins 1.85x as often as the bottom-quartile rep. The median spread between best and worst rep is 41 points of win rate.
This is the cleanest quasi-causal comparison available in the data: same company, same product, same market, same pricing, different behavior. It is reported as the median across the 96 companies that had at least four reps with 25 or more closed deals each.
And the habit that separates them is multithreading. Across those 96 companies, reps in the top half for bringing extra buyer-side contacts into calls won 36.6%. The bottom half won 24.9%. Twelve points of win rate, from one repeatable behavior, which is exactly the kind of thing a talk track built from your own top performers can transfer.
A null result worth knowing: time on platform showed no ramp effect at all. 31.2% in the first 89 days, 27.9% from 90 to 179 days, 29.1% from 180 to 364 days, 32.5% after a year. Whatever faster ramp looks like, it does not show up as a tenure curve in this data.
6. Talking about paperwork on call one is a buying signal
When the paper process came up on the first call, meaning legal, procurement, security review, or the signature path, the deal won 38.4% of the time against 27.8% when it did not. Only about 9% of first calls get there.
| Captured on the first call | Win rate with | Win rate without | Companies agreeing |
|---|---|---|---|
| Timelines | 43.1% | 27.4% | 16 of 19 |
| Paper process (legal, procurement) | 38.4% | 27.8% | 35 of 37 |
| Authority | 37.5% | 27.6% | 10 of 14 |
| Budget | 37.0% | 27.2% | 34 of 48 |
| Compelling event | 31.8% | 28.1% | 43 of 47 |
| Quantified impact | 30.5% | 28.0% | 35 of 41 |
The paper-process result covers 11,631 first calls, and the median per company is 45.1% against 26.2%. Read the table as a whole and the pattern is clear: the predictive first-call topics are the boring commercial mechanics, which are time, money, authority, and process. Not the emotive ones.
7. The qualification moves everyone drills are the ones that do not predict
Identifying pain on the first call: 28.6% win rate with it, 28.8% without. Decision criteria: 28.6% against 28.8%. Metrics: 28.1% against 28.9%. Economic buyer: 29.0% against 28.6%.
Four of the most-coached qualification moves in B2B sales showed no effect at all. Each of these topics appears on between 26,000 and 52,000 first calls, so these are well-powered nulls, not thin samples where the effect is hiding in the noise.
Paired with finding 6, the story writes itself. The framework is not wrong. The drilling is aimed at the wrong half of it. Teams rehearse pain and metrics, which do not move the number, and skip timelines and procurement, which do. If you run MEDDIC or a similar framework, this is where to rebalance what your reps get coached on.
8. Depth of discovery pays, then plateaus fast
First calls where we captured nothing structured won 19.9%. Capture three or four elements and it jumps to 30.5%. Capture seven or more and it is 30.4%, so there is no further gain.
| Discovery elements captured | Win rate |
|---|---|
| 0 elements | 19.9% |
| 1 to 2 elements | 21.9% |
| 3 to 4 elements | 30.5% |
| 5 to 6 elements | 28.6% |
| 7 or more elements | 30.4% |
Holding call count fixed at one, the same shape holds: 16.1% at zero elements, 23.9% at three or four, 25.6% at seven or more. The usable message for a rep is that you need three or four real things, not a checklist of eight. The marginal fifth question earns nothing, and it costs you the conversational tempo from finding 4.
9. Competitors make deals worse and almost never decide the loss
Deals where a competitor came up on the first call won 26.7%, against 29.8% when none did. 36 of 49 companies agree. So a competitive deal is a slightly worse deal.
But competitors almost never end up as the reason for the loss. Across 253,008 labeled loss reasons, “lost to a competitor” was 2.8% of all labels and the number-one loss reason for just 2 of 102 companies.
| Loss family | Top reason for (CRM labels) | Top reason for (free text) |
|---|---|---|
| Product or feature gap | 42 of 102 | 70 of 139 |
| Price or budget | 26 of 102 | 58 of 139 |
| Timing, not ready | 22 of 102 | 9 of 139 |
| Internal change, no champion | 8 of 102 | 1 of 139 |
| Lost to a competitor | 2 of 102 | 1 of 139 |
Two independent loss datasets, one from structured CRM fields and one from free-text notes, agree on the ranking. Price is the loudest loss reason, product gap is the most common, and competitors are noise.
One methodological aside that matters if you benchmark yourself against this: price or budget is 35.5% of pooled labels but only 19.8% of the median company’s. A handful of high-volume accounts make price look bigger than it is across the industry, which is worth remembering before you conclude your team has a discounting problem rather than a price-holding conversation problem.
10. Managers score reps on the skills that do not decide deals
Across 1,231,692 scorecard evaluations on 751,117 calls, reps score highest on demoing and objection handling. They score lowest on recapping, competitive positioning, qualification, and stakeholder mapping.
| Skill family | Mean score (0 to 1) | Evaluations | Companies |
|---|---|---|---|
| Summarizing and recap | 0.416 | 70,221 | 113 |
| Competition | 0.455 | 10,254 | 21 |
| Qualification framework | 0.457 | 312,668 | 165 |
| Stakeholders and champion | 0.513 | 31,772 | 49 |
| Discovery and questioning | 0.543 | 274,313 | 175 |
| Listening and talk time | 0.631 | 457,746 | 169 |
| Next steps, closing the call | 0.642 | 775,591 | 187 |
| Value and solution framing | 0.672 | 549,980 | 192 |
| Objection handling | 0.711 | 479,776 | 182 |
| Demo and product | 0.730 | 477,344 | 152 |
Scores are ordinal levels defined per scorecard, so each one is normalized to a 0 to 1 scale against its own scorecard’s level count before grouping by skill name. The mean overall result is 62.0%.
Now put this next to findings 1 and 3. Reps score 0.631 on the talk-time metric we can show does not predict outcomes, and 0.513 on the stakeholder mapping that does. The measurement system and the outcome data are pointing in different directions, which is a solvable problem once you look at what your call recordings are actually scoring.
11. A slipped close date tells you almost nothing
Deals whose close date never moved won 27.0%. Slipped once: 32.0%. Twice: 29.8%. Three or more times: 30.9%. Among deals that slipped at all, the median slip is 61 days.
| Close date behavior | Win rate |
|---|---|
| Never slipped | 27.0% |
| Slipped once | 32.0% |
| Slipped twice | 29.8% |
| Slipped 3 or more times | 30.9% |
Across 119,395 closed deals in 116 companies, the CRM field every pipeline review runs on has no predictive power. If anything, a date that never moved is often a deal nobody was working.
The positive counterpart is far more useful. Deals where the amount was revised at least once won 54.7%, against 22.2% where it never changed. Someone negotiating your number is worth more than a date on a slide, and it is a large part of why most sales forecasts are wrong.
What are we still working on?
Three results we believe, with a specific reason not to put a headline on them yet.
Longer first calls close better, but the curve is not clean. Under ten minutes, 23.4%. Sixty minutes or more, 36.6%. But the 30 to 44 minute bucket dips to 25.9%, which is almost certainly a call-type confound: ten-minute calls include no-shows and qualification screens. This needs segmenting by call type before it means anything.
Discussing the decision process early is mildly negative, at 27.4% against 29.5%. That sits oddly beside paper process being strongly positive in finding 6. Interesting if it survives scrutiny, embarrassing if it does not. We want to know what each classification is actually catching first.
What sales teams ask an AI about their own calls. 951,746 questions across 613,543 threads, 71% of them anchored to a specific call. The most common ask is not insight, it is memory. 15.8% are about next steps or follow-ups and 10.7% ask for a summary, against 2.1% asking whether the deal will close and 2.0% about stakeholders. The median question is ten words and starts with “what” (162,279 times) or “write” (56,704). Only 29.8% of threads get a second human turn.
That last one is the most interesting dataset we have, and it is the one we are least ready to publish. Keyword buckets left 59.6% of questions unclassified, so those shares are floors, not a distribution. It needs proper classification before it carries a headline.
Four claims the data refuses to support
Every research project generates a list of things you hoped were true. Publishing that list is the part that makes the rest of it credible.
“New reps who did this one thing ramped 40% faster.” No ramp effect is detectable at all. Time on platform is flat across every tenure band, as shown in finding 5. There is no version of this claim our data supports.
“Single-threaded deals die 3x more often.” The real figure is 2.7x the odds of winning between one contact and five or more, or 24.3% against 46.4% on win rate. That is not the same as 3x the death rate, and the true number is still a very good stat.
“Reps do not complete their next steps.” Only 3.7% of extracted tasks are marked complete. But that measures whether people tick boxes in a product, not whether the work happened. It is a product-usage number wearing a sales-insight costume.
“The exact questions that move deals from stage 2 to stage 3.” We can see stage transitions and we can see call content, but nobody has yet linked specific question phrasings to a transition. Feasible, not done. We will not imply otherwise.
How should you read these numbers?
As observational contrasts, not experiments. Nobody randomized reps into single-threading. Deals that progress accumulate calls, contacts, and next steps, so any raw contrast risks measuring progress rather than causing it.
We controlled for that where we could, using single-call subsets, within-company rep comparisons, and per-company medians rather than pooled averages alone. Those controls are exactly why the eleven findings above sit where they do, and why the rest sit in the “still working on it” pile. Where a result only appeared in the pooled figure and vanished when activity was held fixed, we have said so in the finding itself.
All figures are aggregated and anonymized across companies. Nothing here identifies a customer, a rep, or a deal, and nothing published from this work will. The data was computed in August 2026.
What would we do with this on Monday?
Four changes, ranked by how much the numbers say they are worth.
- Count contacts, not calls, in every pipeline review. A deal with one buyer-side contact is at 24.3%, whatever the close date says. Make single-threading a stage gate, not a coaching note.
- Rewrite what a next step means. An agreed action with the buyer’s name on it, then a date. Not a rep promising to send something.
- Move first-call discovery toward timelines, budget, authority, and the paper process, and stop drilling pain and metrics as though they decide the outcome. Three or four captured elements is the target, not eight.
- Re-weight your scorecard. If talk time carries more weight than stakeholder mapping, your measurement system is arguing with your outcome data.
Where the calls come in
None of this is knowable from a CRM. The contact count that predicts the win, the owner of the next step, the moment procurement first came up: all of it happens on calls, and almost none of it survives into a record anyone reviews.
That is the whole reason this analysis was possible. Airspeed captures the call, extracts the next steps with their owners, keeps the deal view current on who is actually in the room, and puts coaching against the behaviors that move win rate rather than the ones that are easy to score. For sales leaders, that turns a benchmark like this one into something you can check against your own pipeline this week.
We will keep publishing what we find, including the parts that did not work.
Book a demo to see what your own call record says about the deals you are working.
Revenue Decoded is where we pull apart how revenue actually gets made, the mechanics, not the theatre.
Frequently asked questions
What is the single biggest predictor of winning a B2B deal?
The number of people on the buyer's side. Across 133,068 closed deals, deals with one buyer-side contact won 24.3% of the time and deals with five or more won 46.4%. That is 2.7x the odds, and 101 of 111 companies in the set showed the same direction. It holds even when you compare deals with exactly one recorded call, so it is not just a side effect of good deals having more meetings.
Does talk-to-listen ratio predict whether a sales deal closes?
No. Reps who talked under 40% of a first call won 31.1%. Reps who talked 70% to 79% won 30.4%. Only reps above 80% were meaningfully worse, at 26.2%. A second measure on a separate sample of 18,752 transcripts found buyer share of words was flat across every band. Contact count spans 22 points of win rate. Talk ratio spans about five, and not cleanly.
What kind of next step actually predicts a win?
One with the buyer's name on it. Deals where at least one agreed next step was owned by someone on the buyer's side won 33.7%, against 18.7% where every action sat with the seller. Due dates help too, but about half as much once you hold call count fixed. Of 4.2 million extracted next steps, only 33.1% are owned by the buyer.
How often do B2B deals get lost to a competitor?
Rarely. Across 253,008 labeled loss reasons, "lost to a competitor" was 2.8% of all labels and the number-one loss reason for only 2 of 102 companies. Product or feature gap was the most common top reason, at 42 of 102 companies, and price or budget the loudest. Two independent loss datasets agree on the ranking.
How was this sales research conducted?
We analyzed 278,233 closed deals across 116 companies on Airspeed, of which 133,068 had at least one recorded call and formed the analyzable set. Every contrast is reported pooled, as a median of per-company rates, and with a count of how many companies reproduce the direction. Headline contrasts were re-run on deals with exactly one recorded call so that "more activity" could not explain the result. All figures are aggregated and anonymized, and no customer is identifiable.
