Key findings
- The demo request is only sometimes a good signal15×the lift from a demo request at one company vs 0.8× at another and everything in between
- Some signals have opposite correlations21 of 108signals went clearly up at some companies and clearly down at others.
- Few signals can predict won opps, many can predict pipeline74%of tests of contact engagement were clear for becoming an opportunity. 7% were clear for winning.
Figure 1
All 108 signal results at a glance: most point one way, and 21 went up at some companies and down at others
Each square is one signal, tested for one outcome at three or more companies. Hover for the signal and its counts; click to see it in the appendix.
- Higher wherever clear 50
- Up at some companies, down at others 21
- Lower wherever clear 15
- No clear difference anywhere 22
"Higher wherever clear": a clear difference at one or more companies, and every clear one higher. Upside's analysis of 10 companies' data, November 2026. Each company's data was anonymized and aggregated on its own before pooling.
A word from the author
I went into this with three hunches. Signals are not the same for every company. Some signals that look like they should matter don't. And every company has a few that make no sense until you ask the people who run the process.
Everyone in B2B carries a list of buying signals: the pricing page view, the demo request, the second buyer on the deal. Most of those lists were built at one company and carried to the next, and nobody checks whether they stay consistent or are really a signal. We now had a way to check. Upside connects to each customer's CRM, email, calendar, calls and web data, and our signal library gives every signal one definition, one time window and, for calls, one yes/no question. So we ran the whole library, the same way, on ten companies' own data, and compared. If you are interested in running this on your own data, let us know, we offer free trials for qualified companies and implementation is just connecting your systems.
The scale: 102 live signals, 19,459 new business deals, 205,036 contacts, 880 first calls read blind by AI, and 1,325 company-level results. All data was anonymized and aggregated before being used in this report.
Here's what held, what flipped, and what never showed up at all.
The direction of a signal travels. Its strength does not.
Of 305 clear results with an expected direction, 285 (93%) pointed the way the library expects. Strength is another matter: 21 of the 108 signals tested at three or more companies went clearly up at some companies and clearly down at others, and even the signals that always pointed up ranged more than tenfold between companies.
Figure 2
Same signal, very different strength: up to 15× at one company, below 1× at another
Each dot is one of the 10 companies: how many times as likely people, deals or accounts with the signal were to reach the outcome as those without it. Left of Same means less likely.
- Clearly more likely at that company
- Clearly less likely
- Could be chance
- Typical company
- Same = no difference
Upside's analysis of 10 companies' data, November 2026. Each company's data was anonymized and aggregated on its own before pooling. Source: the study dataset.
Demo requests, replies to email and a fast sales follow-up stay above 1× wherever they are clear, but their size ranges more than tenfold. The rows lower down cross the line: whether a C-level contact, a personal email address or a two-way first conversation helps depends on the company. In our experience the reason is usually local: what a form is used for, which meetings sync, who gets imported as a contact.
Engagement predicts pipeline, not wins.
Seventeen contact behaviours were clear for becoming an opportunity in 116 of 156 company tests (74%). The same behaviours were clear for winning in 7 of 107 (7%). The things a lead-scoring model rewards tell you who will get into pipeline. They say almost nothing about who will close.
Figure 3
Clear for pipeline at most companies, almost never for wins
The same 17 contact behaviours at the typical company. Left: how many times as likely the contact was to become an opportunity. Right: how many times as likely the deals those contacts joined were to be won. The percentage beside each dot is the share of companies that could test it where the result was clear: 74% for becoming an opportunity (116 of 156 company tests), 7% for winning (7 of 107). Hover a dot for the counts.
- More likely at the typical company
- Less likely
- Same = no difference
Upside's analysis of 10 companies' data, November 2026. Contact behaviours tested on both measures at five or more companies.
Win rates rest on fewer people, because only some of these contacts reach a closed deal. But the win-rate results cluster around 1×, not just short of significance: someone who wrote back was 5.0 times as likely to become an opportunity at the typical company, and that opportunity was about as likely to win (1.1×).
Some signals the library marks must-test barely showed up at all:
- The seller reached out first. Expected to lower the win rate; no clear difference at 8 of 10 companies, with a typical 0.93×.
- Took a meeting. Clear for becoming an opportunity at 5 of 10 companies, and for winning at none of 8. Synced calendars bring in customer, support and delivery meetings alongside sales ones.
- Two or more buyer people on the first call. Clear at 3 of 8.
- Registered for an event or webinar but did not attend. Expected to be a warning sign; clear at only 4 of 9.
- Clicked a marketing email. No clear difference in win rate at any of the 5 companies that could test it.
Seventeen signals held up at seven or more companies.
These were clear in the expected direction at most of the companies that could test them, and never clearly the other way. They split by what they predict. What a contact does early predicts who becomes an opportunity, with lifts of 2× to 7× at the typical company. What the buying group does in the first month after Discovery predicts who wins, with lifts of about 2×.
Figure 4
Seventeen signals that held up at seven or more companies
Clear in the expected direction at most of the companies that could test them, and never clearly the other way. Each row runs from the lowest company to the highest; the mark is the typical company. Hover a row for the counts.
- Lowest company
- Highest company
- Orange = less likely (below Same)
- Typical company
- Same = no difference
Upside's analysis of 10 companies' data, November 2026. "Sales followed up within 7 days" is the seller's choice, so it says as much about whom reps pick as about the buyer.
The strongest results are not really signals.
The biggest numbers in the study are about how data gets recorded, not about buyers. Three stand out because they are large and consistent across companies:
- Buyer gone quiet for 30+ days on an open deal. Lower at 7 of 9 companies, to a tenth of the win rate at the typical company (0.09×). It is read back from the close date, and lost deals are often closed in clean-ups long after the buyer went silent. The first-month version, a 14-day silence after Discovery that does not depend on the close, gives a far more believable 0.34×.
- Two or more buyer contacts on the deal. Higher at 7 of 10 companies, up to 15.6×, and lower at one. Contact roles keep only their latest value and are added as a deal moves forward, so winning deals collect more of them.
- Job titles. "No job title on record" was lower at 9 of 10 companies (0.27×), and "Works in the core buying function" higher at 10 of 10 (2.2×). Titles are often filled in when someone joins a deal, so part of the signal is the deal itself.
Beyond these three, the agents that ran each company flagged 79 results to confirm with the company before anyone acts on them, about eight per company. Most fall into three kinds: a field that keeps only today's value, a signal read back from the close, and a stand-in that measures something else, such as a booking link used for onboarding or a contact form that also takes support requests.
What we found critical
These are the steps that mattered most while running the library across ten companies. Most hold for any team, with or without Upside.
A signal list is a set of hypotheses, not a scoring model
The same signal ran from 0.8× to 15× depending on the company, and 21 of 108 flipped direction. We gave every signal one definition and tested it on each company's own closed deals before anything was shown. The tradeoff: a company with 50 won deals can only call its biggest signals, and testing needs data most teams have not connected yet.
Test every signal on your own closed deals before you route, score or forecast on it, and keep only the ones that come out clear.
Pipeline signals and win signals are different lists
Seventeen contact behaviours were clear for becoming an opportunity in 74% of tests and for winning in 7%. We kept the two outcomes apart on every signal: a person's first 30 days against the win rate of the deals those people joined. The tradeoff: win rates rest on fewer people, so most win signals need a few hundred closed deals before they can be called.
Score leads on engagement, and forecast deals on what the buying group does in the first month after Discovery.
The biggest numbers came from how the data is recorded
"Buyer gone quiet for 30+ days" cut the win rate to 0.09× and "two or more buyer contacts" lifted it up to 15.6×; both read fields that change as a deal moves. We flagged every signal that reads a today-only field or counts back from the close date. The tradeoff: the honest version of a signal is usually weaker. A 14-day silence read forward from Discovery gives 0.34×, not 0.09×.
Read signals forward from a fixed date, and treat any ratio above 5× as a question about the data before it is a finding.
Every surprise had a local explanation
The agents flagged 79 results, about eight per company, that make no sense without asking the company: a booking link used for onboarding, a contact form that also takes support requests, a lead source that means something else.
Before acting on a surprising signal, ask the person who owns the form, the campaign or the field what it is actually used for.
Definitions bend to the company, so write them down
What counts as new business, which stage means qualified and who counts as an employee differed at every company; one company's window was six months after a CRM migration. We recorded every such choice as a judgement call, with the alternative and what it changes. The tradeoff: the numbers are less comparable across companies than they look.
Keep one written list of definitions per company, and re-test when a CRM migration or a new data source changes what a field means.
How we measured
- Companies
- 10 B2B companies using Upside as of 5 November 2026: 9 sales-led and 1 product-led with a sales team; 9 on Salesforce and 1 on HubSpot. Each company's data was anonymized and aggregated on its own before anything was pooled, and no figure in this report rests on fewer than 3 companies.
- Signals
- The 102 live signals in Upside's signal library. Each has one definition, one time window, default thresholds and, for calls, the exact yes/no question an AI reader answers. Where a company's data forced a change, the change is written down as a judgement call.
- Deals
- New business opportunities created in a window of about 18 months per company (six at one company, after a CRM migration), sorted into won, lost after qualifying, and never qualified. Open deals are counted but kept out of rates. A deal signal compares the win rate of closed deals with the signal against those without it.
- Contacts
- People whose first engaged touch fell in a 12-month window, with no engagement in the year before and not already on a deal. Signals are read in their first 30 days. The outcome is joining a new business opportunity within 180 days (30 days at one company whose web and form data started recently).
- Engaged touches
- Meetings attended, emails the buyer sent, clicks, identified web visits, form fills and campaign actions. Never emails the seller sent, opens, deliveries or invitations not attended.
- Accounts
- Account-months. The outcome is a new opportunity at the account within 90 days.
- First calls
- A blind AI read of first-call transcripts against the library's questions, on a fixed random sample of up to about 45 calls per outcome group. Readers never saw the outcome, and transcripts were deleted after the read.
- Clear or chance
- A deal result is clear when a two-sided Fisher exact test gives p under 0.05; a contact result when the two 95% ranges do not overlap. Fewer than 30 on either side is too few to call.
- Expected direction
- Each library signal states the direction it usually goes: positive, warning sign, varies by business or usually flat. A clear result that way is as expected; the other way is opposite.
- Typical company
- The median of the companies' values. A few large companies dominate any all-company total, so every ratio and share is computed per company first.
- Who ran it
- One Claude agent per company, connected only to that company's Upside workspace and writing aggregates only. No name, ID, email or amount left a company's workspace.
The study at a glance
- Companies
- 10 B2B companies; 9 sales-led, 1 product-led with a sales team; 9 on Salesforce, 1 on HubSpot
- Deals
- 19,459 New Business opportunities created between July 2024 and October 2026: 2,143 won, 9,041 lost after qualifying, 5,829 never qualified, 2,446 still open
- Deal window
- About 18 months at most companies (6 to 28)
- Win rate
- 12% of closed deals at the typical company
- Contacts
- 205,036 people in the contact cohorts, of whom 9,528 became part of an opportunity
- Accounts
- 440,688 account-months
Caveats and limits
In our experience these numbers are best read as directional. They show what came before an outcome, not what caused it.
- A convenience sample. The companies are Upside customers, almost all sales-led B2B companies on Salesforce. They are not a sample of the market.
- Many tests. Each company tested around a hundred signal results. At p under 0.05, a few clear results per company will be chance. We lean on patterns across companies, not on any single result.
- Fields that keep only today's value. Job titles, contact roles, deal owners and intent stages are overwritten as a deal moves. A title filled in when someone joins a deal can look like a signal of the deal. The library flags these signals, and so do we.
- Signals read back from the close. "Buyer gone quiet" is measured from the close date. Lost deals are often closed in clean-ups long after the buyer went silent, so the signal partly describes how deals are closed, not how buyers behave.
- Definitions bend to the data. What counts as New Business, and which stage counts as qualified, differs by company. Windows differ too: one company's deal window is six months, after a CRM migration. Every such choice was written down per company as a judgement call.
- Not possible is not the same as doesn't matter. Some companies have no call recordings, no ad data or few identified web visitors. Those signals could not be tested there.
- Small samples for calls. First-call signals rest on AI reads of a sample of calls. Many are too few to call at any one company.
Appendix: every signal tested at three or more companies
108 signal results, each counted across the companies that had enough data to compare it. "Typical company" is the median of the companies' ratios, left blank where fewer than 3 companies had a ratio that could be shown.
Figure 5
Pipeline signals lean one way; win signals mostly sit in the grey middle
108 signal results: 374 company results were clear, 345 showed no clear difference.
- Clearly higher
- Clearly lower
- No clear difference
- Each bar counts the companies that could test the signal, centred on no clear difference. Hover a bar for the counts.
| Signal | Typical company | Companies that could test it |
|---|---|---|
| Becoming an opportunity · 38 signals | ||
| Active in 2 or more channels, or several distinct signals | 3.88× | |
| Active on 3 or more days | 2.23× | |
| C-level or founder | 1.33× | |
| Came from a bought or enriched list | 1.21× | |
| Director, VP or head-of level | 1.18× | |
| First engaged touch was a two-way conversation | 1.44× | |
| No job title on record | 0.27× | |
| Personal or school email address | 0.17× | |
| Record created at the first signal | 1.27× | |
| Sales followed up within 7 days | 6.39× | |
| Took a meeting | 1.48× | |
| Works in the core buying function | 2.19× | |
| Wrote or replied to an email | 4.96× | |
| Arrived through brand: direct visit or branded search | 1.53× | |
| Attended an in-person event, booth or dinner | 0.9× | |
| Demo request or contact-sales form | 7.02× | |
| Registered for an event or webinar but did not attend | 0.53× | |
| Attended a webinar live | 0.9× | |
| Came back to the website on several days | 2.19× | |
| Clicked a marketing email | 0.47× | |
| Downloaded gated content | 0.83× | |
| Viewed product, use-case or integration pages | 3.98× | |
| Viewed the contact or demo-request page | 7.26× | |
| Talked with the seller on the phone | 4.39× | |
| Top-of-funnel content pages | 0.47× | |
| Lead score (scored MQL) without a hand-raise | 1.84× | |
| Read customer stories or case studies | 4.24× | |
| Viewed the pricing page | 6.58× | |
| Watched a webinar on demand | 1.21× | |
| Website chat or meeting booked on the website | 13.6× | |
| Arrived from an AI assistant | 2.61× | |
| Clicked a paid search ad | 6.33× | |
| Filled a lead form inside a social ad | 0.39× | |
| Viewed comparison or alternatives pages | 2.49× | |
| Came from a review site, marketplace or community | ||
| Sales-accepted before the first touch | 1.29× | |
| Self-reported source names a person, podcast, community or event | ||
| Viewed enterprise, security or trust pages | ||
| New opportunity at the account · 7 signals | ||
| A fresh spike of engagement | 1.52× | |
| Two or more people engaged at the account | 2.08× | |
| A closed-lost account re-engages | 3.14× | |
| A former user works at the account | 2.05× | |
| Account reached by paid social ads | 2.2× | |
| Third-party intent stage or fit score | 2.62× | |
| A former customer account comes back | 1.33× | |
| Winning · 63 signals | ||
| Met the buyer on 2 or more days before the deal | 2.01× | |
| Raised a hand with nothing tracked before it | 0.8× | |
| The seller reached out first | 0.93× | |
| Two or more buyer contacts on the deal | 3.16× | |
| Buyer gone quiet for 30+ days on an open deal | 0.09× | |
| Company size or segment | 0.74× | |
| Met the seller months before (prior relationship) | 1.78× | |
| A 14-day buyer silence in the first month | 0.34× | |
| A buyer executive on a call | 1.64× | |
| A buyer's name engaged at another account | 2.06× | |
| A new face from the buyer | 1.74× | |
| Active in 2 or more channels, or several distinct signals | 0.93× | |
| Buying group wider than the CRM lists | 1.96× | |
| Director, VP or head-of level | 1.06× | |
| Record created at the first signal | 1.17× | |
| Sales followed up within 7 days | 1.02× | |
| The buyer emails in the first month | 2.17× | |
| Took a meeting | 1.44× | |
| Two or more buyer people on the first call | 1.45× | |
| Works in the core buying function | 1.06× | |
| Meetings on 3 or more days in the first month | 2.28× | |
| The first call ran 45 minutes or more | 1.59× | |
| The seller adds a leader, solution engineer or second person | 1.96× | |
| Three or more active buyers in the first month | 2.75× | |
| Wrote or replied to an email | 1.08× | |
| Active on 3 or more days | 1.09× | |
| Explained how they will decide | 2.38× | |
| First engaged touch was a two-way conversation | 1.25× | |
| Had heard of the seller before | 0.96× | |
| Named other vendors in the running | 1.32× | |
| Worried about price | 0.94× | |
| A buyer engaged before under another email address | 1.39× | |
| A buying-window trigger | 2.24× | |
| A dated reason to act now | 1.83× | |
| A senior leader is behind it | 1.77× | |
| Came back to the website on several days | 1.19× | |
| Came from a bought or enriched list | 0.86× | |
| Clicked a marketing email | 1× | |
| Demo request or contact-sales form | 0.98× | |
| Doing the work by hand today | 1.14× | |
| Introduced by a partner | 1.96× | |
| Just exploring | 0.43× | |
| Lead score (scored MQL) without a hand-raise | 1.04× | |
| No job title on record | 0.95× | |
| Talked with the seller on the phone | 0.79× | |
| Took the next step themselves | 1.29× | |
| Trouble with the current vendor | 1.09× | |
| Viewed the contact or demo-request page | 1.06× | |
| Already knows the seller | 2.33× | |
| Arrived through brand: direct visit or branded search | 1.15× | |
| Attended an in-person event, booth or dinner | ||
| C-level or founder | 0.99× | |
| Put a number on the problem | 0.82× | |
| Raised a need the product may not meet | 1.01× | |
| Registered for an event or webinar but did not attend | 0.69× | |
| Replacing a tool they have today | 1.38× | |
| Viewed product, use-case or integration pages | 1.04× | |
| Website chat or meeting booked on the website | 1.09× | |
| Downloaded gated content | 0.81× | |
| Named an integration they need | 1.74× | |
| Top-of-funnel content pages | 1.1× | |
| Use case or product line on the deal | 0.77× | |
| Viewed the pricing page | 1.12× | |
Upside's analysis of 10 companies' data, November 2026. Signals tested at three or more companies.