Of 53 pipeline problems I investigated in my client work between February and October 2026, 26 came down to a number that was wrong. The leads weren’t real, the credit rule pointed at the wrong team, the report divided the wrong things or the tracking had broken. Five came down to market demand.
So when pipeline looks wrong, I check whether the number is right before anyone changes a campaign or a budget. In my notes that check settled nearly half the cases, and it costs the least.
Method
I record almost every client call in Granola, about 480 meetings between January and October 2026. I kept client working sessions and my internal sessions about client work, and dropped prospect calls, conferences, vendor pitches and personal meetings. With an AI assistant doing the first pass, I searched those notes for every time someone asked why leads, pipeline, conversion or a reported number looked wrong, checked each case against its meeting notes to confirm what was found, and merged repeat discussions of the same problem into one case with one primary cause.
This is one consultant’s notes, not a survey. Most of the cases come from the engagements where I sit in the weekly pipeline meeting, so a few companies carry most of the count. Two of them come up often below, and I call them client A and client B. The notes are AI summaries of the calls, so they can leave things out. In 15 cases the cause is the team’s working conclusion and wasn’t proven. Problems nobody investigated aren’t here at all.
The causes, by category
In four of the ten groups, the reported number itself was wrong.
| Cause | Cases | Was the reported number wrong? |
|---|---|---|
| Records that weren’t real or were mislabeled | 8 | Yes |
| A site or ad account change | 6 | No |
| Source and credit rules | 6 | Yes |
| Definitions, denominators and comparison windows | 6 | Yes |
| Tracking or reporting broke | 6 | Yes |
| Nothing was broken | 6 | No |
| Market or brand demand | 5 | No |
| Deal or segment mix | 4 | No |
| Sales follow-up and capacity | 3 | No |
| Wrong audience or offer | 3 | No |
| Total | 53 | 26 yes, 27 no |
Leads that weren’t real
The largest group was records that looked like leads and weren’t. They included spam booked as sales meetings, form fills from domains that bounced every reply and bots inflating traffic reports. I go through how to audit for them in junk in qualified leads, and good leads marked as junk.
The same group includes real leads carrying wrong labels. At client B, an automated tag marked partner referrals as out of market, so from the sales side it looked like the partner was sending junk.
Site and ad account changes
Several of these were changes in ad accounts. At client B, lowering campaign budget caps to cut spend disrupted the bidding within a week. At client A, changing the paid search goal months earlier had moved credit for qualified leads from non-brand to brand campaigns, which made non-brand look like it had stopped working. A landing page test everyone thought had ended was still running.
Credit rules and report definitions
Twelve cases were about how the number was built. At both clients with outbound teams, outbound credit overlapped with other sources. At client A, deals credited to outbound had started somewhere else, and the first clue was that inbound and outbound converted at almost the same rate. At client B, prospects contacted by outbound often searched the brand afterward, inside the window where outbound got the credit. I wrote about that pattern in when marketing and sales report different lead numbers.
At client B, a report showed a late-stage pass rate falling and blamed the final review step. It was counting leads turned away at the first screen together with those declined at the end, two decisions made by different people for different reasons. A board report at client A divided the quarter’s opportunities by the quarter’s leads, so leads nobody had called yet counted against the rate. A preset “how did you hear about us” list at client B missed about eight in ten people who had found it through AI (more on that in how much AI-driven demand referral tracking misses). For writing these rules down, the HubSpot stage definitions guide has a template.
Tracking that broke
Six cases were plain breakage. A chat widget was down for a week. Consent settings at another client suppressed about a fifth of tracking and audience data. A checkout handoff dropped the source on every self-serve sign-up, so prospects sent from sales to self-serve showed up as lost. Most of an e-commerce client’s catalog dropped out of its online store. A reporting tool showed the current month’s numbers on old monthly reports. The reports kept running through all of them, which is why testing each path from the visitor into the CRM belongs on the list below.
When nothing was broken
Six times the number was right and nothing needed fixing. The usual version was a planned cut read as a decline. At client B, sales read a planned paid search cut as inbound drying up. At client A, a reported rise in spam turned out to be inside the normal range once someone counted it.
Market demand, deal size and sales follow-up
Market demand was the main cause five times. At client A, slowing growth was read as an execution problem until sixteen months of category search showed a flat market, which the board reporting guide covers. Later, branded search there fell for several weeks and a holiday explained only half of it. The company already won nearly every branded auction, so fewer people were searching for the name.
Deal mix showed up four times, and it makes more leads the wrong answer. At client A, revenue fell behind while qualified leads ran ahead of plan, which I wrote up in when the lead plan is on track and revenue isn’t. At client B, one close rate was hiding requested offers that closed and unrequested ones that never did.
Sales follow-up and a wrong audience or offer came up three times each. In one, a rep covering for a colleague stopped working the best-converting inbound offer because the requests looked low intent. For the other places leads stall, see how to find out why B2B leads aren’t turning into revenue.
What people blamed first
In 26 cases the notes record what someone first thought the cause was. That first explanation was right once. It was wrong in 18 and only part of the answer in seven.
That overstates how often first guesses miss, because notes are more likely to mention a first theory when it gets overturned. It matches what I see in the room, though. Paid search got blamed for a slowdown that came from branded demand, and a partner for leads a tag had mislabeled.
Check your own pipeline in this order
The order follows the counts above. The first five steps check whether the number is right.
- Count the records that aren’t real. Read 20 of last month’s qualified leads. Look for bounced domains and spam meetings, and read a few from any source sales has stopped working.
- List every change from the last 90 days. Buttons, forms, landing page tests, ad goals, budget caps and bidding settings, each with a date. Put them on your weekly lead chart, and confirm finished tests really ended.
- Write down the source rule and who owns it. If outbound and inbound convert at nearly the same rate, check ten outbound deals by hand.
- Check the math in the report. Make sure conversion rates follow a group of leads over time instead of dividing one period by itself, and that comparison windows match the event you’re explaining.
- Test the tracking end to end. Submit your own form, start a chat and go through checkout, then see what reached the CRM and with what source.
- Ask whether the drop was planned. If you cut spend or moved budget, judge it by the conversion rate further down the funnel.
- Then look at demand. Compare category search, your branded search and your share of branded auctions over at least a year.
- Split revenue by deal size and deal type before asking for more leads.
- Check who is working each lead source, including anyone covering for someone who’s out.
FAQ
How often is a drop in pipeline caused by the market?
In my notes, 5 of 53 investigated problems, about 9%, came down mainly to market or brand demand. These are one consultant’s clients, all B2B companies, so treat the number as a reason to check the cheaper explanations first. It isn’t a benchmark.
What should I check first when leads drop?
Check whether the leads are real and whether anything changed on the site or in the ad accounts. Those two groups made up 14 of the 53 cases, and neither check needs new tools or budget.
Can I use the dataset?
Yes. The CSV has one row per case with the category, the symptom as it was first reported, the cause found, whether the first explanation was wrong and how confident the finding is. Client details are removed. Please link to this page if you use it.