What a leaky sales funnel actually is
A leaky sales funnel is one where records enter a stage and too few of them leave it for the next stage. The leads already exist, the spend already happened, and they stop somewhere specific. Nothing about it is mysterious once you can name the stage.
The word leak is doing real work here. A leak is not low volume, and it is not a long sales cycle. It is a gap between how many records arrive at a stage and how many meet that stage's exit rule.
That definition has a practical consequence. You cannot see a leak in a single top to bottom conversion number, because that number averages every stage together. You see leaks only when you compare two adjacent stages against each other.
This page is a diagnostic, not a list of tips. The order is always the same: measure, find the biggest leak, fix one thing, measure again. If you need the stages themselves first, start with the B2B sales funnel template and come back.
Sales funnel leakage is the same thing by another name
Some teams say sales funnel leakage, some say pipeline leakage, and revenue operations teams often say revenue leakage. The vocabulary changes with the department. The measurement does not: prospects enter a stage, fewer prospects leave it, and the gap is the leak.
Pipeline leakage usually means the sales stages only, after a lead has been accepted. Funnel leakage covers the whole path, including the marketing stages before sales ever touches the record. Use whichever word your team already uses, and say which stages you mean.
Signs you have a leaky sales funnel
None of these prove a leak on their own. They tell you the diagnostic is worth running.
- Volume is up and closed business is flat. More is going in, the same amount is coming out, so something in the middle is absorbing the difference.
- The forecast keeps missing in the same direction. Deals sit in one stage far longer than the forecast assumes, then close late or not at all.
- Reps say the leads are bad, marketing says the leads are not worked. Both claims are testable, and the disagreement usually means nobody has counted.
- A stage has a long median time and almost no activity logged. Records are parked there rather than being worked and lost.
- Losses have no written reason. When the reason field is empty or always says "no budget", the funnel cannot tell you anything.
- The same accounts come back next year. They were reachable, but something stopped them the first time.
What a leaky sales funnel costs
The cost is not only lost revenue. The marketing spend that produced those prospects is already gone, sales time went into records that were never going to close, and the pipeline that leadership forecasts from quietly includes deals a leaking stage will absorb.
That last part does the most damage. A funnel that leaks in a predictable place still produces a confident forecast, so the miss arrives at the end of the quarter instead of at the start, when the sales team could still have done something about it.
Three things that look like leaks and are not
Half of the funnel problems people bring to a review are not leaks. Ruling them out first saves weeks of work on the wrong thing.
| What it looks like | What it actually is | How to tell |
|---|---|---|
| A stage where deals sit for months | A speed problem, if records still exit at the normal rate | Compare exits to entries for that stage; if the rate holds, it is slow, not leaky |
| A sudden drop in a late stage | An incomplete cohort: those records have not had time to exit yet | Check whether the cohort is older than the median time in that stage |
| A low top to bottom rate | Often just a wide top stage that was never meant to convert | Read the stage to stage rates instead; a wide top with a fair second stage is a strategy, not a leak |
There is a fourth case worth naming: seasonality. If your buyers disappear in the same weeks every year, a cohort from those weeks will read as a leak. Compare it with the same weeks last year, not with last month.
How to find funnel leaks with stage to stage rates
The measurement is deliberately simple. For every pair of adjacent stages, divide the records that met the exit rule by the records that entered. That is the stage to stage rate, and it is the only rate that points at a stage.
HubSpot's funnel reports name the distinction directly: next step conversion is the share moving from one stage to the next, while cumulative conversion measures from the first stage onward. Diagnosis needs the first number. Reporting can have the second.
Count a cohort, not a calendar month
A month mixes records that entered at different times, so a long deal appears to convert in whichever month it happened to close. Take the records that entered stage one in a closed period, then follow that same group forward through every stage.
Cohort counting is slower to produce and far harder to argue with. It also stops the most common false alarm, which is a recent cohort that simply has not aged into the later stages yet.
Vendors publish average funnel conversion rates measured on their own customer base, with their own stage definitions. This page quotes none of them. The comparison that works is your funnel against itself, one cohort against the next, counted the same way both times.
What to instrument before you can diagnose anything
Most funnels cannot be diagnosed because the data to diagnose them was never recorded. This is the minimum, and most of it is already available if you turn it on.
Two of these are usually free. Salesforce's Opportunity History report shows From Stage and To Stage for stage changes, and HubSpot has read-only stage calculated properties that update automatically as records move through a pipeline, including time spent in each stage.
The web side of the funnel is instrumented separately. Funnel exploration in Google Analytics 4 visualizes the steps users take toward a task and shows where they fall out, with open funnels letting users enter at any step and closed funnels only at the first.
The two halves rarely join up neatly, and that is fine. Use analytics for behavior before the form, and the CRM for everything after it. Trying to force one number across both is how funnel projects stall.
Website traffic is a stage too, and it drops off
For inbound funnels the first drop happens before a record exists. Traffic arrives, reads one page and leaves, and no lead is ever created. That step belongs in the diagnostic, measured in the analytics tool rather than in the CRM.
Keep it as its own stage instead of folding it into deal stage rates. Traffic that never converts is a marketing problem with marketing fixes, and mixing the two makes every sales stage rate below it harder to read.
Find the biggest leak, not the worst rate
Once you have rates, the temptation is to attack the worst percentage. That is usually the wrong stage. A poor rate on a stage that few records reach costs less than a fair rate on a stage that everything passes through.
Rank by records lost instead: entries at the stage multiplied by the share that never exit. Do that for every stage and the list reorders itself, often dramatically.
| Question | What it tells you | Why it is not enough alone |
|---|---|---|
| Which stage has the worst rate? | Where the process is weakest | Ignores how many records ever get there |
| Which stage loses the most records? | Where the money is | A wide top stage will always look bad in raw counts |
| Which stage lost the most records that matched your fit criteria? | Where the money you actually wanted is | Needs fit recorded on the record, not inferred later |
| Which stage changed most against the previous cohort? | What broke recently, rather than what was always weak | Needs two comparable cohorts counted the same way |
The last question is the one most teams skip and the one that pays fastest. A stage that has always converted poorly is a design choice. A stage that converted fine last quarter and does not now is a break, and breaks have causes you can still find.
The funnel leaks table: symptom, cause, fix
This table maps what you see in the numbers to the cause behind it and the first change to make. Confirm the cause before acting on the fix column; the same symptom can have two causes.
| Symptom in the data | Likely cause | How to confirm it | First fix |
|---|---|---|---|
| Records enter, sit untouched, then go cold | Slow follow-up | Compare the created timestamp with the first human contact date | Name one owner per record and one deadline for first contact |
| High volume into stage one, low qualification rate | Bad fit at the top | Score the lost records against your fit criteria after the fact | Tighten fit at the source, then check rates by source |
| Good conversations that never book anything | No clear next step | Read the last activity on lost records for a scheduled event | End every interaction with a dated next step in the calendar |
| Deals stall for weeks in the evaluation stage | Single threaded deal | Count distinct contacts engaged per account at that stage | Ask who else has to approve, and get to them |
| Proposals go quiet with no objection | No reason to act now | Read loss reasons for "timing", "next year" and "on hold" | Tie the decision to a dated event on the buyer's side |
| A stage with long median time and no logged activity | Handoff drop | Check owner changes against the first activity after the change | Write the handoff rule: who, by when, and what happens if not |
| Rates fine per stage, revenue still short | Not a leak: volume or deal size | Compare entries at stage one against the previous cohort | Work on the top of the funnel, not on the middle |
Leak one: slow follow-up
This is the most common leak and the cheapest to fix, because nothing about the offer, the message or the targeting has to change. A record arrived, and nobody contacted it while the buyer was still thinking about the problem.
It hides well. The record looks worked, because eventually someone did work it. The gap only appears when you put the created timestamp next to the date of the first human contact and sort by the difference.
The fix is structural rather than motivational. One named owner per record, one deadline for first contact, and a report that shows breaches by name. A published sales cadence turns the deadline into a sequence rather than a reminder.
Speed cannot rescue a record that never fit. If you fix follow-up while the top of the funnel is still full of the wrong companies, you will move faster into the same losses and read it as proof that speed does not matter.
Leak two: bad fit at the top
Every stage below the top inherits whatever the top lets in. When fit is loose, the second stage rate collapses and every later rate looks worse than it is, because the records that reached those stages were never buyers.
Diagnose it backwards. Take the records lost at stage two, score them against your ideal customer profile, and see how many would have failed fit on day one. If most would, the leak is at the source.
Fixing it means changing what enters, not working the same leads harder: tighter criteria on forms and lists, a qualification question earlier, and rates reported by lead source so the weak source is visible. Our guide on how to qualify sales leads covers the criteria themselves.
Expect the total to fall before anything improves. Fewer records will pass stage one, which looks like a step backward on a volume dashboard and is the point of the change.
Lead scoring does not fix lead quality
Lead scoring is often sold as the fix for this leak, and it is really a measurement aid. A score tells you which leads look qualified. It cannot make unqualified prospects qualified, and a score built on the wrong fit criteria ranks bad-fit leads confidently.
Check the model against outcomes before you trust it. Take the records the process marked as qualified and count how many reached the next stage. If that looks close to random, the score is describing engagement rather than fit.
Leak three: no clear next step
A record leaves a good conversation with nothing on the calendar. Both sides felt it went well, and there is nothing to return to. The deal does not get rejected, it simply has no next event.
In the data this looks like records lost with positive activity history and no scheduled event in the last entry. In a call recording it sounds like "I will send some information over" with no date attached.
Messaging plays a part here. When the landing page, the email and the call each describe a slightly different problem, there is nothing specific for the prospect to agree to next, and the conversation ends politely with no commitment on either side.
The fix is a rule rather than a script: no interaction ends without a dated next step that both sides agreed to, and a record without one is flagged. This applies to email too, where the ask is the whole point of a call to action email.
Leak four: single threaded deals
One contact is carrying the whole deal. They are enthusiastic, they answer quickly, and they cannot sign. When they go on leave, change jobs or lose the internal argument, the deal stops with no warning in the data.
The measurement is a count: distinct contacts engaged per account, at the stage where deals stall. If the median is one, you have found the cause of a long stalling stage without needing to read a single note.
The fix belongs to the stage definition, not to a rep's judgment. Make a second engaged contact part of the exit criteria for the evaluation stage, and see multithreading in sales for how to ask without going around your champion.
Leak five: no reason to act now
The buyer agrees with everything and still does not move. Nothing is wrong with the product, the price or the relationship. Doing nothing simply costs them less than doing something this quarter.
This leak shows up in loss reasons that say timing, next year or on hold, and in deals that reopen twelve months later at the same stage. It is also the leak most often misread as a pricing problem, which leads to a discount that changes nothing.
The honest fix is to find a dated event on the buyer's side that the purchase serves: a contract ending, a system being retired, a target with a deadline, a team arriving.
If no such event exists, the deal is not lost, it is early. It belongs back in lead nurturing with the expected date written down, rather than sitting in the pipeline making the forecast look better than it is.
Leak six: records dropped at the handoff
Every change of owner is a place where a record can stop. Marketing to sales, sales development to account executive, sales to onboarding: each handoff has a moment where nobody is holding the record and nobody notices.
Handoff leaks look different from other leaks. They produce a stage with a long median time and almost no logged activity, rather than a pile of losses with reasons. Nothing was rejected, because nothing was worked.
They are usually alignment problems wearing a process costume. When marketing and sales disagree about what a qualified lead is, records arrive that the sales team does not believe in, and quiet disbelief looks exactly like a slow handoff in the data.
Diagnose it by comparing the owner change timestamp with the first activity after that change. Fix it by writing the rule down: who receives the record, by when they must act, what context travels with it, and what happens when the deadline passes.
The sales handoff entry covers that rule in detail, and MQL vs SQL covers the shared definition both teams have to agree on before any handoff rule can be enforced.
A diagnostic order that saves weeks
Work the funnel from the outside in. Cheap checks that can rule out whole categories come first, and the expensive qualitative work comes last, on one stage only.
The first box is not a formality. If two people would classify the same record differently, the cohort count is measuring disagreement rather than the funnel, and every later step inherits that.
The last box is the discipline that makes the whole thing work. One change, in one stage, with a written prediction and a recount date. Everything else in the funnel stays frozen until that date passes.
Confirm the cause before you fix it
A stage tells you where records are lost. It never tells you why. The step between the number and the change is reading the records themselves, and skipping it is the main reason funnel projects produce activity instead of results.
- Read twenty lost records from that stage. Not a dashboard: the actual activity history, timestamps, notes and last message.
- Sort by the gap that matters. Created to first contact for follow-up, owner change to first activity for handoffs, contacts engaged for single threading.
- Listen to three calls from the stage. The missing next step and the missing reason to act are audible long before they are countable.
- Ask the rep who lost them, one question. What did the buyer say last? The answer is often a cause you had not listed.
- Write the cause as a sentence you could be wrong about. "Records sit four days before first contact" can be checked. "Leads are bad" cannot.
Fixing the leak versus pouring in more leads
When results are short, adding leads is the faster decision and usually the worse one. Every extra record you buy passes through the same leaking stage, so you pay again for the same loss and the funnel gets harder to read, not easier.
There is a clean rule for choosing. Look at where the biggest leak sits.
| Where the leak is | What more leads do | The right move |
|---|---|---|
| Below the top stage | Multiply the loss and hide the cause under new volume | Fix the stage first, then add volume to a funnel that holds it |
| At the top stage, from bad fit | Make it worse, unless the new source is better targeted | Change the source or the criteria, not the quantity |
| Nowhere: rates steady, entries falling | Exactly what is needed | Add volume, and watch the stage rates while you do |
| Unknown, because nothing is counted | Guarantee that you still will not know | Count one cohort first; it takes days, not quarters |
There is one more reason to fix first. A leak fixed keeps paying on every future cohort, while leads bought pay once. That is not an argument against buying leads, it is an argument about sequence.
How to run the diagnostic
Write the stages and the exit rule for each one
List the stages a record passes through and the one observable event that moves it to the next stage. Without exit rules, every later number is an opinion.
Count one cohort through every stage
Take the records that entered stage one in a closed period, follow that same group forward, and record entries and exits at each stage.
Rank the stages by records lost
Multiply entries by the share that never exit. The stage that swallows the most records is the leak to work on, whatever the percentages say.
Confirm the cause before you change anything
Read the lost records, the timestamps and the notes for that stage. Pick the cause you can show evidence for, not the one you already believe.
Change one thing and write down what you expect
State the change, the stage it should move, and the direction. A prediction written before the change is what makes the next reading readable.
Recount the same cohort measure and compare
Wait one full sales cycle for that stage, recount the same way, and compare. If nothing moved, revert the change and pick the next cause.
How to tell whether the fix worked
The recount is where most diagnostics fall apart, usually because it happens too early or with a different counting rule. Both problems are avoidable if you decide the recount date and the measure before you make the change.
- Wait one full sales cycle for that stage. Reading a cohort that has not aged into the next stage measures recency, not conversion. Our sales cycle length guide covers how to calculate that wait.
- Use the same counting rule. Same cohort definition, same exit rules, same source filter. A rule changed mid-diagnostic makes the comparison meaningless.
- Check the stage above and below. A tightened fit rule should lower stage one exits and raise stage two rates. If only one moved, the change did something else.
- Compare against your prediction, not against zero. A change that moved the number in the wrong direction is information, provided you wrote down which direction you expected.
- Revert what did not work. Leaving failed changes in place is how funnels accumulate rules that nobody can explain a year later.
Ask the people who left
Counting tells you where and roughly when. It cannot tell you what the buyer was thinking, and for the leaks that involve a decision, that is the whole answer.
Two sources are worth the effort. A written loss reason chosen from a short fixed list, entered by the person who lost the deal, gives you something countable. A short conversation with a handful of buyers who stopped gives you the sentence the list was missing.
Keep the list short and mutually exclusive. A loss reason menu with fifteen overlapping options produces data nobody trusts, and "no budget" quietly absorbs everything from bad fit to a missing reason to act now.
Where the numbers live
This page does not rank vendors. These are the categories that hold the data a funnel diagnostic needs:
- CRM: stages, owners, stage change history and loss reasons, which is where most of the diagnostic happens.
- Product or web analytics: step sequences before the form, where funnel exploration shows which step users fall out at.
- Marketing automation: the pre-sales stages, email engagement and the moment a record is passed on.
- Conversation recording: the calls behind a stage, for the causes that are audible before they are countable.
- A spreadsheet: genuinely enough for the first cohort count, and faster than waiting for a dashboard to be built.
- Lead routing rules: not a tool you read, but the place where records are assigned, and therefore where handoff leaks are created.
None of these tools finds a leak by itself. A tool reports what the process already records, so a sales team that skips loss reasons or logs activity inconsistently will get clean charts built on nothing. Fix the recording habit before buying anything.
If the pipeline itself is not defined yet, the counting will not work. The how to build a sales pipeline guide covers stages and exit criteria before any of this applies.
Mistakes that keep funnels leaking
- Reading one top to bottom conversion rate and calling it the funnel.
- Counting calendar months instead of cohorts, so long deals land in the wrong bucket.
- Attacking the worst percentage rather than the stage that loses the most records.
- Changing three things at once, then having no way to tell which one worked.
- Recounting before the cohort has aged through the stage you changed.
- Diagnosing from a dashboard without reading a single lost record.
- Comparing your rates against published benchmarks built on someone else's stage definitions.
- Buying more leads to cover a leak that sits below the top stage.
- Leaving failed changes in place because nobody wrote down what they were supposed to do.
The leak review note
The note below was written for this page. It exists so that a funnel review produces one decision with a date on it instead of a discussion, and so the next review can tell whether the last one was right.
Cohort: {{cohortPeriod}}, {{recordCount}} records entered stage one Worst stage by records lost: {{stageName}} Entries: {{entries}}. Exits: {{exits}}. Median days in stage: {{medianDays}}. Evidence I read: {{evidence}} Cause I am betting on: {{cause}} The one change: {{change}} What I expect to see: {{prediction}} Recount date: {{recountDate}} Everything else in the funnel stays as it is until that date.
You fill it in from a dashboard without reading a single lost record. The note then records a guess with a date on it, and the recount cannot tell you whether the change worked or whether the cohort was simply different.
Frequently asked questions
What is a leaky sales funnel?
A leaky sales funnel is one where records enter a stage and too few of them reach the next stage. The leads exist, the spend already happened, and they stop somewhere specific. The word leaky only becomes useful once you can name which stage.
How do you find funnel leaks?
Count one cohort of records through every stage and divide exits by entries between adjacent stages. Then rank the stages by how many records never exit. The stage that swallows the most records is the leak to work on first.
What causes a leaky sales funnel?
Six causes explain most funnel leaks: slow follow-up, poor fit at the top, no clear next step, a single contact carrying the deal, no reason for the buyer to act now, and records dropped at the handoff between teams.
Should I fix the funnel or just get more leads?
Fix the funnel first when the leak sits below the top stage, because every extra lead you buy passes through that same leak. Add leads when the top stage is genuinely starved and the later stage rates are already steady.
How do I measure stage to stage conversion?
Divide exits by entries between two adjacent stages, on the same cohort. HubSpot calls this next step conversion and distinguishes it from cumulative conversion, which measures from the first stage. Top to bottom numbers hide which step is leaking.
Why should I count a cohort instead of a month?
A calendar month mixes records that entered in different months, so a long deal looks as though it converted in whichever month it closed. Following one entry cohort forward keeps entries and exits attached to the same group.
What should I instrument to diagnose a leaky sales funnel?
The stage a record sits in, the timestamp of every stage change, the owner, the source, the date of the first human contact, the number of contacts involved, and a written loss reason. Most CRMs record the first few automatically.
How long should I wait before judging a fix?
One full sales cycle for the stage you changed, and no less. Reading a stage before the records in it have had time to exit produces a number that mostly reflects how recently the cohort entered.
Can Google Analytics show me where the funnel leaks?
For the web part of it, yes. Funnel exploration in GA4 visualizes the steps users take toward a task and shows where they fall out of the sequence. It covers behavior on your site, not what happens after a record reaches the CRM.
Is a long sales cycle a funnel leak?
Not by itself. A stage that is slow but still passes records on is a speed problem. It becomes a leak when records sit long enough that the buyer's priority moves and the record leaves sideways instead of moving forward.
What is the difference between a funnel leak and bad targeting?
Bad targeting is a leak at the top stage: the records were never going to qualify. It shows up as healthy volume with a low qualification rate, and it is fixed at the source, not by working the same leads harder.
How does the handoff between teams cause leaks?
Records change owner and context is lost. Nobody is named, no response deadline applies, and the record sits. Handoff leaks show up as a stage with a long median time and no activity logged, rather than as losses with reasons.
How many leaks should I fix at once?
One. Changing fit criteria, cadence and messaging in the same month means the next reading cannot tell you which change moved the number, and a change that made things worse stays hidden behind one that helped.
Do benchmark conversion rates help me find my leak?
No. Published funnel rates come from someone else's market, price point and stage definitions. The comparison that works is your funnel against itself, one cohort against the next, counted the same way both times.
- Google Analytics Help, Funnel exploration, for open and closed funnels and how users fall out of a step sequence, checked Sep 23, 2026.
- HubSpot Knowledge Base, Funnel (legacy) report types, for next step conversion versus cumulative conversion, checked Sep 23, 2026.
- HubSpot Knowledge Base, Use stage calculated properties, for automatic time in stage properties on pipeline records, checked Sep 23, 2026.
- Salesforce Help, Opportunity History Report, for the From Stage and To Stage columns on stage changes, checked Sep 23, 2026.
- Jeluvi entries this guide builds on: B2B sales funnel template, How to build a sales pipeline, How to qualify sales leads, Sales handoff, Multithreading in sales.
- The leak table, the diagnostic order and the review note were written for this page. No conversion rates, response times or benchmark figures are quoted, because none of them would be yours.