Sales Funnel Leaks: How to Find the Real Gap Before You Try to Plug It

2026-09-19

"Sales funnel leak" is a plumbing metaphor, and the metaphor is doing more work than it should. It implies you can see the crack, and that the fix is to plug it. What you are actually looking at is a gap between two numbers — how many people your records show at one stage, and how many they show at the next.

That gap is the signal, not the diagnosis. It can mean people dropped out. It can also mean one of the two counts is wrong. Verification is what tells them apart. Until you have done it, naming a cause is guessing with a percentage attached.

Start by reconciling the whole chain, not the worst-looking stage

Line up the counts end to end for the same period: impressions, clicks, page views, sessions, actions taken on the page, leads recorded, bookings or purchases. Then find the largest unexplained gap and investigate that one. If conversion looks bad and nobody has checked whether people are arriving at all, this is where to start — and if two of your systems disagree with each other about the same stage, that disagreement is the finding, not a rounding issue to wave away.

Skip this if you have already reconciled the chain for this period and every gap in it is explained.

Two things break this exercise before it starts.

Undefined events. Write down exactly what behaviour counts at each stage and what population it is divided by, before you quote any rate. If "lead" means a form submission in one system and a qualified enquiry in another, the two counts were never comparable — and the space between them can read as lost people when it is a translation problem. If a number sounds alarming and nobody in the room can say what it counts, define it before you act on it. If every metric in play already has a written definition with an explicit denominator, this one is done.

Blended populations. Report outcomes separately by traffic source, by period, by page version and by offer version. Average cold and warm traffic into one figure and that figure cannot tell you which one moved. Keep the combined total for operations, not as a test result. If there genuinely is only one source, one version and one offer in the period, there is nothing to separate.

Confirm the gap is real before you name a cause

Four checks, each with its own applicability.

Run a measurement check on the data itself. Date range, timezone, filters, campaign status, destination URL, which variant is actually live, event configuration, sample size. Do this before interpreting anything — and do it first when something looks suddenly and dramatically wrong. A sharp overnight move is a reason to inspect the configuration before you diagnose performance; it does not tell you that the configuration is what changed. If you ran this check on this data set already and nothing has changed since, move on.

Go to the system that actually receives the outcome. For every outcome, one system is authoritative — the CRM, the calendar, the checkout, the payment processor — and platform reporting is a secondary account of it. Reconcile the two. When someone says there are no leads, check the authoritative system before you believe the sentence. This matters enough that we wrote a separate piece on how to tell whether your CRM pipeline numbers are decision-ready. If platform numbers and first-party records already agree within a margin you understand, skip it.

Fire the event yourself. Complete the action on a real device and watch for it to arrive. If it never lands, the number you have been treating as a leak was never measuring that action. Do this before concluding anything from an event's absence. If the event has been verified since the last change to the page, the funnel, or the tracking, you are covered.

Instrument every handoff between tools. Draw the actual technology path a visitor walks, and assign a verified event to each point where the domain, platform or tool changes. Those crossings are where records can go missing or arrive disconnected from the session that produced them, which is why each one needs a verified event rather than an assumption. A thank-you page view proves the action only where you have confirmed that reaching that page requires completing it. If the funnel runs on one platform with no boundary crossings and completion is verified, this does not apply.

One more, where buying takes time: link each sale back to the enquiry that started it, and report cohorts that have not finished converting separately from matured ones. If your sales cycle is longer than your reporting window, a recent period can show enquiries with few sales attached — which is what an immature cohort can look like, and also what a leak looks like. Treat a half-finished cohort as a verdict and you risk diagnosing a stage on evidence that has not finished arriving. Immaturity does not clear the stage either; it means you cannot yet tell. Where purchase is effectively immediate, same-period comparison is valid and this does not arise.

Then work the stages in order and stop at the first one that fails

When no single constraint is obvious, order matters, because a failure at an early stage makes any conclusion you draw at a later one unreliable. Work down this list and stop where it breaks:

1. Viability — does a qualified, reachable, willing market exist at economics this business can support?
2. Measurement readiness — is the data real enough to make decisions on?
3. Offer, audience and message — does the right person understand what this is and why it is for them?
4. Traffic — is the right person arriving, in enough volume to learn from?
5. Page and funnel — does the path from arrival to purchase hold together?
6. Email and lifecycle — is the value of people who did not buy immediately being captured?
7. Economics and scale — do the unit economics support spending more, and can fulfilment absorb it?

It is tempting to start at stage five — the page, the handoffs, the follow-up — and stay there. That is fine if stages one through four are settled. If one of them has failed, your stage-five conclusions are unreliable until that failure is dealt with. That is a statement about interpretation, not about arithmetic: a broken offer does not make your page view count false, it makes what the count means impossible to read while the earlier stage is broken.

On stage one specifically: if acquisition metrics have been optimised repeatedly without revenue following, or if nobody has ever counted how many people meet every condition required to buy, that count is the thing to produce. If you already have repeat customers at an acquisition cost your margin supports, viability is answered and the constraint sits downstream.

Choosing which gap to work when several look bad

Two rules, and they apply in different situations.

If improvements are being planned for lower sections of the funnel while most visitors never reach them, find the first point where a meaningful share of intended visitors leave and work there. Not the section that annoys you, not the one nearest the money. If the earliest consequential loss is already understood and addressed, that is not your constraint.

If several stages all work adequately and you are choosing between them, estimate what a plausible improvement at each would do to qualified outcomes, and work the one with the largest effect. Where one stage is clearly broken, skip the estimating and fix the break.

Before either: walk the whole path on real devices, looking for outright breakage. A styling fault, a broken button, a form that fails on mobile. Fix what is broken before drawing conclusions from data the breakage produced. If the path has been walked since the last change and nothing is broken, that particular check is finished — the definitions, event verification, first-party reconciliation and cohort checks above still stand on their own.

Three visible leaks that may point somewhere else

Strong engagement, weak downstream conversion. When click rate is healthy and opt-ins or purchases are not, investigate who is arriving before rewriting the ad. That is an instruction to check the traffic, not a conclusion that the traffic is the cause — the creative may still be part of it. This pattern is worth checking specifically when conversion fell after an audience or geography change, or when cheap traffic is being bought because it is cheap. Our piece on tracing unqualified leads back to the stage that produced them works this in detail. If engagement and downstream conversion moved together, the ad and the traffic are telling you the same story and there is nothing to separate.

Drop-off that compounds across many steps. Draw the whole path and write down the job of each step. Any step whose job could be done elsewhere is a candidate for removal or combination — particularly steps that exist because two tools are separate, or because funnels usually have one. If every step has a stated customer, qualification, measurement, safety or fulfilment purpose, leave the structure alone.

The nurture sequence added as a rescue. Before inserting an email step because conversion is weak, name the reason the buyer needs time. A free resource and an email step wedged between the ad and the actual offer can add another place to lose people. Where the purchase genuinely requires consideration and the sequence serves that consideration, it belongs there.

How much evidence before you act on a gap

Decide how much evidence this particular decision needs before you look at the result, based on how expensive and how reversible the decision is. Rebuilding a checkout and changing a headline are not the same bet.

Then count the actual events, not the percentages. A stage converting at "20%" on a handful of events is a different object from the same rate on many, and asking what one more event would do to the number tells you which one you have. The exception: if you are only confirming that an event fires, one successful test is enough — that is not a sampling question.

Where this ends

These checks tell you where to look. They do not exhaust the possibilities. Insufficient volume, execution problems, timing, and interactions between two stages can all survive every check above. If working the chain and the sequence does not isolate a cause, what you have is evidence that is not sufficient to diagnose it — and the next move is more first-party investigation, not a plausible-sounding label applied to a gap you have not verified.

If you want a second read on which gap in your own chain is worth working first, GrowthHackr's Campaign Leak Audit walks your setup against these checks and costs nothing.