Your Funnel Is Not Broken. Your Assumptions About It Are.

Most funnel problems are not technical. They are diagnostic. Marketers spend time patching the wrong stage, optimising the wrong metric, and drawing the wrong conclusions from data that was never designed to answer the question they are asking. The funnel is not broken. The mental model behind it usually is.

If you have ever looked at a funnel report and felt like it was telling you something without quite telling you anything useful, you are reading it correctly. Funnel data shows you where people stopped. It rarely tells you why. And that distinction is where most conversion programmes quietly go wrong.

Why Funnel Thinking Goes Wrong Before You Even Open the Data

The funnel is one of the oldest frameworks in marketing, and like most old frameworks, it has survived not because it is perfect but because it is useful enough to keep being used. That longevity has a cost. People stop questioning it.

The standard representation, broad at the top and narrow at the bottom, implies that your job is to push volume in at one end and pull revenue out of the other. Optimise each stage. Reduce friction. Repeat. It is clean, logical, and only partially true.

Real customer journeys are not linear. Someone discovers your brand through a social post, ignores it, sees a display ad three weeks later, ignores that too, searches for a competitor, reads a review that mentions you, comes back to your site directly, and converts. Which stage of your funnel gets credit for that? In most attribution models, the answer is the last click. In most funnel reports, that person appears as a direct visit that converted. The actual experience is invisible.

I spent years running performance channels where the funnel view was gospel. We had dashboards that tracked every stage, weekly reviews where we dissected drop-off rates, and a culture of relentless bottom-funnel optimisation. We got very good at it. We also spent years not noticing that we were primarily optimising the conversion of people who were already going to buy. The funnel looked healthy. The growth was not there.

If you want a broader grounding in how conversion optimisation actually works across the full customer experience, the CRO and Testing hub covers the discipline end to end, from research methods to commercial measurement.

The Drop-Off Fallacy: Confusing Location With Cause

Open any funnel report and you will see a drop-off percentage at each stage. Session to product page. Product page to cart. Cart to checkout. Checkout to purchase. The numbers are precise. The interpretation is almost always wrong.

Seeing that 68% of users drop off at the checkout page does not tell you that the checkout page is the problem. It tells you that 68% of users who reached the checkout page did not complete a purchase. Those are different statements. The problem might be the checkout page. It might equally be that the people arriving at checkout were never going to buy, that the product page set incorrect expectations about price or delivery, that a competitor appeared in a browser tab at the same moment, or that the user simply wanted to save the cart and come back later on a different device.

Funnel data is a symptom report. It tells you where the temperature is high. It does not tell you what caused the fever. The moment you treat a drop-off location as a diagnosis rather than a signal, you start optimising the wrong thing.

I have seen this play out at a significant scale. An ecommerce client had a persistent checkout abandonment rate that the team had been trying to fix for months. They had simplified the form, added trust signals, tested button colours, reduced the number of steps. Nothing moved meaningfully. When we finally ran session recordings and exit surveys, the answer was straightforward: a significant proportion of users were hitting an unexpected delivery cost at the final summary screen. The checkout page was not broken. The product pages were not surfacing delivery information clearly enough. The fix was upstream. The funnel report had been pointing everyone in the wrong direction.

Tools like Hotjar’s funnel analysis and session replay can help you move from location to behaviour, which is a more useful starting point for diagnosis. But the tool is only as good as the questions you bring to it.

The Problem With Optimising Each Stage in Isolation

Stage-by-stage optimisation is the standard approach. It is also the approach most likely to produce local improvements that do not add up to global growth.

Improving your click-through rate from ad to landing page is useful. Improving your landing page to checkout conversion is useful. But if those two improvements are pulling in different directions, the sum is less than the parts. An ad optimised for curiosity clicks will bring a different audience than an ad optimised for purchase intent. If your landing page is built for purchase intent, the curiosity audience will drop off immediately, and your landing page conversion rate will fall even as your CTR improves. You have optimised two stages and made the funnel worse.

This is not a theoretical problem. It happens constantly in performance marketing. Paid search teams optimise for impressions and clicks. Landing page teams optimise for on-page engagement. CRO teams optimise for checkout completion. Nobody owns the through-line. The handoff between stages is where the value leaks, and nobody is measuring the handoff.

Semrush has a useful breakdown of how TOFU, MOFU, and BOFU interact as a funnel framework. The model is helpful for thinking about content strategy, but it reinforces the same siloed thinking if each stage is treated as a separate optimisation problem rather than part of a connected system.

The question to ask is not “how do we improve stage three?” It is “what does the person entering stage three believe, and does what they find there match those beliefs?” That is a different question, and it requires you to understand the experience before the stage, not just the stage itself.

Micro-Conversions: Useful Signal or Measurement Theatre?

Micro-conversions are one of those ideas that sounds rigorous and often is not. The logic is sound: if you cannot measure the macro-conversion directly, measure the smaller behaviours that lead to it. Track email sign-ups, content downloads, video views, time on page, scroll depth. Use those as proxies for intent.

The problem is that a proxy is only useful if it actually correlates with the thing it is proxying. Most teams never check. They set up micro-conversion tracking, see the numbers improve, report the improvement, and assume the macro-conversion will follow. Sometimes it does. Often it does not.

I judged the Effie Awards for several years. One of the things that stands out from that experience is how rarely the metrics presented in case studies had been validated against actual business outcomes. Teams would show impressive engagement numbers, rising micro-conversion rates, improving brand scores. The commercial result was sometimes there. Sometimes it was absent. The measurement had become the point, rather than the business outcome it was supposed to represent.

If you are going to use micro-conversions as a funnel performance indicator, do the work first. Take a historical dataset and check whether users who completed the micro-conversion went on to complete the macro-conversion at a meaningfully higher rate than those who did not. If the correlation is weak, the micro-conversion is not a useful proxy. It is a number that makes your dashboard look busy.

Crazy Egg’s work on AI-assisted heatmaps and session recordings is a better approach to behavioural signal: looking at what people actually do on a page rather than whether they completed a predefined event. Behaviour is messier to interpret, but it is closer to reality.

What Goes Into the Top of the Funnel Determines What Comes Out

This is the point that most conversion optimisation programmes ignore, and it is the one that has the most commercial leverage.

Your conversion rate is not just a function of how well your funnel is designed. It is a function of who you are sending into it. A funnel filled with high-intent, well-matched audiences will convert at a higher rate than an identical funnel filled with low-intent or poorly-matched audiences. This is obvious when stated plainly. It is consistently overlooked in practice.

Think about it from a retail perspective. Someone who walks into a clothes shop, picks up a jacket, and tries it on is far more likely to buy than someone who walked past and glanced in the window. The in-store experience matters, but the biggest variable is who walked through the door and why. Optimising the changing rooms does not help if you are not attracting people who want to try things on.

Earlier in my career, I was fixated on lower-funnel performance. I believed, genuinely, that the conversion rate was the primary lever. If we could just squeeze more out of the people already in the funnel, we would grow. It took a few years of working across broader channel mixes, and watching businesses that invested in upper-funnel brand activity consistently outgrow those that did not, to recalibrate that view. Much of what performance marketing takes credit for is capturing intent that existed before the ad appeared. The person was going to buy. The search ad intercepted them. That is valuable, but it is not growth. Growth comes from reaching people who were not already planning to buy from you.

When I was growing an agency from around 20 people to over 100, one of the most important commercial lessons was that our new business conversion rate was almost entirely a function of the quality of the leads we were talking to. When we tightened our targeting and only pursued opportunities where we had genuine fit, our win rate went up significantly without any change to our pitch process. The funnel had not changed. The input had.

Mailchimp’s ecommerce CRO resources touch on audience segmentation as a conversion lever, which is the right instinct. Segmentation is not just a personalisation tactic. It is a funnel quality control mechanism.

The Attribution Problem Inside the Funnel

Funnel analysis assumes you can see the funnel. In most businesses, you cannot. You can see fragments of it, stitched together by tracking pixels, UTM parameters, and CRM data that was never designed to talk to each other. The funnel you are optimising is a partial reconstruction of a reality you cannot fully observe.

This is not a reason to stop measuring. It is a reason to hold your measurements more lightly than most teams do.

Cross-device journeys are largely invisible in standard funnel reporting. A user who researches on mobile and converts on desktop appears in your data as two separate users unless you have a login or identity resolution layer. The mobile session looks like a bounce. The desktop session looks like a direct visit that converted. Your funnel report shows strong direct conversion performance and poor mobile performance. Neither conclusion is accurate.

Offline touchpoints are even harder. A customer who saw a TV ad, heard a podcast mention, and then searched for your brand will appear in your funnel as a branded search click. The brand investment that primed them is invisible. This is why businesses that cut brand spend often see short-term performance metrics hold steady before the conversion rate starts to decline six to twelve months later. The funnel looks fine until it does not, and by then the damage is done.

The honest position is that your funnel data is a useful but incomplete picture. The job is to make good decisions with incomplete information, not to pretend the information is complete. That requires judgment as much as it requires analytics.

Testing Inside the Funnel: What A/B Tests Can and Cannot Resolve

A/B testing is the standard answer to funnel optimisation questions, and it is a good answer for a specific class of problems. It is not a good answer for all of them.

Testing works well when you have a clear hypothesis, sufficient traffic volume to reach statistical significance in a reasonable timeframe, and a single variable you can isolate. Most real funnel problems do not fit that description. They are multi-variable, low-traffic, or driven by factors that cannot be tested on a page, like audience quality or messaging consistency across channels.

The other limitation of A/B testing inside a funnel is that it measures the decision at a single point, not the downstream outcome. A variant that improves checkout completion by 12% sounds like a clear win. If that variant attracts a slightly different user behaviour that leads to higher returns or lower customer lifetime value, the test result was misleading. You need to follow the cohort further down the line than most testing programmes do.

Crazy Egg’s overview of split testing methodology is a solid primer on the mechanics. The discipline of testing is sound. The mistake is treating a statistically significant result as a business answer rather than an input to a business decision.

Qualitative research is the complement that most testing programmes underuse. Before you test a variant, understanding why users are behaving the way they are gives you a better hypothesis to test. Hotjar’s ecommerce CRO guidance makes this case well: session recordings and on-site surveys surface the friction points that funnel data can only hint at. The combination of qualitative insight and quantitative testing is more powerful than either alone.

Content’s Role in Funnel Performance

Content is the part of funnel performance that most conversion optimisation programmes treat as someone else’s problem. CRO teams focus on page design, form fields, and checkout flows. Content teams focus on traffic and engagement. The gap between them is where a significant amount of funnel value sits.

Content shapes what a user believes before they reach a conversion point. If the content they consumed on the way in created the wrong expectation about price, product fit, or process, the conversion page cannot fix that. The user arrives with a belief that the page cannot overcome, and they leave. The funnel report records a drop-off. The CRO team tests button colours.

Moz has written about using blog content as a conversion funnel driver, which captures the right instinct. Organic content that ranks for mid-funnel queries, comparison searches, and category-level intent terms is doing conversion work before the user ever reaches a product page. Treating it as purely a traffic play misses the point.

The most effective funnel programmes I have worked on have always had strong alignment between content strategy and conversion strategy. Not because someone mandated it, but because the data made it obvious. When we looked at which traffic sources produced the highest conversion rates, it was consistently organic visitors who had consumed multiple pieces of content before reaching a product page. They arrived better informed, with more accurate expectations, and converted at a higher rate. The content was doing the selling.

Building a Funnel View That Is Actually Useful

Given everything above, what does a useful funnel analysis actually look like? Not a dashboard that shows drop-off rates at each stage, though that is a starting point. A useful funnel view answers a different set of questions.

First: who is entering the funnel, and does that match the audience you designed the funnel for? If your funnel was built for a B2B buyer with a three-month consideration cycle and you are sending it traffic from a broad awareness campaign, the conversion rate will look poor regardless of how well the funnel is designed.

Second: what do users believe at each stage, and is that belief accurate? This requires qualitative input. You cannot answer it from quantitative data alone. Exit surveys, on-site polls, and user interviews are not optional extras. They are the mechanism for understanding the gap between what your funnel communicates and what users understand.

Third: where are the highest-volume drop-off points, and what is the most plausible cause? Not the most convenient cause, not the one that supports the test you already wanted to run. The most plausible cause, based on all available evidence.

Fourth: what would need to be true for the fix to work? This is the hypothesis discipline that separates useful testing from activity for its own sake. If you cannot articulate what you believe is causing the drop-off and why your proposed change would address it, you are not running a test. You are guessing with extra steps.

The broader discipline of conversion optimisation, done properly, is about building this kind of systematic understanding rather than running isolated experiments. There is more on that approach across the CRO and Testing hub, which covers everything from research methodology to how to make the commercial case for investment in the discipline.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions