If your funnel report says “Paid Social is dead,” but your bank account says “That’s funny, revenue looks alive,” congrats. You have met the modern funnel: part measurement system, part improv theatre.
Most funnels do not intend to lie. They just do what dashboards do best: confidently summarise a messy reality into neat little boxes. And in 2026, the reality is extra messy. People bounce between devices, browse in private modes, click a link, get distracted by a dog video, come back later via a “direct” visit, and finally purchase after reading five reviews and asking a group chat that includes one friend who refuses to buy anything not approved by Consumer Reports.
Google literally calls the space between trigger and purchase the “messy middle,” because “endless loop of uncertainty and comparison” apparently did not fit in a chart label.
If it’s not converting, it’s not telling the truth.
What I want to do today is give you a data-first way to tell whether your funnel is showing a real performance problem, or a measurement problem wearing a performance costume. (It happens. Often.) You will leave with a decision framework you can use for a proper funnel performance audit, especially if email is part of your mix and you are tired of being emotionally manipulated by click-through rates.
The big lie: “The funnel is linear, and attribution is factual”
The first reason funnels lie is philosophical. The funnel assumes people move politely from awareness to consideration to purchase, like they are following floor arrows in an IKEA. In real life, the journey is circular and chaotic, which McKinsey has been yelling about since before “growth” became a job title.
The second reason is mathematical. Attribution is not a truth machine. It is an approximation that depends on what you can track, how you stitch identities together, and which model you choose. Even Shopify, which has every incentive to make marketers feel calm and in control, straight-up says there is no such thing as 100% “true” attribution.
So if your funnel is built on a linear story and your reporting is built on an approximation, the output can be tidy and still be wrong. That is not you being bad at marketing. That is the system doing what it does.
Why this got worse: privacy changes turned your funnel into Swiss cheese
When people say “tracking got harder,” it can sound vague, like “eating healthy.” What I mean is: entire categories of signals you used to treat as reliable are now noisy, delayed, missing, or intentionally blurred.
In email, Apple’s Mail Privacy Protection can hide IP addresses and pre-load content, which means senders can’t reliably know if and when someone opened an email, and location gets fuzzy too. Apple explains the mechanics pretty plainly in their support docs. Third parties like Postmark have also documented what this does to open tracking in practical terms: opens inflate, and “open rate” becomes a wobbly engagement proxy instead of a behavioural fact.
In ads and web measurement, cookies and identifiers have been under pressure for years, and Google’s plan for third-party cookies has shifted multiple times. Google announced in July 2024 that it was moving toward an approach that “elevates user choice” rather than simply deprecating third-party cookies. Reuters later reported Google would not roll out a new standalone prompt and would keep current settings, while still continuing Privacy Sandbox work. Translation: your ability to follow users around the internet is not getting simpler.
And in apps, Apple’s App Tracking Transparency changed what data can be accessed for tracking, and frameworks like SKAdNetwork exist specifically to enable privacy-preserving attribution in a more limited way.
None of this is “bad” or “good” in a moral vacuum. A lot of it is genuinely about user privacy and consent. But it does mean your funnel is increasingly built on partial information. Partial information is fine, as long as you do not treat it like the full story.
A funnel performance audit is not a dashboard review. It’s a cross-examination.
A typical funnel review goes like this: you pull up a report, notice a dip, panic slightly, and decide to “fix conversion rate” by changing button colour or sending more emails.
A funnel performance audit is different. You are not just asking “where did conversions drop?” You are asking two questions in order:
First: Is the drop real, or is it a measurement?
Second: If it’s real, what is the highest-leverage behaviour to change?
To get there, I use a simple framework I call the Funnel Lie Detector. It has five tests. No tools required. Just honesty, and the willingness to admit your favourite report might be vibes-based.
Test 1: Pick a “source of truth” conversion, and reconcile everything to it
If you do not have one conversion number you trust, you do not have a funnel. You have competing fan fiction.
For e-commerce, this is usually paid orders in your backend. For subscriptions, it is successful payments. For lead gen, it might be qualified leads accepted by sales, not form fills that include “asdf@gmail.com.”
Once you pick the source of truth, you do a reconciliation exercise: “Do my analytics conversions roughly match the backend over the same time window, minus known differences?” If they are wildly off, your funnel is not lying. It is hallucinating.
This is where a lot of teams discover issues like duplicate purchase events, missing consented tracking, broken tags after a site release, or payment providers that redirect in a way analytics cannot stitch properly. None of these are sexy. All of them will ruin your decision-making.
If you only do one thing this week: reconcile. It is the marketing equivalent of checking whether the scale is on carpet before declaring your life is over.
Test 2: Identify whether you have a “measurement leak” or a “behaviour leak”
A behaviour leak is real friction. People bounce on pricing. They abandon checkout. They do not understand your value prop. Email content doesn’t match landing pages. The offer is weak. The onboarding is confusing. All normal, all fixable.
A measurement leak is when the behaviour happens, but your systems fail to observe it correctly. Common examples include:
You are using a last-click model that credits the final touch, so email “wins” by default because it shows up late in the journey. Shopify’s own definitions make it clear that last click and last non-direct click behave differently, and each can meaningfully change which channel gets credit.
Your lookback windows are too short for your buying cycle, so early influence disappears. In Google Analytics, the default conversion window for many events is 90 days, and you can change it, but that default can still be a mismatch for longer consideration cycles.
Your email “opens” spike because of privacy pre-loading, so your nurture logic marks people as engaged when they are not. Apple explicitly notes remote content can be downloaded in the background, which is exactly why opens can mislead.
Here’s the key: you cannot optimize behaviour with measurement fixes, and you cannot fix measurement with copy tweaks. If you misdiagnose the leak type, you will spend two sprints “improving conversion” while the tracking is quietly broken, like a smoke alarm that only chirps when you have guests.
Test 3: Audit your attribution assumptions, not just your attribution model
A lot of marketers think the attribution model is an assumption. It is not. It is just the final layer of assumptions.
The assumptions you need to surface are things like:
Are you implicitly treating “last touch” as “cause”? Google Ads literally frames last click as giving all credit to the last-clicked interaction and points out this ignores other interactions along the way.
Are you double-counting conversions across platforms? (Yes, you are.) Platform dashboards are incentivized to take credit. Your funnel is not “lying” here. It is being marketed by other marketers, which is honestly kind of poetic.
Are you treating clicks as persuasion? A click is a mechanical action, not a psychological commitment. Sometimes people click because the button is big. Sometimes they buy because their friend told them your product saved their sanity.
This is why I like the IAB framing: multi-touch attribution is about distributing credit across touchpoints, but it still has constraints and does not automatically prove causality. If you want causality, you need experiments, not prettier charts.
Test 4: Triangulate with at least one signal that is hard to game
When funnel data is noisy, you need a triangulation habit. I am not talking about “more dashboards.” I am talking about bringing in one additional signal that is conceptually different, so the same bias does not infect everything.
Good triangulation signals include:
Backend revenue and refunds (because money has fewer opinions).
Customer self-report, like “How did you hear about us?” surveys, as long as you treat them as directional, not courtroom testimony.
Email click-to-purchase rates for tagged campaigns, because clicks are still observable even when opens are not, and they connect more directly to on-site behaviour than an open does. (Still imperfect, but better.)
Holdout or incrementality tests, because they answer the question attribution cannot: “What happened because of this marketing, versus what would have happened anyway?” Google explicitly positions incrementality as causal measurement and has been rolling out updates to make these experiments easier for advertisers. Industry bodies also describe incrementality testing as a key multi-channel measurement approach.
If your funnel says Channel A drove 0 sales, but a holdout test shows Channel A produces incremental lift, your funnel is not just lying. It is actively trying to get your budget cut.
Test 5: Check whether email is being used as the “closer” in reporting, not in reality
This one is personal, because email gets blamed and praised for things it did not do.
Email often shows up late. People search, browse, hesitate, then buy when a cart reminder or promo lands. That makes email look like the hero in last-click or last non-direct click models, because those models credit the final marketing interaction.
But here’s the uncomfortable truth: email can be the final push without being the original persuasion. If your reporting treats email as the “cause,” you will over-invest in send volume and under-invest in the upstream work that made the customer receptive in the first place, like product pages, pricing clarity, reviews, and yes, ads.
This is also where privacy changes mess with lifecycle logic. If you are still using opens as the primary engagement trigger for segmentation or suppression, Mail Privacy Protection can cause you to send more to people who are not actually engaged, which hurts deliverability and customer trust over time. Apple is clear that MPP is designed to limit what senders can learn about mail activity. The ethical move is to shift your engagement definition toward clicks, site behaviour, and downstream value, not to get clever about “workarounds.”
If you are thinking, “But how will I know who is engaged?” welcome to modern email, where the answer is: with more humility and better measurement design.
The part nobody wants to hear: your funnel might be “accurate” and still mislead you
Even if your tracking is perfect (it is not), funnels can still mislead because they turn a dynamic journey into a static story.
Google’s “messy middle” research describes people oscillating between exploration and evaluation before purchase. McKinsey’s consumer decision journey similarly frames decision-making as a loop, including post-purchase experience feeding back into future consideration.
In other words, someone dropping out at “consideration” might not be lost. They might be looping. Someone converting at the bottom might have been emotionally convinced weeks ago and simply needed payday to arrive.
So when you do a funnel performance audit, do not just ask “where did they drop?” Ask “What loop are they stuck in?” Email is especially powerful here because it is one of the only channels that can show up consistently across that loop, assuming you respect consent and do not treat inboxes like a free billboard.
A practical decision framework for your funnel performance audit
When you find a “leak,” run it through this decision filter before you decide what to fix:
First: Is this a measurement leak, a behaviour leak, or both? If you cannot answer, pause and reconcile.
Second: If it is behaviour, is it an intent problem or a friction problem? Intent problems look like low engagement across touchpoints, low brand search, low email click rates, and high bounce. Friction problems look like decent engagement but drop-offs at specific steps, like checkout or trial activation.
Third: Is email the right lever, or is email just holding the bag? If email is “winning” attribution because it is last-touch, you might need to invest upstream rather than sending another urgency subject line that says “Last chance” for the seventeenth time this month. Your subscribers can count. They just choose not to reply.
Fourth: Can you validate with a triangulation signal or an experiment? If you can’t validate, treat your conclusion as a hypothesis, not a fact.
This is how you stop your funnel from becoming a dictatorship run by one report tab.
- Your funnel can “lie” because the customer journey is messy and attribution is an approximation, not a truth source
- Privacy changes make key signals noisier, especially email opens due to Mail Privacy Protection, so engagement and automation logic need to evolve
- A funnel performance audit should separate measurement leaks from real behaviour leaks before you try to optimize anything
- Attribution models tell you who touched the conversion, not what caused it, so triangulation and incrementality tests matter when budgets are on the line
- Email often looks like the “closer” in reporting, so be careful about over-crediting it and under-investing in the work that built intent upstream
If you want more practical frameworks like this, plus the occasional gentle roast of marketing metrics that deserve it, subscribe to The Click Brief. If you are in the middle of a funnel performance audit and your dashboards are fighting each other like toddlers, you can also find me on LinkedIn or Instagram and tell me what’s going on. I love a good measurement mystery, as long as it does not involve “But Meta says…” as the only evidence.