Every marketing team is running an attribution model. Almost none of them are measuring what they think they're measuring.
Last-click attribution - still the default in most businesses - tells you which channel got credit for the sale. Not which channel caused it. Not which channel started the conversation. Not which touchpoint shifted the consideration. Just the last thing a customer clicked before they bought.
In a world where someone sees a LinkedIn post, googles the brand, reads a Reddit thread, gets forwarded a newsletter, sees a retargeting ad, and then converts on a Google search, last-click tells you Google did it. The rest of the story is invisible.
This isn't a technical problem. Google Analytics could be more sophisticated. The tools exist. The real problem is that most organisations don't want a more accurate picture - they want a defensible one. Last-click survives not because it's right, but because it's legible to a board that wants to see a number next to a channel name.
What nobody tells you is that solving attribution doesn't mean finding a better model. It means getting honest about what you can and can't measure and building a measurement framework around that honesty rather than around the comfort of false precision.
Here's the actual state of attribution in 2026 and what a good marketing team does about it.
📊 STRATEGY - WHY LAST-CLICK IS A FICTION YOUR ORGANISATION HAS AGREED TO BELIEVE
The measurement problem nobody has actually fixed
Last-click attribution has one thing going for it: everyone understands it. The channel that got the last click gets the credit. Simple. Reportable. Completely misleading.
The problem is the customer journey it ignores. Research consistently shows that B2C purchases involve an average of six or more touchpoints before conversion. B2B is higher. High-consideration purchases insurance, software, expensive consumer goods — can involve dozens of interactions across weeks or months. Last-click takes that entire journey, credits one moment of it, and discards the rest.
What gets discarded is usually the most important part. The LinkedIn post that introduced the brand. The newsletter that explained why the product was worth considering. The podcast mention that shifted the framing. The word-of-mouth recommendation that made the search happen in the first place. None of these get a click attributed to them. All of them did work.
Multi-touch attribution was supposed to fix this. The idea is sound: distribute credit across all the touchpoints in the journey, weighted by their position or influence. The practice is messier. Multi-touch models require you to track the full journey, which means cookies, cross-device matching, and a logged-in user state that most customers don't provide. Privacy changes (iOS 14, third-party cookie deprecation) have made the tracking problem significantly worse. What you get is a model that's more sophisticated than last-click but still operating on an incomplete picture.
Data-driven attribution - where machine learning distributes credit based on actual conversion patterns is better still, and now available in GA4 and most major ad platforms. But it requires volume, requires data integrity, and still can't account for the channels it can't see.
CORE INSIGHT: No attribution model solves the measurement problem. They each make different assumptions about what they can't see and allocate credit differently as a result. The question isn't "which model is right?" — it's "which model's blind spots are least damaging to our decisions?" That's a much more honest question to take into a budget conversation.
→ Takeaway: Run your last 6 months of conversion data through at least two attribution models — last-click and data-driven if available in your platform. Look at which channels gain credit and which lose it between models. The delta between models is where the interesting decisions live.
What "direct" traffic is actually telling you and why it matters more than most teams realise
Open your analytics right now. Find your direct traffic channel. For most brands, it's in the top three sources by session volume, often the top two.
Direct traffic, in most analytics setups, is the channel where measurement broke down. It's the users who arrived without a referrer being tracked — which means someone typed the URL directly, yes, but also: someone clicked a link in a WhatsApp message, an email client that strips parameters, a Slack conversation, a Discord channel, a PDF, a private Instagram DM, or any of a dozen other places that don't pass referrer data.
That's dark social, the enormous volume of sharing and recommendation that happens in private channels, completely invisible to standard analytics.
The scale of it surprises people who haven't looked. Studies have consistently found that dark social accounts for the majority of online sharing. When someone shares a newsletter in a team Slack, forwards an article to a colleague, pastes a link into a WhatsApp group, or DMs a LinkedIn post to someone who might find it useful — none of that appears in your social channel data. It appears as direct. Or it doesn't appear at all.
This matters for two reasons. First, it means your "best organic channel" measurement is wrong. Content that gets heavily shared in dark social channels will show up as direct traffic, you'll see the traffic spike without seeing the cause. Second, it means that channels you're underinvesting in because they "don't show ROI" may be generating significant dark social sharing that's showing up somewhere else in your data.
The newsletter is a particularly good example. Email open rate, click rate, list growth — all measurable. But the newsletter issue that gets forwarded to a colleague, shared in a Slack channel, discussed in a podcast, or copy-pasted into a team presentation produces attribution-invisible impact that the metrics will never fully capture.
CORE INSIGHT: Dark social is not a niche analytics problem. For most B2B and considered-purchase B2C brands, it's where the majority of word-of-mouth actually happens. The brands that grow fastest aren't better at measuring dark social — they're better at creating content worth sharing in places that can't be measured.
→ Takeaway: Tag your major content assets with UTM parameters and share them through multiple channels in a single week. Watch where untagged direct traffic spikes relative to when you publish. The correlation - imprecise as it is - gives you a proxy signal for which content is getting dark social traction that your attribution model can't see.
📣 PPC - WHAT ATTRIBUTION ACTUALLY MEANS FOR YOUR CHANNEL BUDGET
How to make budget decisions in a world where measurement is structurally broken
The budget conversation is where the attribution problem becomes real. If you can't trust your attribution model, how do you justify spend by channel? How do you cut underperforming channels without cutting the ones doing invisible work?
Three things that actually help:
1. Incrementality testing. The most honest way to understand a channel's contribution is to turn it off and measure what happens. Geo holdout tests where you suppress a channel in one geography and compare conversion rates against a matched control, give you a direct read on what the channel is actually contributing, net of what would have happened anyway. It's disruptive, it requires statistical rigour, and it's significantly more honest than any attribution model. Major platforms including Meta and Google offer incrementality testing tools. Most brands don't use them.
2. Media mix modelling. MMM takes aggregate data, total spend by channel, total revenue, external factors like seasonality and uses regression analysis to estimate the contribution of each channel. It can't track individual journeys, but it can handle the channels that individual tracking can't reach: TV, outdoor, podcast, dark social, word of mouth. It's the only method that gives you a view of the full picture, including the unmeasurable parts. The cost of proper MMM has fallen significantly with modern tooling; it's no longer only accessible to enterprise brands.
3. Brand tracking. Measuring unaided brand awareness, consideration, and purchase intent over time gives you a signal that's orthogonal to your channel data. If brand awareness is rising in a demographic you're targeting, something is working, even if the attribution model can't tell you what. Brand tracking used to require expensive research panels; tools like Tracksuit have brought it into reach for mid-market brands.
CORE INSIGHT: If you're making channel budget decisions based solely on last-click attribution, you're optimising for the thing you can measure rather than the thing that matters. The solution isn't a better model it's a portfolio of measurement approaches that triangulate toward the truth rather than pretending any single model provides it.
→ Takeaway: Pick one of the three above that you're not currently using. Incrementality testing if you're PPC-heavy. MMM if you have meaningful offline or unmeasurable channel spend. Brand tracking if you're investing in any awareness activity at all. None of these replace attribution, they complement it honestly.
🗣️ STRATEGY - WHAT TO ACTUALLY TELL LEADERSHIP
The conversation most marketing teams get wrong and how to frame it better
The hardest part of the attribution problem isn't technical. It's organisational. At some point, you're sitting across from a leadership team that wants to know which channels are working, and the honest answer is: "We're not entirely sure, and neither is anyone else, and here's why."
Most marketers don't say that. They pick the most defensible model, present the numbers with confidence, and avoid the conversation about its limitations. This feels safer in the short term and creates worse decisions in the long term.
The alternative is to reframe the conversation around what measurement can genuinely tell you and be explicit about what it can't.
The framework that works: present attribution data as directional, not definitive. "Last-click shows Google Search driving 40% of conversions, but this doesn't account for the touchpoints that started the consideration. Here's what we're doing to get a fuller picture." Then show the supplementary data brand tracking trend, dark social proxy signals, incrementality test results if you have them.
Leadership doesn't actually need perfect attribution. They need confidence that the marketing team understands the measurement landscape and is making decisions in good faith within its limitations. False precision - claiming your model is accurate when it isn't, destroys that confidence the moment it's questioned. Honest imprecision, explained clearly, builds it.
CORE INSIGHT: The goal of the leadership conversation isn't to show that your attribution model is right. It's to show that you understand what your measurement can and can't tell you and that your decisions are calibrated accordingly. That's a higher bar than presenting a number, but it's the one that earns long-term budget trust.
→ Takeaway: Before your next channel performance review, write down three things your current attribution model cannot measure. Present those alongside the numbers. It reframes the conversation from "defending a model" to "honest measurement in a complex environment" and that's a conversation marketing almost always wins.
🔧 TOOL OF THE WEEK
Northbeam (or Rockerbox if you're mid-market) - multi-touch attribution platforms built for the post-cookie world.
Both tools work by combining first-party data, modelled signals, and platform-reported data to build a more complete attribution picture than GA4 can provide on its own. They're not perfect - nothing is - but they're significantly more honest about their assumptions than last-click.
Specific use this week: if you're on one of these platforms, run a model comparison report, last-click vs data-driven vs position-based. Look at which channels gain or lose credit across models. Any channel that consistently underperforms on last-click but gains substantially on other models is probably doing more work than your reporting suggests.
If you're not on a paid attribution platform, GA4's model comparison tool does a simpler version of the same thing and it's free.
YOUR ONE ACTION THIS WEEK
Have the honest conversation about one metric you currently present with more confidence than it deserves.
Pick one number in your regular reporting, a channel attribution figure, a ROAS claim, a cost-per-acquisition and write down the three biggest assumptions that make it look the way it does.
You don't need to present this to anyone yet. Just know what it is. The gap between "number we present" and "what we actually know" is where the best marketing teams spend their thinking time and where most teams don't go.
