In This Post
In this postCollapse -Open +
- The attribution illusion: why sophistication doesn't fix incomplete data
- How privacy and fragmentation broke traditional measurement
- What CMOs are actually doing instead
- A practical framework for measuring what matters
In this postCollapse -Open +
The Illusion of Certainty
Your attribution model just told you that 34 percent of conversions came from retargeting ads. Confidence level: high. But here's what it actually means: of the small fraction of customer behavior your tools could see, retargeting touched the most recent visible step before a person bought something.
That's not causation. That's an educated guess with a spreadsheet behind it.
The problem isn't new. But in 2026, it's getting worse. AI is making attribution models faster and more sophisticated, and that sophistication is creating a dangerous illusion: that we know what actually drives customer decisions when, in reality, we're flying partially blind.
What Actually Changed
Three years ago, marketing measurement was already complicated. But it was comprehensible. You could trace a path: ad seen on Facebook, click, conversion. The signal was linear enough that attribution models could make reasonable guesses.
Then a few things happened at once:
Privacy killed cross-device tracking. A customer might see your ad on phone, research on a laptop, and convert on desktop. Without cross-device IDs, each step looks like a separate, unrelated event. Your model sees a "direct" conversion and has no idea the ad campaign created it.
Platforms started hoarding data. Apple blocks it. Google now blocks it after they finish deprioritizing it. TikTok's data sharing is limited. That means the platform making money from the conversion knows more about what happened than you do. Your view of the journey is always incomplete compared to theirs.
Customer journeys became genuinely non-linear. A customer doesn't see one ad and convert. They might see your content in a newsletter (which you can't track to the conversion), get mentioned in a Slack group, read a blog post, see a retargeting ad, check reviews, and then buy when their coworker recommends you.
Many of those moments aren't measurable events. Some are invisible.
AI filled the gap by guessing harder. Rather than abandon attribution during this messier era, vendors are using machine learning to fill in the missing pieces. They're modeling unmeasured behavior, inferring offline influence, and predicting which touches "probably" mattered. The confidence level looks high because the algorithm says so. But they're still just guesses.
When channels don't share data, each step looks disconnected. Your model sees the visible pieces and fills in the blanks.
The Fragmentation Problem No Model Can Solve
Let's walk through a real customer journey and see where measurement breaks:
A potential customer sees a LinkedIn post about your product (no event tracked, because it's on LinkedIn and you don't own the data). They think about it. Later, they Google your product name and land on your site (attributed to "organic search"). They browse, don't convert, and leave.
A week later, they see a retargeting ad on TikTok (event tracked). They click through, read the pricing page, and close the tab. No conversion yet.
The next day, they get an email from a contact that includes your company (no event tracked, you don't know about this email). The email is what finally convinced them. They click the link in that email and buy.
Now: your attribution model will credit the TikTok retargeting ad with the conversion because it was the most recent tracked event. The email conversation won't show up because it's not in your measurement system.
The LinkedIn post won't show up. The Google search will get partial credit in a multi-touch model, but your model can't know it happened a week earlier and played a different role.
The model looks at the visible signals, draws a line, and tells you that TikTok ads drive conversions. So you increase TikTok spend. But the real driver was an offline conversation that your model has no visibility into.
This is happening millions of times a day in your customer data.
The shift in 2026 is not that attribution got harder. It's that CMOs stopped pretending it's reliable.
Why AI Attribution Sounds Better Than It Is
Modern AI attribution tools are genuinely sophisticated. They can model complex patterns, incorporate first-party data, and apply statistical techniques that older models can't. But sophistication doesn't fix the core problem: they're modeling journeys based on incomplete information.
Here's what these tools actually do:
They take the data you can see (which is already incomplete) and train models to recognize patterns. They spot correlations between ad exposures and conversions. They weight recent touchpoints more heavily. They estimate the probability that an invisible touchpoint mattered.
All useful. But all predicated on the data they're given.
A sophisticated model built on incomplete data is still predicting based on incomplete information. It just sounds more confident because it's built by a machine learning team instead of a marketing analyst with a spreadsheet.
<a href="https://www.braze.com/resources/articles/challenges-of-marketing-attribution" rel="nofollow noopener noreferrer" target="_blank">Research from Braze on the state of marketing attribution</a> shows that attribution models are reliable for understanding relative channel performance within your data, but they're not reliable for understanding causation without additional verification through experimentation.
Your dashboard looks confident. But confidence in measurement and accuracy in measurement are two different things.
What CMOs Are Actually Doing About This
The smartest operators have stopped treating attribution as a source of truth. Instead, they're using it as one input among many. Here's the shift:
From: "My attribution model says this channel works, so I'm investing more."
To: "My attribution model is showing a correlation. Let me run an incrementality test to see if it's causal."
Incrementality testing means holding back a percentage of your audience from a campaign and comparing the behavior of the holdout group against the group that got the campaign. It's expensive, slow, and creates gaps in your marketing. But it actually tells you whether a channel drives incremental value or just captures users who would have converted anyway.
<a href="https://www.bcg.com/publications/2025/how-cmos-scaling-gen-ai-in-turbulent-times" rel="nofollow noopener noreferrer" target="_blank">According to BCG's research on how CMOs are scaling generative AI</a>, strong measurement of incremental growth and ROI is now a core priority for marketing leaders. Companies like Braze have started including incrementality testing in their standard measurement offering. Google started recommending it.
Meta is pushing it. The pattern is clear: the industry is shifting away from "trust the model" to "verify with experimentation.
Another shift: first-party data and journey context. Smart teams are building systems to understand full-journey context, not just individual touchpoints.
They're connecting email opens, website behavior, in-app interactions, and support tickets into a single view of the customer. That context helps a CMO understand not just which channel touches a customer last, but what that customer was experiencing across the entire relationship.
When you see that a customer was actively browsing your pricing page, received a targeted email about a discount, and then bought, the story is clearer. Your model doesn't have to guess as much because the context is there.
A third shift: customer lifetime value (CLV) over conversion. The biggest mistake in 2026 marketing is still optimizing for the first conversion instead of the second one. Attribution models are built to explain single moments. But a customer who buys once and churns is not the same as a customer who buys, stays, and upgrades.
Smart CMOs are measuring what touches contribute to retention, not just acquisition. That means tracking email opens six months after purchase, in-app engagement, support interactions, and renewal decisions. It's harder to tie these back to which ad someone saw, but it's infinitely more useful for business growth.
The Cannabis and Compliance Angle
If you're marketing in cannabis, this problem gets sharper. Cannabis advertising is already restricted on most platforms. Your ad can't mention "high" or "intoxication." Your landing pages are limited. Your audience targeting is neutered because you can't use detailed behavioral data.
But your customers are still making decisions. They're just deciding in channels you can't easily measure. They're talking in private group chats. They're asking friends. They're visiting dispensaries and asking the budtender for recommendations.
Your attribution model thinks a search campaign worked. Actually, a friend recommended the product in a text message, and the search campaign just captured a customer who was already decided.
So for cannabis brands, the move is even more dramatic: stop trusting channel-level attribution entirely. Instead, focus on:
- Dispensary sell-through and budtender feedback
- Customer surveys about how they found you
- Store traffic patterns
- Offline word-of-mouth tracking
The measurement that matters is not in your ad platform's dashboard.
What to Do Right Now
If you're a CMO reading this, you're probably not comfortable admitting that your attribution reports might be confidence fiction. Here's a practical path forward:
1. Run incrementality tests on your top three channels. Pick your biggest advertising spend. Hold back 5-10 percent of your audience and see what actually moves. You'll probably be surprised.
2. Start tracking CLV instead of just CAC. Look at which customers are staying, spending more, and referring others. That's your real ROI.
3. Build first-party data systems. CRM, email, website analytics, in-app data, support data all stitched together. When you own the data, you own the measurement.
4. Interview customers. Ask them how they found you. Ask what made them decide to buy. Ask what made them stay. The truth is messier than your dashboard, but it's closer to reality.
5. Shift your ad spend accordingly. Once you know what actually works, don't just accept the attribution model's recommendations. Use them as a starting point for testing, not as final answers.
The CMOs who move fastest are the ones who stop asking "What does my attribution model say?" and start asking "What would it take to actually know?"
FAQ
A: No. Attribution models are useful for spotting correlations and patterns within the data you can see. They're valuable for optimization within platforms and for understanding relative channel performance. They're just not reliable for answering "which channel actually drove this conversion" without additional confirmation through testing.
A: Not completely. Use them as one input. But run experiments to verify recommendations before making major budget shifts. A good attribution platform will tell you which channels correlate with conversions. Incrementality testing will tell you which ones actually cause conversions.
A: It won't, not for years. Privacy regulations will keep tightening. Platforms will keep changing their APIs and data sharing. Customer behavior will keep fragmenting across channels. The old idea of "final" attribution is gone for good. The new skill is knowing how to measure under uncertainty.
A: Customer lifetime value, retention rate, repeat purchase rate, referral rate, and customer satisfaction. These are harder to tie to a single ad, but they're what actually matters to the business.
A: No. Intuition is worse than imperfect data. But use data differently: run experiments, talk to customers, and let patterns emerge. Don't use data to claim certainty when uncertainty is honest.
A: Sophisticated, but not necessarily better at the core problem. A better AI model is still predicting on incomplete information. Where AI helps is spotting patterns in the data you do have and scaling experimentation faster. Use it for those things, not for magic. --- The hard truth: in 2026, the CMO who admits "I don't fully know what's working" is being more honest than the one who trusts their dashboard. The next competitive advantage goes to the teams who measure what actually matters, experiment relentlessly, and know the difference between correlation and causation. Your attribution model is a tool, not the truth. Start measuring like you know that.