I’ve had the same conversation with every CMO I’ve met over the past five years. Usually over coffee, after the formal meeting ends, they’ll lean in and admit something: they don’t really trust their attribution model. They use it. They present it to the board. They make decisions based on it. But deep down? They know it’s showing them a version of reality that’s at best incomplete, at worst completely backwards.
Here’s what nobody wants to say out loud: attribution modeling isn’t failing because we need better technology or more data. It’s failing because we’re trying to use spreadsheet logic to understand human behavior, and frankly, most of us are more invested in models that justify our current budgets than ones that challenge them.
We Have More Data and Less Confidence
There’s something deeply weird happening right now. We have more tracking pixels, more sophisticated algorithms, and more computing power than ever before. Marketing technology has never been more advanced. Yet according to Gartner’s 2023 research, only 14% of marketing leaders say they have “high confidence” in their attribution insights. That’s down from 23% just three years ago.
Think about that for a second. As the tools get better, trust is going down. That’s not a technology problem. That’s a fundamental mismatch between what we’re trying to measure and how people actually make decisions.
The Customer Journey Doesn’t Exist (Not the Way You Think)
Attribution models treat buying decisions like GPS directions. Point A to Point B to Point C. Click this ad, see that email, convert here. The model crunches the numbers, assigns weights to each touchpoint, and outputs a clean percentage breakdown of what drove the sale.
But people don’t work like this. Let me walk you through a real customer journey from a campaign we ran last quarter:
Tuesday morning, someone scrolls past our client’s Facebook ad. Doesn’t click. That evening, they mention the product to their spouse over dinner. Wednesday, the spouse gets curious and searches on Google, clicks a paid search ad. Thursday, the original person sees an Instagram Story from the brand. Friday, they’re browsing Reddit and see the product mentioned in a thread they were already reading. Saturday morning, they finally Google the brand name directly and buy.
Now, tell me: what caused that conversion?
Your attribution model never saw the dinner conversation. It probably missed the Reddit mention unless you’re running promo codes. It might give all the credit to that final branded search, or maybe it splits credit between Facebook, Google, and Instagram using some formula that was probably chosen because it made the reporting look reasonable.
But here’s what really happened: Facebook planted the seed. The spouse conversation provided social proof. The search ad legitimized the product at the moment of curiosity. Instagram was a timely reminder. Reddit offered third-party validation. The branded search was just the final step in a process where every touchpoint mattered but none of them would have worked alone.
This is where the whole thing breaks down. We’re trying to isolate cause and effect in a system where everything influences everything else.
The Uncomfortable Truth About Budget Politics
Let me tell you something that doesn’t get said in conference rooms: most companies don’t actually want accurate attribution. They want attribution that confirms their existing organizational structure.
Picture a typical marketing department. You’ve got:
- A search team running $2M in Google Ads
- A social team managing $1.5M across Meta platforms
- A video team with $800K on YouTube
- A programmatic team handling $1.2M in display
Each team has managers, specialists, dashboards, and KPIs. They’ve built their careers on specific metrics. Their bonuses depend on CPCs and CTRs and channel-specific ROAS.
Now imagine your new attribution model reveals that YouTube is actually responsible for 60% of your conversions, while search is mainly just capturing demand that other channels created. What happens?
In theory, you shift the budget. In practice, you question the model.
This is why last-click attribution stuck around for so long. Everyone knew it was wrong. But it was simple, defensible, and it made the search team look like heroes. The people with the power to change the model were the same people benefiting from keeping it.
Modern multi-touch attribution is more sophisticated, sure. But it suffers from the same political reality. We unconsciously tune our models to match our org charts rather than our customers’ actual behavior.
The Feedback Loop Problem
Here’s something that keeps me up at night: attribution models learn from attributed conversions. That creates a self-reinforcing loop that gets further from reality over time.
Watch how this plays out:
- You run ads across multiple channels
- Your model tracks what it can track-clicks, impressions, pixel fires
- It assigns credit based on these visible interactions
- You optimize toward the channels getting credit
- Those channels generate more attributed conversions
- The model “learns” these channels are most valuable
- You double down on them
Meanwhile, what’s invisible to the model?
- Water cooler conversations about your brand
- Podcast ads that don’t include tracking URLs
- Billboard impressions during someone’s commute
- Word-of-mouth from great customer service
- PR coverage that builds background awareness
- Organic social discussions you’re not part of
- That newsletter from another brand that happened to mention you
- Your competitor’s awful experience that made you look good by comparison
Your attribution model is like someone looking for their keys under a streetlight. Not because that’s where they lost them, but because that’s where they can see.
The scary part? As you optimize toward the channels your model can see, you might be slowly killing the channels that are actually doing the heavy lifting.
The iOS 14.5 Lesson We Already Forgot
When Apple dropped App Tracking Transparency in 2021, it should have changed everything. Overnight, massive amounts of data vanished. Facebook’s pixel got kneecapped. Cross-device tracking fell apart. The machine that powered a decade of “data-driven” marketing stopped working.
For about six months, there was real panic. Real soul-searching. Brands were forced to acknowledge how much they’d been depending on surveillance-level tracking to make decisions.
But here’s what actually happened: after the initial shock, most brands didn’t fundamentally rethink anything. They just found workarounds. Server-side tracking. Probabilistic matching. Conversion APIs. First-party data strategies with increasingly creative definitions of “consent.”
We treated it as a technical problem to engineer around rather than a wake-up call about the limitations of deterministic attribution.
The question we should have asked but didn’t: If we can’t precisely track individual journeys anymore, should we have been making decisions as if we could in the first place?
A Better Way: Think Incrementality, Not Attribution
After spending the past several years running campaigns with monthly budgets ranging from $50K to over $2M across every major platform, I’ve learned something crucial: the brands that consistently grow aren’t the ones with the most sophisticated attribution. They’re the ones who stopped trying to attribute every conversion and started thinking about incrementality instead.
Here’s the shift:
Old Way of Thinking:
“This Facebook campaign generated 500 conversions at $50 CPA according to our last-click attribution model.”
New Way of Thinking:
“When we increased Facebook spend by 30%, total conversions across all channels went up 12%, suggesting Facebook is creating demand that shows up everywhere.”
This requires actually testing things, which is slower and messier but gets you closer to truth:
Geo-Holdout Tests
Run campaigns in some markets but not others. Compare total business performance, not just attributed conversions. Yes, it takes longer. Yes, it’s less clean. But it shows you what’s actually incremental versus what’s taking credit for sales that would have happened anyway.
Budget Variation Tests
Systematically turn spend up and down by channel. Watch what happens to total revenue, not just the conversions that platform claims credit for. The relationship between spend changes and business outcomes tells you what’s really working.
Brand Lift Studies
Measure whether awareness, consideration, and preference are actually moving. Not everything that matters shows up as an immediate conversion. Some channels build equity that pays off over time.
Just Ask People
Survey your customers. Ask them how they heard about you and what convinced them to buy. It’s imperfect-memory is unreliable-but humans are often better at understanding their own decision-making than algorithms are. Use these insights to reality-check your quantitative models.
How We Actually Approach This
At Sagum, we’re borderline obsessive about data. Every client gets a custom BI dashboard. We track everything that can be tracked. We built our entire operation around a data-first approach.
But we’ve also learned that the most valuable insights come from combining rigorous analysis with genuine empathy for how customers actually think.
Here’s our framework:
Layer 1: Platform Data
Track what Facebook, Google, TikTok, and Pinterest tell you. But understand that every platform is designed to justify its own existence. They all overstate their impact. That’s not dishonesty-it’s just the nature of last-touch measurement in a multi-touch world.
Layer 2: Unified Analytics
Pull everything into one place so you can see the full picture. We use Grow for this because it lets us combine data from every source. When total revenue moves but attributed conversions don’t, that’s telling you something important.
Layer 3: Incrementality Testing
Run controlled experiments. Pause channels temporarily. Vary budget levels. Use geographic splits. These tests are expensive and take time, but they reveal what’s actually driving incremental growth versus what’s just intercepting existing demand.
Layer 4: Customer Understanding
Talk to real customers. Survey them. Interview them. Understand their decision process in their own words. This qualitative context helps you interpret all the quantitative signals correctly.
Layer 5: Synthesis
Combine everything into a probabilistic understanding of what’s working. Accept that you can’t know everything with certainty. Make decisions based on weight of evidence from multiple sources, not false precision from one flawed model.
The Channels Attribution Gets Most Wrong
Based on running thousands of campaigns across every major platform, here are the channels that standard attribution consistently misunderstands:
Most Undervalued: YouTube Pre-Roll
YouTube typically gets evaluated on view-through conversions, which most attribution models either ignore or heavily discount. But we’ve run enough incrementality tests to know that YouTube drives enormous lift in branded search, direct traffic, and performance across other channels.
The reason is simple: video creates memory. Someone doesn’t click your YouTube ad, but three days later when they encounter your product somewhere else, they think “oh yeah, I’ve heard of that.” That familiarity reduces friction.
Traditional attribution gives YouTube almost no credit for this. Incrementality testing often reveals it’s one of the most efficient channels for creating new demand.
Most Overvalued: Prospecting Search
Branded search ads deserve every penny they get. But broad prospecting search? Attribution models love it because it’s often the last click before conversion.
But here’s the actual sequence: someone becomes aware of their problem through social content, learns about potential solutions through video or display, develops intent, and then searches for product categories and clicks an ad.
The search ad gets all the credit. The channels that created awareness and consideration get nothing. You optimize toward search, starve the upper funnel, and wonder why growth plateaus.
Most Complicated: Retargeting
Retargeting is simultaneously overvalued and undervalued depending on how you measure it.
Overvalued: On a last-click basis, retargeting looks incredibly efficient. But you’re mostly showing ads to people who were already coming back. The conversion would have happened anyway.
Undervalued: If you only count truly incremental conversions, you might cut retargeting entirely. But smart retargeting actually shortens consideration time and reduces cart abandonment in ways that help the whole funnel work better.
The key is frequency capping, creative variation, and measuring time-to-conversion rather than just conversion rate.
Most Invisible: Creative Quality
Here’s something attribution models can never capture: the performance difference between mediocre creative and exceptional creative in the same channel with the same targeting.
We’ve seen identical audience strategies produce 3-5x different ROAS based purely on creative quality. But attribution models treat all impressions equally. They can’t tell you that your competitor’s ad is more effective because it’s more emotionally resonant or more memorable.
This is why the industry’s obsession with targeting and attribution has coincided with a noticeable decline in creative quality. We’re optimizing what we can measure-audience, placement, bid strategy-while ignoring what we can’t: emotional impact, memorability, persuasiveness.
The Privacy Future Is Actually Good News
Here’s what everyone knows but few are preparing for: attribution is going to get harder, not easier.
Privacy regulations keep expanding. Browsers keep deprecating cookies. Platforms keep locking down data. The surveillance infrastructure that powered attribution for the past decade is going away, and it’s not coming back.
Most agencies treat this like a crisis to manage. They’re investing in workarounds and finding clever ways to keep tracking people who increasingly don’t want to be tracked.
But here’s a contrarian take: this might actually force us to make better marketing decisions.
When you can’t track every click, you’re forced to:
- Build distinctive brands that create lasting memory structures
- Focus on actual business outcomes like revenue and customer lifetime value
- Think strategically about channel mix instead of tactically about optimization
- Create content people actually want to engage with rather than ads they tolerate
The brands that win over the next decade won’t be the ones with the most sophisticated tracking. They’ll be the ones who build real relationships with customers, create genuinely memorable experiences, and understand that marketing is about influencing behavior, not just measuring it.
What to Do Starting Tomorrow
If you’re reading this thinking “okay, but I still have to allocate budget on Monday,” here’s what I’d recommend:
This Month:
Stop presenting attribution as fact. In every report, acknowledge the limitations. Present attribution data as one perspective among several, not as objective truth.
Add business metrics to every dashboard. Track total revenue, total conversions, and customer acquisition trends alongside channel-specific metrics. When the two tell different stories, investigate why.
Survey 50 recent customers. Ask simple questions: “How did you first hear about us?” and “What convinced you to buy?” You’ll be surprised how often the answers contradict your attribution model.
Next Quarter:
Run your first incrementality test. Pick your biggest channel and vary spend by 20-30% for a month. Measure impact on total business performance, not just attributed conversions. The results will probably surprise you.
Consolidate your data. Get everything into one dashboard so you can spot cross-channel patterns. We use Grow because it creates an environment where data drives productive conversations and tests rather than just justifying decisions already made.
Build a testing calendar. Schedule ongoing experiments: new channels, creative variations, audience tests, incrementality studies. Treat your marketing like a lab where you’re constantly learning, not a vending machine where you expect predictable outputs.
Long Term:
Shift from attribution to contribution thinking. Stop asking “which channel gets credit?” Start asking “how does each channel contribute to the system as a whole?”
Invest in brand measurement. Track awareness, consideration, and preference over time. Understand that marketing impact doesn’t always show up in the same period as the marketing activity.
Build a culture of skepticism. Reward people who question data, not just people who hit their numbers. The goal is understanding reality, not confirming assumptions.
Accept uncertainty. The most sophisticated approach to attribution is acknowledging that customer journeys are complex, multi-channel, and partially unknowable. Make decisions based on weight of evidence from multiple imperfect sources rather than false precision from a single flawed model.
The Real Bottom Line
Attribution modeling isn’t broken because we haven’t found the right algorithm yet. It’s broken because we’re asking it to provide certainty about something inherently uncertain: why individual humans decide to buy things.
The marketers I know who consistently drive growth year after year share a common trait: they use attribution data as one input among many, never as gospel. They combine quantitative analysis with qualitative insights, controlled experiments with customer empathy, sophisticated modeling with old-fashioned common sense.
They understand that marketing is both art and science, and that the most important things-emotional resonance, brand perception, customer loyalty-are precisely the things that are hardest to measure and attribute.
So yes, use your attribution model. Build it, refine it, learn from it. But never forget it’s a map, not the territory. And sometimes the most valuable insights come from putting down the map and actually talking to the humans you’re trying to reach.
After all, if you really want to understand why someone bought from you, there’s a revolutionary attribution method that’s been available this whole time: you could just ask them.
