Fairing CEO Matthew Bahr on the Limits of AI Attribution
As Media Mix Modelling dominates the post-cookie narrative, Fairing CEO Matthew Bahr argues that predictive AI and pixel tracking still miss the most critical measurement signal: the customer’s own account of the purchase journey.
Topics
Marketing attribution is currently caught in a tug-of-war between two extremes. On one side, the industry is rushing toward AI-driven Media Mix Modelling (MMM) to bandage the signal loss left by cookie deprecation and stringent iOS privacy frameworks. On the other, brands are grappling with the reality that discovery increasingly happens in unmeasurable spaces—podcasts, closed communities, and creator ecosystems where pixels simply do not fire.
Matt Bahr, Co-Founder & CEO of Fairing, operates in the critical space between these extremes. His company has built its architecture on buyer-declared, zero-party data, betting that a direct question often yields more strategic accuracy than a complex predictive algorithm. Yet, as Fairing deepens its server-side integrations with giants like Meta and pushes into AI-assisted categorisation, it faces a shifting landscape where data sovereignty is fiercely protected, and native ecosystem widgets threaten commoditisation.
In this interview, Bahr dissects the limitations of machine learning in attribution, the ongoing friction between enterprise data ownership and platform algorithms, and why the last measurable click is rarely the one that actually closed the sale.
The industry is currently obsessed with Media Mix Modelling (MMM) and AI-driven predictive attribution to solve the signal loss from iOS 14 and cookie deprecation. Fairing champions buyer-declared, zero-party data. Do you see self-reported data as a permanent pillar of the marketing stack, or is it just a stopgap until AI attribution gets ‘perfect’?
Self-reported data needs to be a permanent part of the measurement stack because it captures something models and pixels can’t directly observe: what customers say actually influenced their decision.
There’s no single measurement methodology that gives brands the complete picture. MMM, incrementality, platform attribution and customer-reported data all answer different questions, and they’re most useful when you bring those signals together rather than expecting one methodology to become the single source of truth.
AI will make attribution models better, but it doesn’t remove the underlying measurement problem. As more discovery happens through creators, podcasts, CTV, and AI tools like ChatGPT and Claude, there will continue to be influences that don’t leave a reliable digital trail. Self-reported data helps fill those gaps by giving brands a stable signal directly from the customer.
Fairing recently rolled out a Meta CAPI integration, allowing brands to sync survey responses directly back into Meta’s ecosystem. While this improves short-term ad targeting, doesn’t it defeat the broader enterprise goal of data sovereignty? Are brands essentially handing their most valuable zero-party data back to the algorithms they are trying to gain independence from?
Nearly every major brand already uses Meta’s Pixel and Conversions API, which gives Meta broad visibility into customer purchases. Those integrations exist because brands want Meta to optimise campaigns toward the outcomes that matter to them, whether that’s a sale, an app install or something else.
Fairing adds a signal that comes directly from the customer on top of those integrations. It also helps Meta credit conversions it couldn’t match on its own, for example, when an ad blocker or browser privacy settings stop the pixel from firing, or the purchase happens on a different device than the ad click. There’s a real performance benefit to giving platforms more conversion data, though brands should always opt in with a clear understanding of the data rights they’re granting.
Data sovereignty is really about where the complete dataset lives. A brand’s full survey history, including responses about podcasts, creators, CTV and word of mouth, stays with the brand in Fairing, and the brand decides which partners it flows to. Meta gets a signal that it can optimise against; it doesn’t get the whole picture.
The channels that introduced customers to the brand get undercounted, while those that closed the sale get overcredited.
At Advertising Week NY, you advocated for pixel-based attribution for podcast ads. As historically hard-to-measure channels like audio and CTV become more natively measurable, does the reliance on the ‘How Did You Hear About Us’ survey diminish?
Better pixel-based measurement doesn’t make HDYHAU (“How did you hear about us?”) surveys less relevant. It gives marketers another signal to combine with what customers tell them. The strongest approach is to triangulate multiple sources rather than choosing one methodology over another.
Pixels are very useful for identifying measurable exposure and conversion behaviour, but customer-reported data can capture influence that doesn’t leave a digital trail. For example, someone hears about a brand on a podcast or from a creator and buys weeks later.
As these channels become more complex, we’re taking HDYHAU further. Advanced Attribution moves beyond broad responses such as “Podcast” or “YouTube” to identify the specific show, creator or source that influenced the purchase. Importantly, it reflects how consumers actually remember an interaction rather than forcing them into a marketer’s channel taxonomy, turning those responses into structured, actionable attribution data.
Competitors in the zero-party data space are heavily touting dynamic, individualised survey logic over what they call ‘one-size-fits-all’ approaches. Meanwhile, platforms like Shopify could theoretically build native, ‘good enough’ post-purchase widgets. What is Fairing’s true moat today—is it the data collection mechanism itself, or the analytics layer you process it through?
It’s not one or the other. Accurate survey data depends on three things: how you collect the data, how you clean and normalise it, and how that data can then be used. We’ve invested heavily across all three layers.
On collection, Fairing surveys have been viewed more than one billion times across 3,500+ brands. That scale, and more than seven years of testing, allows us to continually improve question design and the survey experience across different devices and surfaces. Our integrations with platforms including Vibe, Podscribe and Agentio can also automatically update survey options based on where a brand is actually spending.
Then there’s the work that happens after someone responds. Brands can bulk-reclassify “Other” and write-in responses, and we’re building towards rules-based and AI-assisted reclassification. We can also resolve different customer responses back to the same placement – so, for example, “Facebook > Influencer > @BasketballGirl” and “Influencer > @BasketballGirl > Facebook” become one canonical answer.
Finally, every response is tied to the order and its revenue data, and that clean data can feed into more than a dozen media and measurement partners, including Meta, TikTok, Vibe, Agentio, Podscribe, Triple Whale, Fospha, WorkMagic and Google Analytics. A “good enough” widget can collect an answer. The harder part is turning those answers into accurate, usable signals that can inform measurement and campaign optimisation.
Because Fairing collects buyer-declared data across thousands of brands, you have a broad view of the ecommerce landscape. When you compare what standard analytics platforms report with what customers say actually influenced their purchase, what is the biggest discrepancy you’re seeing in marketing measurement today?
The biggest discrepancy is where credit lands. Standard analytics platforms tend to credit the last measurable touch, usually branded search, direct traffic or a retargeting ad. When we ask customers what actually influenced them, they often point somewhere else entirely: a podcast, a creator, a friend’s recommendation or something they saw weeks earlier. The channels that introduced customers to the brand get undercounted, while those that closed the sale get overcredited.
That gap shows up most in the channels brands are spending more on today, which also happen to be the hardest to measure: podcasts, creators and CTV.
No single methodology gives brands the complete picture, and buyer-declared data isn’t meant to replace the others. It adds the customer’s own account of what drove the purchase, which is often the part dashboards can’t see. That helps marketers understand what actually influenced a purchase and make more informed decisions about where they invest. Without it, it’s easy to cut the channels building demand because the report credits the ones capturing it.
ALSO READ: Mixpanel’s CEO Jen Taylor on Turning Technology into Measurable Business Growth
