Marketers: 2027 AI Impact on Incrementality

The digital advertising world is grappling with a monumental shift, making incrementality testing when AI agents strip UTMs and referrers a pressing concern for every performance marketer. With nearly 60% of web traffic now attributed to non-human entities, according to a recent report by Barracuda Networks, the traditional methods of measuring marketing effectiveness are crumbling. How can we possibly gauge the true impact of our campaigns when the very signals we rely on are being systematically erased?

Data Point 1: 58% of Web Traffic is Non-Human, and Growing

Barracuda Networks’ 2025 report on bot traffic stated unequivocally that 58% of all internet traffic originates from non-human sources. This isn’t just about malicious bots; it includes an increasingly sophisticated array of AI agents, scrapers, and privacy-focused browsers that actively obfuscate user data. My interpretation? This isn’t a fringe problem; it’s the new baseline. When more than half of the “users” interacting with your digital touchpoints aren’t real people, and many of the legitimate ones are using tools that strip identifying parameters, relying on client-side tracking for attribution becomes a fool’s errand. I had a client last year, a mid-sized e-commerce brand selling artisanal cheeses, who was pouring money into a niche social media platform. Their dashboard showed fantastic click-through rates, but sales weren’t budging. We dug in, and it turned out a significant portion of their traffic was from a new AI shopping assistant that visited product pages but never converted. It looked like legitimate traffic in their analytics, but it was just an agent browsing, not a potential customer. This taught me a hard lesson: vanity metrics are dead if you can’t verify the source.

Data Point 2: 70% of Marketers Still Rely on Last-Click Attribution

Despite the obvious flaws, a survey by Statista in late 2025 revealed that 70% of marketing professionals still primarily use last-click attribution models. This number, frankly, appalls me. It’s like trying to navigate a dense fog with a broken compass. Last-click attribution assumes that the final touchpoint before a conversion deserves all the credit. This model was already flawed in a multi-touchpoint world, but in an era where AI agents are actively scrubbing referrers and UTM parameters, it becomes actively detrimental. We’re giving credit to channels that might just be the last stop for an AI agent, or worse, misattributing conversions entirely. My professional take? This isn’t just inertia; it’s a dangerous complacency. If you’re still clinging to last-click, you’re making decisions based on data that is, at best, incomplete and, at worst, fabricated by non-human interactions. The marketing dollars you’re allocating using this model are likely misspent, inflating the perceived performance of certain channels while underfunding others that truly drive value.

Data Point 3: The Average E-commerce Site Loses 25% of Its Attribution Data to Privacy Features and Ad Blockers

A recent report from the IAB (Interactive Advertising Bureau) in early 2026 detailed that the average e-commerce website is losing approximately a quarter of its potential attribution data due to a combination of stricter browser privacy settings, ad blockers, and now, AI agents. This isn’t just about general trends; it’s about specific, measurable data loss. When I consult with clients, I often highlight this number as a baseline for what they should expect to be missing. It means that if your analytics dashboard shows 100 conversions, you could be missing the attribution path for another 25. Think about the implications for incrementality testing. If your control group is losing data at the same rate, but your test group is also being impacted by AI agents stripping those crucial UTMs, your “incremental lift” calculations become suspect. We ran into this exact issue at my previous firm when evaluating a new programmatic display campaign for a regional car dealership in Cobb County. The campaign showed a modest lift in reported leads, but when we cross-referenced with their CRM, the true incremental leads were significantly higher. The discrepancy? Safari’s Intelligent Tracking Prevention and various AI agents were preventing a clear attribution path for many initial website visits, making the display campaign look less effective than it truly was. It underscores the critical need to move beyond client-side data capture.

Data Point 4: Companies Implementing Server-Side Tagging See a 15-20% Improvement in Data Accuracy

According to a 2025 study by Google Analytics, businesses that transitioned to server-side tagging saw a 15-20% improvement in the accuracy and completeness of their first-party data capture. This is a game-changer for incrementality. By moving your tracking logic from the user’s browser to your own secure server, you can sidestep many of the issues caused by browser privacy settings, ad blockers, and those pesky AI agents. Imagine a scenario where an AI agent visits your site, but your server-side Google Tag Manager (GTM) container has already processed the initial referrer information before the agent has a chance to strip it. This allows you to retain valuable context. For brands serious about understanding true campaign performance, server-side tracking isn’t an option; it’s a mandate. Tools like Google Tag Manager’s server container documentation provide detailed steps on how to implement this. My advice? Start experimenting with this now. It requires a different skillset and often involves collaboration with development teams, but the data integrity benefits are immense.

Challenging Conventional Wisdom: The Death of the “Pure” A/B Test for Incrementality

Many marketers still preach the gospel of the “pure” A/B test – randomly splitting audiences and measuring the difference. While conceptually sound, this approach is increasingly compromised in the current environment. With AI agents stripping referrers and UTMs, and browsers aggressively limiting cookie lifespan, maintaining truly isolated control and test groups for attribution purposes is becoming a logistical nightmare.

Here’s my controversial take: the obsession with perfectly randomized, cookie-based A/B tests for incrementality is outdated and often yields misleading results in the age of AI agents.

Instead, we need to embrace more sophisticated, often messier, methodologies. I advocate strongly for geo-lift testing, where you segment regions based on similar demographic and behavioral characteristics, then run campaigns in one region while holding out another as a control. This bypasses the whole UTM stripping problem because you’re measuring aggregate sales or leads in a defined geographic area, not individual user journeys. For example, if you’re a national retailer, you might run a specific ad campaign only in the Atlanta metro area (Fulton, DeKalb, Gwinnett, Cobb, Clayton counties) and use the Charlotte metro area as your control, provided historical data shows them as comparable. You then compare sales lift in Atlanta versus Charlotte during the campaign period. Yes, there are confounding variables, but with robust statistical modeling and careful selection of control groups, it provides a much more reliable signal of true incrementality than a cookie-compromised A/B test. Another powerful approach is holdout testing at the channel or tactic level. Can you pause a specific ad type for a period and measure the impact on overall conversions? It’s not as clean as a user-level A/B test, but it provides a clearer picture of whether that ad type is truly driving incremental value or just cannibalizing other channels. The future of incrementality testing isn’t about perfect user-level attribution; it’s about smart, aggregate measurement that accounts for the inherent fuzziness of the modern data landscape.

Data Point 5: 35% of Digital Ad Spend by 2027 Will Be Verified via Data Clean Rooms

eMarketer predicts that by 2027, 35% of global digital ad spend will be verified through data clean rooms. This is a significant shift. Data clean rooms, offered by platforms like Google Ads Data Hub documentation or various third-party providers, allow advertisers to securely combine their first-party data with publisher data for measurement and analysis without exposing individual user identities. This is paramount for understanding incrementality when traditional attribution signals are degraded. When AI agents strip UTMs and referrers, the ability to match anonymized user IDs or hashed email addresses within a privacy-safe environment becomes critical. This allows you to say, “Okay, this segment of users exposed to my campaign eventually converted, even if I don’t know the exact click path.” It’s not perfect, but it’s a massive leap forward in preserving measurement integrity. I strongly advise marketers to start exploring data clean room solutions now. Understanding how to query and interpret data within these environments will be a core competency for any performance marketer in the next few years. It’s complex, yes, but ignoring it means operating in the dark.

The challenge of incrementality testing when AI agents strip UTMs and referrers is not a passing trend; it’s a fundamental shift demanding a complete overhaul of how marketers approach measurement. By embracing server-side tracking, adopting geo-lift and channel-level holdouts, and investing in data clean rooms, you can move beyond the unreliable signals of the past and build a robust framework for understanding your true marketing impact.