‍Idea Peddler Founder Cimin Ahmadi Cohen on Deconstructing Data

Media Transparency · Programmatic Advertising · Attribution · Marketing Data · Agency Accountability

If your ROAS keeps climbing while your business results stay flat or quietly decline, especially in the competitive ecommerce space, you are not alone. And according to Cimin Ahmadi Cohen, Founder and CEO of Idea Peddler, you are probably not looking at the real problem.

Cimin recently joined David Finkelstein and Jessie Lezak on Deconstructing Data, the BDEX podcast that helps marketers understand how data actually drives (and sometimes undermines) marketing decisions. The episode, titled The Attribution Illusion: Why Your Data Looks Better Than It Is, is one of the most candid conversations in recent memory about what is broken in media measurement and what it actually takes to fix it.

What Is the Attribution Illusion?

The concept came out of something Cimin and her team kept running into in the field: a persistent, uncomfortable gap between what the numbers said and what was actually happening in the business.

"There was kind of this explosion of ROAS being used to measure success, almost myopically so," she explained on the show. "But then we noticed a couple of years ago there started — or stopped, rather — being a correlation between ROAS and the business outcomes. Return on ad spend was going up and up and up, and business outcomes—and more importantly, profit margin—would remain flat or even down in some cases. The maths weren't mapping."

To get to the bottom of it, Idea Peddler did something unusual. They hired a New York Times investigative journalist to help synthesize their first-party research, industry data, and field experience into a coherent picture of how and why attribution so consistently overstates performance. The result was their Attribution Illusion white paper, which distills the structural problems into three core findings.

The Three Forces Behind the Illusion

1. Suspicious Segments

Privacy legislation, Apple's ATT update, and the erosion of third-party identifiers have quietly gutted the addressability of most precision audience segments. Few vendors are being honest about it.

Cimin described what they found: panel data sizes cut in half, footfall attribution falling to 1% of actual visitation, and log file transfer match rates shrinking to fractions of what they once were. "There's a massive drop-off in the actual match rate that you can glean depending on your data source and data provider," she said.

The result is what she calls the ad industry's "junk food problem." When a vendor cannot address a meaningful portion of a precision audience, they fill the gap with bots, generic lookalikes, or whatever gets the campaign to scale. "It's become very easy to say, 'I want a precision audience targeting luxury travelers interested in African safari,' and just trust that the vendor on the other side is delivering that audience. But what we're finding is that there's only a fraction of the audience that can be matched in that way, and everything else is getting filled in with filler and fluff data."

There is also a compliance risk most brands are underestimating. "You can't trust every data provider and every media vendor that they are abiding by the laws and regulations" around mobile geolocation, Cimin noted. She cited Uber Media, covered in a Fast Company article around the time of its bankruptcy, as a prominent example of a vendor operating outside the law on PII and geolocation data. With states like Oregon and Vermont now banning the sale of mobile geolocation data entirely, the exposure is real and growing. Sephora's million-dollar-plus fine for privacy non-compliance puts a concrete number on what that exposure looks like.

2. Vanishing Signal

The second force is what Cimin calls vanishing signal: the degradation of the data infrastructure that used to make precision targeting work.

"The single source walled garden grading of homework was not accurate. It was overly inflated." But beyond just overstating results, she argued that relying on any single data source carries genuinely negative long-term consequences for a brand.

Meta and Google's lookalike targeting once rested on robust infrastructure. iOS 14 changed that materially. There is also a secondary problem that David Finkelstein added detail on: when a user opts out of tracking, their mobile ID refreshes repeatedly, creating multiple non-addressable IDs that still get categorized and rolled into audience pools. Ad servers build campaigns against what looks like an audience of ten million people, while only a small fraction are actually reachable. The campaign fills in the rest with whoever it can find.

"You think you're reaching a specific audience, but you're really not," Finkelstein said. Cimin acknowledged the same, noting that most ad platforms have no financial incentive to clean up the problem even when solutions exist.

3. Phantom Privacy

The third finding is perhaps the most under-discussed. It is the illusion that compliance is happening even when it is not being scrutinized closely enough to know for certain. Cimin described it as a "phantom privacy reality" that brands have been too comfortable accepting.

The implication is direct: it is on the brand and the agency to ask these questions explicitly rather than assume everything is above board.

On AI: Scaling the Problem, Not Solving It

The conversation turned pointed when Jessie Lezak asked whether agentic media buying might solve some of the targeting problems they had been discussing. Cimin's answer was one of the clearest-eyed takes on AI in advertising you will find anywhere right now.

"I remain a little bit suspicious until proven otherwise that AI is going to be a panacea for ad targeting problems. We all know the rule of garbage in, garbage out. So if a lot of the data is stuffed with filler data, that just doesn't scale. Agentic being brought in to scale bad data feels like the most likely outcome."

She named the incentive problem directly. Platforms like Google have little motivation to clean up bot farms if those bots make their CPC numbers look better. "It's far easier to say AI's got it handled. The reality is I don't trust that AI has a good ability to weed that inventory out of the system, which is just total ad waste."

Her practical position: Idea Peddler is actively testing AI-driven tools, including Google AI Max and AI search products, but always measures results against holistic, independent data rather than platform self-reporting. "When we do that, we're looking at what is the business outcome of that, as opposed to what is a platform telling us."

The real promise of AI, she suggested, is still ahead of us. When platforms are built AI-first rather than having AI bolted on, the technology might genuinely learn in real time what is working without compromising on privacy compliance. Getting there requires infrastructure that does not yet widely exist.

Directional Data vs. Decision-Grade Data

One of the most actionable parts of the conversation was Cimin's framework for understanding what data is actually worth acting on.

Decision-grade data requires at minimum two sources, and at least one must be genuinely independent. "Everyone grading their own homework will obviously give themselves an A," she said. The second source does not have to be expensive. It can be a marketing mix model, a qualitative study, or a quantitative survey. What matters is that it comes from a source with no financial stake in the outcome. Her rule of thumb: allocate two to five percent of media budget to unbiased measurement before optimizing anything.

Directional data (platform dashboards, lookalike audiences, self-reported metrics) can still be useful, but only with discipline. Her team uses lookalike data within platforms like Meta to compare segments against each other rather than treating the outputs as absolute truth. Texas versus California. High-value customer type A versus type B. The goal is to run structured hypotheses regarding metrics like CAC rather than accept the platform's headline number.

One warning she flagged specifically on lookalikes: in Meta, lookalike audiences always appear to outperform other segments, not necessarily because they are more accurate, but because you cannot take them elsewhere. It is a lock-in mechanism. Understanding why the numbers look good matters as much as the fact that they do.

The Bottom Line

What Cimin described on Deconstructing Data is not a cynical indictment of digital advertising. It is a practitioner's map of where the system breaks down, built from years of noticing that the maths were not mapping.

The attribution illusion persists because it is comfortable. Dashboards with improving numbers are easy to present in a meeting. Asking harder questions about match rates, about who is actually measuring performance, about what gets stuffed into a precision segment when the addressable audience runs out, takes more work and sometimes produces less flattering answers.

The business that operates on accurate information makes better decisions than the one running on optimistic dashboards. That much has not changed.

As Cimin put it near the close of the episode: "The first step is acknowledging this is real. And to just try to start to do better bit by bit rather than thinking we're going to stop this overnight."

Listen to the full episode of Deconstructing Data, The Attribution Illusion: Why Your Data Looks Better Than It Is, on Apple Podcasts, Spotify, and bdex.com/deconstructing-data-podcast. New episodes stream live every Thursday at 4:15 PM ET on LinkedIn and YouTube.

Connect with Cimin on LinkedIn and learn more about Idea Peddler at ideapeddler.com.

‍