By Vlad Zhovtenko, CEO and co-founder of RedTrack
Digital advertising runs on data. Data makes decisions, budgets, and entire careers possible. And it is poorly understood.
In an advertising context, there is no fixed idea of data. “Data” refers to whatever an ad platform decides is data on a given day, based on whatever attribution model, deduplication rule, and reporting policy they’ve chosen to apply. There’s something absurd here, not unlike Humpty Dumpty in Through the Looking Glass telling Alice that any word he chooses “means just what I choose it to mean — neither more nor less.” Marketers are on the receiving end of ad platforms that work this way. Whatever the platform decides “data” means is the data they get.
Most advertisers do not devote serious thought to the subjectivity of their data. They believe they are running a data-driven operation. They’re in fact running an interpretation-driven one, data as interpreted by the parties they buy traffic from.
Metrics aren’t data (and vice versa)
There’s an overlooked distinction between metrics and data.
Metrics are processed outputs. ROAS, CPA, conversion counts, blended cost figures, and dashboard cards are not the underlying record of what happened. They record how an ad platform read the raw data and subsequently applied its own attribution model, deduplication rule, and reporting window.
The actual event record sits underneath. For instance, the click landed on the site at noon, but the impression was served two days earlier. The conversion fired with its full set of identifiers, before any ad platform decides which campaign should get credit. Customer journey artefacts build up across sessions and devices, and most advertisers never see them. They have only ever seen the ad platform’s reading of it.
When advertisers say they are data-driven, what they are running on is a derivative (or a derivative of a derivative). The ad platform’s reports are aggregated outputs of attribution logic applied to events invisible to the advertiser. Usually there is no second copy to compare the platform’s reports against. The ad platform’s number is the only number, so advertisers have no choice but to treat it as truth.
Why the status quo stopped working
The arrangement worked when humans ran campaigns by hand. A media buyer received a partial and biased report from each ad platform. They cross-checked the numbers against site behaviour and adjusted bids and budgets. The buyer noticed odd behaviour and compensated for flawed input.
Now the algorithm inside each ad platform consumes the same flawed input directly, at machine frequency, with no human in between. The signal goes from the ad platform’s interpretation of events straight into the bidding system, which scales spend toward whoever fits the pattern that interpretation describes. If a human in the loop made bad data visible, then an algorithm in the loop makes it invisible.
Bad inputs fail and scale quietly. The same flawed signal that once produced visible problems in a hand-run campaign now produces more spend pointed at a noisy proxy of the right buyer. The damage shows up later in retention and lifetime value (LTV), not in the dashboard the buyer is reading.
To test where they actually stand, advertisers must answer three questions.
The first is to determine whether an event was recorded on the ad platform’s side or in a system they control. The second is to determine whether the data is stored by the advertiser or someone whose business model depends on the advertiser not having a clean copy. The third is to determine whether they can export and use it independently of the system that captured it.
If the ad platform comes out on top of any of these, the ‘data-driven’ claim doesn’t hold.
The events that drive every decision are held by parties whose interests are not the same as the advertiser’s. The interpretation of those events is delivered back as a finished product, not a working set.
Whatever comes next happens on the ad platform’s terms. The advertiser is the platform’s customer, and the platform will behave accordingly.
Optimisation absent control
Without an independent record, an advertiser cannot deduplicate conversions across ad platforms. Each ad platform sees only its own claim and logs it as a win.
If Meta, Google, and an affiliate network all claim the same conversion, an advertiser can’t settle the dispute unless they have a separate copy. Reported conversions usually exceed real conversions, and the gap widens with every ad platform added to the mix.
Since the ad platform’s model is the only model the advertiser sees, its choices about lookback windows, view-through credit, and assist weighting become the advertiser’s choices by default.
When the number on the dashboard is the ad platform’s preferred reading of the outcome, the algorithm within it learns from a signal biased toward that ad platform. It scales spend toward whatever the flawed reading marks as a winner. The reporting bias now shapes the very bidding system that allocates the next dollar.
The advertiser pays for this twice. First, in spending pointed at the wrong audiences, and again in the worsening decisions that compound on top of those allocations, month by month.
Data as control
Owned data is the only fixed point an advertiser actually has. If you own your data, it is an operating input. It gives an advertiser real influence over the bidding system, rather than the appearance of it.
From there, an advertiser can compare performance across ad platforms using the same evidence on each side, rather than trust each ad platform’s reading of its own contribution. The advertiser can push that same evidence back into ad systems as a conversion signal rather than sending whatever the pixel happens to capture. They can accumulate a private history of their own buyers, and know exactly when events happened.
That captured record is what you work from. Every month of events held in that system makes the next month’s decisions sharper until the captured history is itself a piece of the business no ad platform can take back.
When you feed the algorithm inside platforms your own capture events, reconciled and held in a system you control, you can route data back into ad systems on your own terms.
Verified first-party signals versus platform-reported data is now the dividing line. And if you’re still feeding the algorithm whatever the pixel happens to deliver once the ad platform has run it through its own interpretation layer, you’re operating on the wrong side of it.
About the author
Vlad Zhovtenko is the CEO and co-founder of RedTrack, the unified control plane e-commerce brands and media buyers use to run, improve, and scale paid acquisition. With over two decades in digital advertising, he has become one of the more vocal advocates for advertiser-side data ownership, arguing that as platform AI takes over budget allocation decisions, the conversion signal advertisers feed those systems has become the primary variable they still control. Before RedTrack, Vlad co-founded AdxGeeks, a programmatic media-buying platform, and advised companies on performance marketing strategy and lead generation operations.
