How to measure conversions without user-level identifiers

A city traffic engineer who wants to know whether a new highway on-ramp is working doesn't need to follow every car home. She can't see where each driver started or where they're headed. What she can see is how many cars pass each road sensor every hour, how that changed after the ramp opened, and how it compares with rush hours, holidays, and the week a stadium concert snarled everything downtown. Get the patterns right and she can tell how much traffic the ramp actually moved, without knowing a single driver's name.

Marketing measurement has landed in the same spot. Privacy rules and platform changes have made it much harder to follow individual shoppers from ad to purchase, and the tools that depended on that tracking now miss big chunks of the journey. But the revenue still shows up in your storefront, on Amazon, and at retail every day. Brands that learn to measure from those aggregate patterns can keep making confident budget calls, while brands still waiting for user-level tracking to come back are steering by numbers that undercount their best campaigns and overcredit the ones standing closest to the checkout.

Key takeaways

• User-level tracking has become less reliable because of browser restrictions, Apple's App Tracking Transparency, and privacy laws, and multi-touch attribution has drifted toward over-crediting last-touch channels as a result.
• Privacy-preserving measurement isn't always identifier-free, since clean rooms and platform-modeled conversions often still rely on matched IDs behind the scenes.
• A tracked conversion doesn't prove an ad drove it, so the real goal is estimating how much revenue each campaign actually added.
• Marketing mix modeling measures conversions from aggregated spend and revenue over time, but traditional and open-source MMMs often struggle when spend rises and falls with demand, which leads to baseline leakage.
• Prescient's marketing mix model treats demand as something marketing builds and fades over time, so it can separate baseline demand from marketing impact and capture halo effects without user tracking.
• A model that predicts total revenue well can still split credit between channels incorrectly, so accuracy checks need to test how credit is assigned, not just how well the totals line up.
• Platform attribution is still useful for optimizing inside a platform, but budget decisions across channels need independent, aggregate measurement.

Why user-level attribution stopped telling the full story

For years, digital attribution worked by stitching together a person's journey through cookies, device IDs, and pixels. That approach was never perfect, but a series of changes has made those stitches much weaker. It's worth being precise about what actually changed, because "cookies are dead" isn't quite right.

The result isn't a single day when tracking disappeared. It's a patchwork where some journeys are visible, many are partly visible, and a lot are invisible, which is harder to work with than a clean break.

Multi-touch attribution drifted toward last-touch bias

Multi-touch attribution (MTA) was supposed to spread credit across every touchpoint in a customer's path. When the identity chain breaks, though, the touchpoints it can still see tend to be the last ones before purchase, like a branded search click or a retargeting ad. Upper-funnel channels like YouTube, CTV, and TikTok often reach people who never click, so they lose credit even when they're creating the demand.

That leaves MTA over-crediting demand capture and starving demand creation. Brands that follow those numbers end up shifting budget toward the campaigns that look closest to the sale and away from the ones filling the funnel.

Privacy-preserving doesn't always mean identifier-free

A lot of tools now describe themselves as privacy-safe, but they don't all work the same way. Data clean rooms and some platform conversion modeling still match users with hashed emails or first-party IDs inside a protected environment, then return aggregate results. That can be useful, but it still depends on identity matching, and it inherits the same gaps when consent or match rates drop.

Measurement that runs directly on aggregated spend and revenue totals is different, since it never needed individual identifiers in the first place.

A tracked conversion isn't proof an ad drove the sale

This is the part signal loss made obvious, though it was true all along. Even a perfectly tracked journey only shows which ads a shopper touched, not which ads changed their mind. Someone who clicks a retargeting ad might have bought anyway, and someone who never clicked a CTV spot might have bought because of it. Measurement has to estimate what each campaign added, not just record which ones were present.

Measuring conversions at the business level

Once you stop trying to follow every shopper, the measurement question shifts to three simpler ones that leadership actually cares about:

• What happened, meaning how many conversions and how much revenue came in.
• What drove it, meaning how much of that came from marketing and which campaigns.
• What should change, meaning where budget should grow, shrink, or move.

Different approaches answer those questions with very different data requirements.

Marketing mix modeling stands out because it starts from what's observable (your actual revenue over time) and works backward to estimate what drove it. That's why it can measure channels that rarely get clicks, like CTV, audio, and direct mail, alongside paid social and search.

Why many aggregate models still get conversions wrong

Marketing mix modeling solves the identifier problem, but that doesn't mean every MMM measures conversions accurately. Traditional MMMs and many widely used open-source frameworks share a set of assumptions that break down in exactly the conditions marketers deal with every day.

The pie-chart assumption

Most traditional and open-source MMMs treat revenue like a pie. They slice each day's sales into a baseline piece plus a separate piece for each channel, then add the pieces back up. That assumes channels work independently and that baseline demand sits apart from marketing, when in reality a YouTube campaign can lift branded search, and last year's marketing is part of this year's baseline.

Spending rises and falls with demand

Marketers naturally spend more during holidays, promotions, and product launches, which are the same periods when shoppers are already more interested. When spend and natural demand climb together, there are many different ways to split the resulting sales between baseline and marketing that fit your history equally well. The model has no way of telling from the data alone which split is right.

Baseline leakage pushes credit in both directions

When a model can't separate those two forces, credit leaks between them. Sometimes holiday demand gets handed to your ads, which makes campaigns look stronger than they are and encourages overspending at peak. Other times, real marketing impact gets absorbed into trend or seasonality, which makes your brand look naturally strong and your campaigns look weak.

Why statistical fixes and more data don't solve it

Many open-source MMMs handle this with statistical penalties or starting assumptions that nudge the model toward one answer when several fit the data. Those techniques make the output stable, but they don't find the correct split. They pick one plausible answer and repeat it with confidence.

More data doesn't fix that either. If a model's structure can't represent how marketing actually works, adding months of history just makes it more confident in the same wrong split.

How Prescient measures conversions without identifiers

Prescient's marketing mix model takes a different approach to the same aggregated data. Instead of slicing revenue into independent pieces, our model treats demand as something your marketing builds, sustains, and lets fade over time. Baseline demand and marketing impact are estimated together, which makes it much harder for one to absorb the other.

That approach rests on a few principles about how marketing actually behaves:

• Funnel direction: Upper-funnel campaigns build interest that lower-funnel campaigns, like branded search and retargeting, often capture later. Our model credits the campaign that created the demand, not just the one that captured it.
• Halo effects: A campaign can drive conversions through other channels and sales destinations, like a YouTube campaign that lifts branded search or a Meta campaign that drives Amazon purchases. Our model measures those halo effects without cookies or user tracking.
• Flexible response shapes: Returns don't always shrink as spend grows. Our model doesn't force every campaign into an automatic saturation ceiling, which matters because our saturation research found many brands underspend on campaigns that still have room to scale.
• Varying delay and decay: Ads keep working after they run, and their effects fade at different speeds. Our model lets those patterns differ by campaign instead of applying one fixed decay formula to everything.

Because all of this runs on aggregated spend, impressions, and revenue, Prescient can measure campaigns at the campaign level with daily updates across your website, Amazon, and retail, without needing to know who any individual shopper is.

How to check whether a privacy-safe model is accurate

If an MMM doesn't track individuals, it's fair to ask how you'd know it's right. The answer depends on what kind of accuracy you're testing, and this is where many evaluations go wrong.

Revenue prediction and credit accuracy aren't the same thing

A model can predict your total revenue almost perfectly while still splitting credit between channels incorrectly. If Meta and Google spend always rise together, a model might get the total right without knowing how much each one contributed. Since budget decisions depend on that split, predictive accuracy alone isn't enough.

Useful accuracy checks look at both:

• Predictive accuracy: How well does the model forecast revenue on data it hasn't seen?
• Credit accuracy: When tested against simulated data where the true answer is known, how close does it get to each channel's real contribution?
• Uncertainty: Does the model show how confident it is, so you know when two campaigns' results are too close to call?

Prescient has tested its model against widely used open-source MMMs on simulated data with known answers, at the channel level, so the comparison would be fair across models that don't all report at the campaign level. Prescient also shows a model confidence score alongside its results.

Incrementality tests are a data point, not a final answer

Many brands want to check their MMM against incrementality tests, which makes sense. Prescient's view is that these tests are locally accurate and globally inaccurate: they can describe one campaign in one window, but they don't hold as a rule across seasons or spend levels. That's why Prescient shows brands their model's accuracy both with and without a test included. You make the final call, and a test you trust can still be built into the model as a starting assumption.

What leadership gets from privacy-safe conversion measurement

For CMOs and analytics leaders, the case for aggregate measurement comes down to a few business outcomes. These are the ones that tend to matter most in budget conversations.

• Less exposure to privacy changes: Because aggregate MMM works on spend and revenue totals, it doesn't depend on cookies, device IDs, or user consent for cross-site tracking. You'll still need to handle first-party data responsibly, but new browser rules don't break the model.
• One view across sales destinations: Website, Amazon, and retail revenue can be measured in the same framework, which matters for omnichannel brands whose ads drive sales in places pixels can't see.
• Budget plans you can defend: Our Media Forecasting feature forecasts what happens when you scale a campaign's spend or shift budget into it from other channels, using what the model has learned about your seasonality and halo effects.

Platform attribution still has a role, too. Ad platforms need fast conversion signals to optimize bidding and delivery, so their reporting is useful for decisions inside each platform. It's just the wrong tool for deciding how much budget each platform deserves.

MTA vs. traditional MMM vs. open-source MMM vs. Prescient

Every approach makes tradeoffs, and those tradeoffs shape the numbers you'll see. Here's how they compare on the issues that matter most for privacy-safe conversion measurement.

A plain-English guide to the technical terms

Some of the concepts behind privacy-safe measurement come wrapped in heavy jargon. This quick reference translates the most common ones into language that's easier to bring to a leadership meeting.

Common mistakes to avoid

Even with a strong model, a few habits can undercut privacy-safe measurement. These are the ones that come up most often.

• Treating platform-modeled conversions as if they were observed, verified sales.
• Assuming server-side tracking solves privacy limits, when it doesn't restore cross-platform identity.
• Judging an MMM only by how well it predicts total revenue.
• Comparing platform-reported ROAS with Modeled ROAS as if they measure the same thing.
• Expecting one tool to handle in-platform optimization, cross-channel budgeting, and forecasting equally well.

Where Prescient comes in

Prescient's marketing mix model measures conversions from aggregated data, with no user-level identifiers, and it was built to handle the problems that trip up many aggregate models. Our model estimates baseline demand and marketing impact together, credits the campaigns that create demand, and measures halo effects across your website, Amazon, and retail at the campaign level with daily updates.

That gives omnichannel brands a measurement approach that holds up as privacy rules keep shifting, plus a model confidence score and our Media Forecasting feature to turn results into budget plans. If you want to see what your campaigns are really driving without relying on user tracking, book a demo.