> Site index: https://usehardal.com/llms.txt
> Every content route also serves markdown at <url>.md

# What's left when you can't track anyone?

If cookies and device IDs disappeared, marketing mix modeling can still read weekly spend and sales. Here is how to start.

Source: https://usehardal.com/blog/whats-left-when-you-cant-track-anyone
Published: 2026-11-19T21:00:00
Updated: 2026-11-19T21:00:00
Author: Hardal
Category: Marketing Measurement

---

Imagine every cookie and device ID disappeared tomorrow. Would you still know what works?

A lot of marketing measurement is built on following a person from an ad to a purchase. When that trail gets thinner, teams often feel as if measurement itself is disappearing. It is not. You are losing one way of reading the picture, not the picture.

Picture a farmer who cannot track each drop of water, but can see which fields grew after rain and irrigation. The farmer still learns which fields need more water, and which already have enough. The unit of study is the field over time, not a single drop.

## How marketing mix modeling reads the field

Marketing mix modeling works the same way. It uses aggregated weekly spend and sales per channel, so it needs no user IDs or cookies. A common guideline is about two years of weekly data, so seasonality has room to show.

MMM complements attribution. It does not replace it, and it is not exact. Attribution is still useful for the short, user-level questions. MMM is useful for the longer ones: how much did this channel contribute over months, and what happens if we spend more or less?

The model also needs non-marketing factors. Price changes, promotions, seasonality and holidays all move sales on their own. If you leave them out, the model will often hand that lift to whatever campaign happened to be on at the time.

Two other patterns are worth knowing, in plain language. Ads can keep working for a while after you stop, which is why a quiet week is not always a true baseline. And each extra dollar often brings back less than the one before it, which is why doubling a budget rarely doubles the result.

The model should be checked against the real world. Incrementality tests and geo-lift tests are the usual way to do that. If the model says a channel is strong, and a holdout in a region says otherwise, give the test more weight and recalibrate the model.

## You do not have to build this from scratch

Free open-source models such as Google Meridian and Meta Robyn have made this reachable for mid-size teams. You still need clean inputs. That is the modest role of a first-party data layer: daily or weekly conversion and revenue figures that you can trust, not a promise that someone else will build the model for you.

Growth teams can start with three questions. Do we have two years of clean weekly spend and revenue per channel? Which channels do we credit only because they are last click? How will we test the model, with incrementality or geo-lift tests?

When you cannot follow a single person, you can still read the pattern in the field.

## Today's small task

Export last year's weekly spend per channel next to weekly revenue and look at it as one chart. You do not need a model to see the first story in that picture.

If you want cleaner first-party conversion and revenue data to feed an MMM, start at [hardal.ai](https://hardal.ai).

Hardal is the first-party data and agentic analytics platform. We help enterprises send accurate signals to their ad platforms and grow revenue, using server-side data from their websites, mobile apps and AI agents.
