Skip to content
Get in touch

Incrementality and media mix models

| Article: 3 min read | Analytics

Incrementality is the share of sales that advertising delivered on top of what the store would have received without it. If a person would have bought from you anyway, the ad added nothing with them, even though the report records them as a conversion. A media mix model (MMM) is a statistical model that estimates each channel's contribution from data on spend by channel and sales.

In plain words

Attribution answers the question "who gets credit for the purchase." Incrementality answers a different one: "what would have happened without this ad." The difference is large for ads that reach people who are already ready to buy: for example, an ad shown for a search containing your store's name.

Haus gives an illustration: a platform reports 100 conversions, the control group without ads gives 40 anyway, so the real lift is 60, and the return falls from 1.5 to 0.9 per unit of spend. To get from the report to reality they suggest an "incrementality factor": the lift divided by the conversions in the report. If the reported ROAS is 4 and the factor is 0.7, the adjusted ROAS is about 2.8, which may be below the break-even point (Haus). The examples are illustrative, and Haus sells exactly this kind of measurement.

How it is tested

  1. A holdout test on the platform. For Meta this is Conversion Lift: randomly chosen users do not see the ads. According to Haus, Meta's approval is required, the test lasts at least seven days and needs at least 10% of the audience in each group. Check the exact spend and conversion thresholds in your account.
  2. A geo test. Some regions see the ads and others do not, and then total sales are compared. Haus describes modern variants with a synthetic control that weights many regions without ads. Google has an open tool, Meridian GeoX, with three designs: holdback (some regions are switched off), go-dark (ads are switched off) and heavy-up (ads are increased) (Google, Meridian GeoX).
  3. Switch it off for a while (practical advice, not from the sources). The simplest approach, but a rough estimate: turn off brand ads for a week or two and see whether branded sales fall. Changes are also explained by seasonality, so the result is approximate.

For Ukraine a geo test is possible by oblast, but whether regional sales volume is enough for conclusions depends on the store. There are no recent separate Ukrainian studies on incrementality.

Media mix models

An MMM calculates how sales depend on spend by channel, season, prices, promotions and other factors. Google offers the open Meridian model; GeoX results can be turned into priors to calibrate an MMM (Google, Meridian GeoX). Data volume requirements for an MMM are not given in open sources, so there are no figures.

Limitations: it is an estimate with an error margin, not a fact; setting it up takes an analyst who understands models. For a small store it is more practical to start with a simple test.

Illustrative example

A store spends UAH 100,000 a month on ads, and the report shows revenue of UAH 400,000 (ROAS 4). A test with a control shows that about 70% of these purchases would have happened without the ads. The lift is UAH 120,000, so the adjusted ROAS is 1.2. With a 30% margin the advertising loses money. That is why budget decisions are made on the adjusted number.

Common mistakes

  • Making the test too short.
  • Treating a test result as permanent: it holds for a period and a channel.
  • Testing everything at once.
  • Demanding precision from an MMM where there is little data.

FAQ

What is incrementality in plain words? How many sales the ads really added on top of those that would have happened anyway.

How is incrementality different from attribution? Attribution distributes credit among touchpoints, while incrementality estimates the difference between "with ads" and "without."

What is an MMM? A statistical model that estimates channel contributions from spend and sales data without tracking individual people.

Does a small store need it? Not necessarily. Start with correct order tracking and a simple test (for example, switching off brand ads).

Sources

Want to check this on your store?

Audit

We review the customer journey, catalog, technical state and data. A prioritized plan of changes in three days.

About the service →

Share

Related

  • Session recordings and heatmaps

    | Article: 3 min read

    A session recording is a replay of how one visitor used a site: where they moved the cursor, where they clicked, how they scrolled. A heatmap is a summary of such actions by many visitors on one page: where they click and how far they scroll. They show what…

  • LTV (customer lifetime value)

    | Article: 5 min read

    LTV (lifetime value, also CLV or CLTV) is the amount a customer brings to the store over the whole time of the relationship. For management decisions it is better to calculate it not in revenue but in margin: how much money the store keeps from all of the…

  • CAC (customer acquisition cost)

    | Article: 5 min read

    CAC (customer acquisition cost) is the amount a store spends to get one new customer: all marketing and sales costs for a period divided by the number of customers who bought for the first time in that period.

  • Data warehouse

    | Article: 3 min read

    A data warehouse is a separate database into which data from all of a store's systems (site, analytics, advertising, CRM, warehouse, accounting) is regularly collected in a single format for reports and analysis. You need one when numbers from different…