Market Core · Marketing Channels · Module

Testing and incrementality

Whether a channel contributes demand that would not otherwise have arrived, established by withholding it rather than by observing it.

The idea

How it works

Observed performance and incremental contribution are different quantities. A channel reaching people who were already going to buy reports excellent numbers and adds nothing, and no amount of tracking distinguishes the two.

A holdout answers it. Withhold the channel in one region or one period and compare. It costs some demand and it is the only method that produces an answer rather than a model.

Working with it

In practice

  1. 01

    Test the largest lines first

    The channels with the most spend have the most to gain or lose from the answer.

  2. 02

    Withhold long enough to see the effect

    Short holdouts measure noise. Most channels need several weeks before the difference is readable.

  3. 03

    Decide in advance what result changes what

    Agreeing the response beforehand prevents an inconvenient result being explained away.

  4. 04

    Accept that some channels add nothing

    It is a common finding and an expensive one to ignore.

One level in

The components of testing and incrementality

A component is something that exists afterwards which did not exist before — a deliverable or a mechanism, not an intention.

  1. The test design

    What is withheld, from whom, for how long.

    Learn
  2. The result

    The measured difference, with the noise band it has to exceed.

    Learn
  3. The decision

    What the result changes, agreed before the test ran.

    Learn

Agree what each possible result would change before running the test. Afterwards, inconvenient results get explained away.