How to act on incrementality results

From test result to budget decision

Acting on incrementality means ranking channels by what they cause, not what they are credited with, and moving budget accordingly.

Incremental CPI: spend divided by incremental installs.

Incremental ROAS (iROAS): revenue from incremental conversions divided by spend. This glossary entry on return on ad spend covers the attributed version it is adjusting.

Both are always equal to or worse than their attributed versions, because incremental conversions are a subset of attributed ones.

Step 1: put every tested channel on the same footing

For each channel you have tested, take its attributed installs and revenue for a normal month and multiply by its measured incrementality.

Sample data: a made-up app to show the method, not a benchmark.

On the dashboard, Channel A is two and a half times cheaper. On what it causes, Channel B is cheaper and is the only one returning more than it costs. The ROAS lab lets you change each input and see when the order flips back.

Step 2: use the cautious end

Each incrementality figure comes from a test with an interval. Plan with the low end for channels you are scaling up, and the high end for channels you are cutting, so the decision survives the uncertainty. If the decision changes depending on which end you use, the test needs more volume before you move much budget. The results guide covers reading intervals.

Step 3: move budget in steps, not all at once

Diminishing returns: each extra dollar in a channel tends to buy fewer incremental installs than the one before, as the channel runs out of the most responsive people.

That means a test result describes incrementality at the spend level you tested. Doubling a channel's budget will not double its incremental installs. So:

1. Move a modest share of budget (for example a fifth) from the worst incremental performer to the best.
2. Run for a few weeks and watch total installs and revenue, not just attributed numbers.
3. Retest the channel you grew once its spend has settled at the new level.

This budget reallocation framework describes a weekly rhythm for moving spend on attributed data; incrementality results tell you how much to discount each channel inside that rhythm.

Step 4: decide what to do with low-incrementality channels

A low result rarely means "switch it off" straight away.

• Brand search: if competitors bid on your name, some spend may protect installs you would otherwise lose. Cap it and judge it on incremental installs.
• Retargeting: narrow the audience to users who have lapsed longer, where the nudge is more likely to matter, and retest.
• Prospecting with a low result: check creative and targeting before cutting. A test on a weak campaign measures that campaign, not the channel's potential.

Channels attribution cannot see well

Some spend, such as connected TV, outdoor or influencer bursts, shows weakly in click-based attribution, so attributed ROAS understates or misstates it. The way to measure the true return of a CTV campaign for app installs is the same as any other channel: a geo test. Run CTV in some regions and not in matched ones, read total installs, and compute incremental CPI and iROAS from the difference. The test types guide has the geo set-up.

Include the full cost

Incremental CPI is only as honest as its spend figure. Include agency fees, creative production and measurement tools when comparing channels with very different cost structures. MMP Price Watch has a rundown of MMP costs that do not show on the price page if measurement is a meaningful line in your budget.

When to retest

• After moving more than about a quarter of a channel's budget in either direction.
• When the season changes sharply (festival periods, year-end).
• When the product, pricing or store listing changes in a way that shifts organic demand.
• At least once a year for every channel that takes a large share of spend.