# First incrementality test plan

From vexpower.com/mmm/incrementality-testing-tools (Mike Taylor, October 2026). Fill in every line before you spend a dollar on the test. If you can't fill one in, that's the first problem to solve.

## 1. The decision

- What will we do differently if the test shows a large lift? A small lift? No lift?
- Who signs off on that decision, and have they agreed to the design in advance?
- Channel (and campaigns) being tested:
- Primary outcome (orders, revenue, new customers, sign-ups), and where it's measured:
- Secondary outcomes we'll look at but not decide on:

## 2. The route

- Route chosen: platform lift study / DIY open-source geo test / vendor or app
- Tool: (Meta Conversion Lift, Google Conversion Lift, GeoLift, Meridian GeoX, CausalPy, GeoLift by Recast, Haus, Measured, other)
- Why this route over the other two:

## 3. Design

- Design type: user holdout / geo go-dark / geo heavy-up / multi-cell
- Unit of randomisation (users, DMAs, states, postcodes):
- Number of test units / control units:
- How test units were chosen (randomised, matched, tool's market selection):
- Pre-period length (aim for several times the test length):
- Test length (cover at least one purchase cycle):
- Cool-down window after the test for delayed effects:
- Spend change in test units (and total spend at stake):

## 4. Power, before launch

- Smallest lift worth detecting (the lift that would change the decision):
- Minimum detectable effect from the power analysis at this design:
- If the MDE is bigger than the lift worth detecting: extend, add units, or increase the spend change. Don't launch an underpowered test.

## 5. Things that could break it

- Other marketing changes planned in test regions during the test (promotions, launches, price changes):
- Spillover risk (people in control regions seeing test ads, or shopping across borders):
- Data delays or tracking changes expected:
- Who is watching for these and how they'll log them:

## 6. Reading the result

- Estimate and interval we'll report (lift, incremental conversions, incremental ROAS or CPA):
- Comparison with what the platform or attribution tool reported for the same period:
- What we'll do with the result in the MMM (prior, calibration, or a note for the next refresh):
- Date of the readout meeting: