In short
Geo holdout states are the regions where you pause Meta ads while leaving them on elsewhere, then compare Amazon ship-to orders. The share you pause trades power against cost: a larger holdout sees smaller lifts sooner but parks more spend and sales. This calculator takes a target share, your monthly Amazon sales and AOV, and returns how many of 51 equal-weight regions to pause, an example geographic starter list, and the minimum detectable lift for that unequal split. The state list is an illustration. Always rematch on your own order history so the groups sold in step before the test. Don't pause from the example table alone.
What are geo holdout states?
In a Meta → Amazon geo test you split the map. Holdout states pause the ads. Ads-on states keep them. You compare Amazon orders by ship-to state before and during the window. The states themselves aren't special; the match quality and the power of the design are.
People still want a starter list: which states might I pause for a 30% holdout? This tool answers that under equal weight across 51 regions (50 states + DC), then checks whether that share can see the lift you hope for given your sales volume.
Think of the share as a dial between learning speed and paused cost. A 15% holdout parks less Meta spend and less Amazon lift during the test, but the minimum detectable effect stays larger, so weak ads can still look like noise. A 40% holdout learns faster and costs more in paused media. Most Amazon brands planning Meta tests land somewhere in the middle once they see both numbers on the same screen.
How to choose geo holdout states
holdout count = round(target share × 51), kept between 2 and 49 actual share = holdout count ÷ 51 CV (planning) = max(1 ÷ √(orders per region-week), 25%) MDE = (1.96 + 0.84) × CV × √(1/(n_off×weeks) + 1/(n_on×weeks))
- Equal weight treats each region alike. Your real sales aren't equal; rematch on orders.
- Starter list walks a fixed geographic spread so the example is stable and shareable.
- MDE uses the unequal group sizes you actually selected, not a forced 50/50 split.
For brand-size screening first, use the brand size holdout calculator. For full CV control, use the holdout test calculator.
Worked example
A made-up brand wants about a 30% holdout for 6 weeks, with $60,000 monthly Amazon sales at $40.00 AOV, hoping for a 20% lift. These are the calculator's example values.
- Pause 15 of 51 regions (29.4% actual share).
- Example codes start AK, HI, MT, WY, ND…
- Minimum detectable lift ≈ 13.5%, at or below the hoped-for lift.
Before launch, replace the example list with states that tracked each other in your baseline. A pretty map that never sold in step will bias the result.
What's a good holdout share?
Common planning bands sit around 20–40%. Smaller shares pause less spend but need more weeks for the same MDE. Larger shares learn faster and cost more in paused media and forgone lift.
A good share is one where the MDE clears your decision threshold and the paused spend is money you can stand to park. Check paused spend in the results when you enter weekly Meta spend, and compare incremental economics afterward with the incremental ROAS calculator.
Common mistakes
- Pausing the example list blindly. It's a geographic illustration, not your matched design.
- Parking only tiny states. Power comes from balanced, comparable groups, not from hiding the holdout.
- Changing the map mid-test. Fix the holdout set before day one.
- Ignoring washout. After ads return, give sales a quiet period before you reuse those states.
How this connects to true incrementality
Choosing geo holdout states is step one of measuring cause. After the test, read waste with the ad waste calculator and pressure-test dashboard ROAS with the Meta ROAS reality check.
Spillmark plans geo holdout tests for Meta ads across US states from your Amazon order history and reports the range. See how lift is calculated.
Frequently asked questions
Which geo holdout states should I pause?
States that historically sold in step with your ads-on group, at a share your test can power. Use the example list only as a starting illustration, then rematch on your Amazon ship-to history.
Why 51 regions?
Fifty US states plus DC, the usual grain when Amazon order reports show ship-to state.
Can I pause by DMA or ZIP instead?
Yes if you can geo-target Meta that way and measure sales at the same grain. This calculator is built for state-level equal-weight planning.
How large should the holdout be?
Large enough that the minimum detectable lift sits below the lift you care about, and small enough that paused spend is tolerable. Often 20–40% in planning.
Does equal weight match Amazon sales?
No. California isn't Wyoming. Equal weight is for a transparent example. Production designs weight and match on your orders.