What are incremental sales?

Every campaign sits on top of sales you would have made anyway: repeat buyers, brand search, and people who find you in organic results. If an ad reaches those shoppers right before they buy, the platform may credit the sale. That credit isn't the same as cause.

In numbers:

incremental sales = sales with the ads − sales without the ads
incremental ROAS   = incremental sales ÷ ad spend

You never observe "sales without the ads" for the same people at the same time. Every method for measuring incrementality is a way of estimating that missing piece. Count the same lift in dollars as incremental revenue, or frame the budget question as marketing incrementality.

Suppose two groups of states usually sell in step. During a four-week test you pause Meta ads in the holdout states. The ads-on states sell 1,200 Amazon orders; the holdout states sell 1,000. You spent $4,000 on Meta in the ads-on states. Treating each order as $50 for a plain example:

incremental orders ≈ 1,200 − 1,000 = 200
incremental sales ≈ 200 × $50 = $10,000
iROAS              ≈ $10,000 ÷ $4,000 ≈ 2.5×

A real test also matches regions on history and reports a range around the lift. The incremental ROAS calculator and the holdout test calculator put those steps into numbers before you spend.

Attributed sales follow platform rules. Meta credits a sale if the buyer clicked or saw an ad inside a window. Amazon Attribution credits tagged clicks within 14 days. Those numbers answer "who got credit?" Incremental sales answer "what changed because of the ads?"

• Attributed can be too high. Loyal customers who would have reordered still count when they see a retargeting ad.
• Attributed can be too low. Someone sees Instagram creative, doesn't click, and searches your brand on Amazon the next day. Neither Meta nor Amazon Attribution sees that path clearly.

Researchers have compared large Facebook experiments with common non-experimental methods and found those methods often misstated the effect (Gordon and others, 2019).

You need a holdout group: people or places that don't see the ads but are otherwise alike.

• Hold out people. The ad platform keeps some users from seeing ads and compares purchases it can track. That misses Amazon orders the platform never sees.
• Hold out places. A geo test pauses ads in some regions and compares Amazon ship-to orders. That is the usual approach for Meta → Amazon brands.

Pick regions with matched market testing, compare with difference-in-differences, and read the full method on how lift is calculated.

Spillmark runs geo holdout tests for Amazon brands that spend on Meta. It plans matched regions, locks the design before the test starts, compares Amazon sales there with everywhere else, takes off Amazon fees and your product costs to show you the real profit, and keeps every test on record for your brand.

Start with the holdout test calculator to see if a test can detect a realistic lift, then read how lift is calculated.