> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dema.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# How incrementality testing works in Dema
> Learn how Dema's incrementality testing isolates the true impact of your marketing campaigns on both sales and profit.
Incrementality testing helps you measure the **true causal impact** of your
marketing campaigns. Instead of relying on correlation-based metrics (like
attribution models), it uses controlled experiments to see what really happens
when you **increase or decrease** marketing spend.
**Key idea:** Incrementality testing answers the question: *Would we have seen
the same results without this marketing spend?*
## How Dema approaches incrementality testing
Dema's methodology is based on **geo-experiments**, where we compare treatment regions (where we modify marketing spend) to control regions (where spend continues as usual). This helps isolate the actual effect of marketing.
Treatment regions have ads paused or reduced (holdout treatment) to measure
how much performance drops without marketing.
Treatment regions receive new channel ads to see if it drives incremental
sales and profit beyond existing marketing efforts.
Treatment regions have ad spend increased or decreased to measure the
incremental impact of adjusting marketing efforts.
Unlike traditional methods that only focus on **sales**, Dema's platform
allows you to measure **profit impact** as well. Since we track **NGP2 and NGP3
(Net Gross Profit)**, we can calculate **effective ROAS (epROAS)**—a
profit-based ROAS that includes costs like returns and marketing spend.
## Three-phase approach to testing
Dema's testing framework consists of three key phases:
**Selecting treatment and control regions**
* We use historical data to find similar regions based on marketing activity, sales trends, and seasonality.
**Using synthetic control groups**
* Instead of simple A/B splits, we create a **weighted blend of control regions** that best mimic the treatment region's past performance.
* **Example:** If Los Angeles is a treatment region, our model might determine that 50% of Houston, 30% of Phoenix, and 20% of Denver together best match LA's historical trends. This makes the comparison more accurate than a random split.
**Grouping zip codes by commute zones**
* We recognize that people often shop and commute across city boundaries.
* Instead of using rigid geographic lines, we cluster zip codes into **commute zones** where consumers travel frequently.
* This ensures our treatment and control regions reflect actual shopping behavior.
**Region assignment**
* Dema provides the list of treatment and control regions
* Ensures alignment with **real-world consumer movement patterns**
**Campaign adjustments**
* Your team adjusts campaign settings in ad platforms
* Either increasing spend (new channel treatment) or pausing campaigns (holdout treatment)
**Test execution**
* The experiment runs for a set duration to collect enough data for analysis
Once the test is complete, we analyze results using **synthetic control modeling**:
**Comparing actual vs. predicted outcomes**
* Check how sales and **profitability** in treatment regions compare to control group trends
* Evaluate performance against baseline expectations
**Measuring incremental lift**
* Calculate the increase or decrease directly caused by marketing spend changes
* Isolate true causal impact from other factors
**Calculating incremental ROAS and epROAS**
* Unlike standard ROAS based on clicks/views
* **Incremental ROAS** isolates true marketing impact
* **epROAS** incorporates profit margins, costs, and return rates
By focusing on **incrementality**, Dema ensures you're making data-driven
decisions based on the **actual business impact** of your marketing, not just
attributed conversions that might have happened anyway.