Run holdouts that hold up.
Stella designs the test, picks control markets that match your test markets, runs your data through every model, and shows you the one that actually predicts. Start free and see it on your own numbers.
- Location selection that matches control markets to your test markets.
- A multi-model approach that surfaces the best pre-period fit, not the best-looking iROAS.
- Results you can defend to a CFO, with the metrics to back them.
Included in every plan, free and up. No credit card.
Trusted by growth teams who left enterprise measurement vendors
Platform ROAS counts conversions that would have happened anyway.
Attribution credits the last click, not the cause. The only way to know the revenue your ads actually caused is to hold them out in matched regions and measure the difference. Done right, that number is defensible. Done wrong, the control markets never matched and the result is noise. The design is everything.
Pick control markets that actually match.
Most holdouts are won or lost here. Stella reads your pre-period history and selects control markets that have moved with your test markets, then builds a weighted synthetic control that tracks them closely, even for dense, multi-location footprints. If the markets did not move together before the test, no model can rescue the result after it.
We present the model that fits, not the one that flatters.
Stella runs your study through an ensemble of models and surfaces the one whose control best tracked your test market in the pre-period, before ads went off. We do not care if a model makes your iROAS look great. We care that the control actually matched, so the lift survives scrutiny.
Everything a defensible holdout needs.
Guided test design
Stella recommends test and control cells and walks you through setup. No data-science degree required.
Add confounders
Bring in control variables like another channel, promotions, or seasonality so the model isolates what you are testing.
Defensible output
Pre-period R² and MAPE on every result, plus a clear iROAS range and statistical significance, ready for the budget meeting.
Already run a holdout? Put it through Stella, free.
Drop a holdout you ran with another vendor or in-house into Stella's free tier in our template. See what our location selection would have recommended, and what the validated result looks like, before you upgrade anything.
Hear it from the brands we run measurement for.
Incrementality is included in every plan, free and up.
Have us run it, or run it yourself.
Most brands want measurement done for them. That's our consultancy, and it's where most clients start. Not ready for done-for-you? The same platform is self-serve, from $6,000 down to free. There's an option for everyone.
$10k isn't for everyone. There's a plan for every budget.
Prefer to run it yourself? Same platform, self-operated: the full loop at $6,000, proper holdouts from $3,000, or start completely free. Walk down to the plan that fits.
- ✓Chat with your data: AI assistant for instant insights
- ✓2 MMM runs / month
- ✓2 incrementality runs / month
- ✓Manual CSV upload (no connectors)
- ✓Automated state-level ingestion, up to 3 connectors (Shopify + Meta + Google)
- ✓Unlimited state-level geo holdout experiments
- ✓Unlimited regional (state-level) MMM
- ✓Unlimited manual data uploads
- ✓Chat with your data
- ✓Everything in Starter
- ✓Always-On Incrementality: daily iROAS, no holdout setup
- ✓Marginal growth curves per channel & campaign
- ✓Cross-platform halo effects (Amazon, Shopify, retail)
- ✓Automated DMA-level ingestion, up to 5 connectors + DMA-level holdouts
- ✓MCP: query your data from Claude Code & your AI tools
- ✓Everything in Professional
- ✓Multi-Touch Attribution (cross-channel journeys)
- ✓Reverse ETL: feed incremental conversions to Meta, Google & TikTok so they bid for incremental profit
- ✓Up to 12 connectors at DMA / zip-level volume
Incrementality questions, answered.
How does Stella pick test and control locations?
Location selection is where most holdouts are won or lost. Stella analyzes your pre-period history and selects control markets that have moved together with your test markets, so the comparison is valid before the test even starts. It works even for dense, multi-location footprints using a weighted synthetic control.
What is a weighted synthetic control?
Instead of matching one control region to your test region, a weighted synthetic control blends many control markets into a synthetic twin that best reproduces your test market history. The closer that twin tracks before the test, the more trustworthy the lift estimate after it.
How do you decide which result to trust?
We run your data through multiple models and present the one whose control best tracked your test market in the pre-period, before ads went off, not the one with the best-looking iROAS. You see the pre-period R² and MAPE (how closely the control matched before the test), plus the iROAS range and statistical significance, so it holds up when finance pushes back.
Can I add confounders like other channels?
Yes. You can add control variables (for example spend on another channel like Google, or promo and seasonality signals) so the model isolates the channel you are testing rather than crediting it for everything that moved.
How long does a holdout take?
A geo holdout typically runs 20 to 28 days, longer for high ad-stock channels. If you want a usable read sooner, Always-On Incrementality (in Professional and up) produces daily directional iROAS while the holdout incubates the rigorous number.
What can I do on the free tier?
Run real holdout studies, no credit card. Already ran a holdout with another vendor or in-house? Drop it into Stella in our template and see what our location selection would have recommended and what the validated result looks like, before you upgrade.