Incrementality testing
Prove which channels actually cause growth
Geo-based experiments that measure real causal lift — on profit and new customers, not only sales. Not clicks, not modeled estimates, not platform claims.
Geo lift test — Meta Prospecting, DE
Daily profit, treatment vs control regions · 8 weeks
+14%
Incremental lift
2.8
Incremental ROAS
96%
Confidence
Trusted by commercial teams at leading brands
The experiments
Four test designs, one for each question
Every test compares matched treatment and control regions — the only difference between them is the spend change you're testing. Which change you make is the design: pause, introduce, scale up, or scale down.
Is this channel actually incremental?
Ads are paused or reduced in the treatment regions — a holdout — while control regions keep running as usual. If sales fall in treatment, the channel was genuinely driving them, and the gap is its true incremental contribution. If nothing moves, you just found budget.
The most common first test: run it on your biggest line item, because that's where a wrong number costs the most.
Lift test
Weekly sales index, treatment vs control regions
−11%
Sales in holdout regions
2.4×
Incremental ROAS
96%
Confidence
Setting up a test
Designed like a statistician would. Set up in minutes.
Everything that makes geo-experiments hard — matching regions, power analysis, contamination, lagged effects — is handled in the setup flow. You decide what to test; Dema makes sure the test can actually answer it.
New experiment
01
Frame the question
Pick the test design, channel, country, and storefront. Include or exclude specific funnels and campaigns, so the test measures exactly the spend you're deciding on — brand always-on stays out of a prospecting test.
Suggested design
Treatment
Bayern, Hessen +4
Control
NRW, Berlin +7
02
Dema designs the experiment
You get statistically optimized designs to choose from: matched treatment and control regions, plus the power analysis — minimum detectable lift, false-positive rate — so you know the test can find an answer before spending anything on it.
Experiment timeline
03
Run it, read it honestly
A treatment window of at least four weeks, then a post-treatment period so lagged conversions land before anything is finalized. Results come with confidence intervals and p-values — not vibes.
The metrics
Read on profit and new customers, not only sales
A channel can look great on gross sales and still lose money after returns and margin — or only reach customers you already had. Every experiment reads on six outcome metrics from the same regions and the same window, so you see the lift where it actually matters.
Gross sales
Read as ROAS
The loudest number — and the easiest one to inflate.
Net sales
Read as eROAS
After returns, discounts, and cancellations.
Net gross profit
Read as epROAS
What the spend actually earned after product costs.
New-customer profit
Read as epROAS
Growth from customers you didn't already have.
Returning-customer profit
Read as epROAS
Demand you might have gotten anyway — now you know.
New customers
Read as CAC
The true incremental cost of acquiring one.
The result
See the real incremental effect on profit and customers
Every test compares treatment vs control regions. The result is a causal measurement of lift, not correlation, not modeled estimates. This is what actually happened because of your marketing.
Profit ROAS — tested vs platform-reported
Geo-holdout tests · last 8 weeks
Meta Ads
2.8×
reported 4.6×
Google Search
2.5×
reported 3.1×
Google Shopping
1.9×
reported 2.4×
TikTok Ads
1.5×
reported 3.8×
Display / Programmatic
0.6×
Not incremental
The tick is 1.0× — break-even on profit. Platforms grade their own homework; the test measures what actually happened. Display claimed 5.2× and delivered 0.6× — that budget moved.
Profitable growth in real time
“Since we started using Dema, we can optimize our budgets for the most profitable growth in real time.”
Where the results go
One test keeps every other number honest
A finalized experiment doesn't end as a slide. It calibrates the MMM, corrects your attribution through causal factors, and grounds the budget moves the agent prepares — so the evidence compounds.
Meta — response curve
Calibrates the MMM
Mark a finalized test as a calibration anchor and the model's response curves get pinned to experimental truth — so predictions hold at spend levels you've actually verified.
Meta — incremental factor
Calibrates attribution
Every test feeds causal factor attribution: a benchmark built from experiments across all of Dema, narrowed by your own results, sets the multiplier that corrects what MTA and the ad platforms claim.
Prepared budget moves
Meta Prospecting
iROAS 2.4×
Display / Programmatic
Not incremental
Google Search
At saturation
Grounds the agent's moves
The agent combines tested lift with MMM curves and prepares budget moves for your approval — built on what's proven, not on what platforms report about themselves.
Go deeper
New to incrementality testing?
Our complete guide covers treatment and control groups, the four test designs, how to read statistical significance, why results should be read on profit rather than revenue, and what incrementality testing cannot tell you.
Trusted by category leaders
Real teams, real numbers — from cutting acquisition costs to doubling profit to automating whole planning cycles.
They tested scaling Google spend — and incremental profit went up with it.
Axel Arigato is a contemporary fashion and sneaker brand born at the intersection of streetwear, music, and culture. As growth accelerated, the team wanted to ensure that scaling their digital investments would translate into sustainable, profitable growth, not just higher ad spend. That's when they turned to Dema.
>100% Incremental epROAS on Google scalingEnterprise-ready
Built for production.
Where your data is processed, how it is protected, and what an agent is allowed to change are controls you set — not defaults you inherit.
EU & US inference
Choose where your data is processed. Model inference and data pipelines run on European or US infrastructure with full residency control.
Never used to train models
Your data is never used to train models. Processing stays in your chosen region and runs in isolated, encrypted sessions.
Always the latest models
New reasoning models roll out as they ship, so your agents keep improving automatically without migration work.
Permissions and approvals
Each agent has its own scope and integrations. Actions can require a named approval before anything changes.
What you can create
Real setups, linked to how they run.
Every example below is a real prompt from the library. Open the use case behind it to see the full run — what it checks, what comes back, and where it lands.
Questions
Frequently asked questions
Incrementality testing is a controlled experiment that measures the causal effect of marketing spend. Dema uses geo-based tests: spend changes in treatment regions while control regions stay unchanged, and the difference in outcomes is the incremental lift. Because it's an experiment rather than a model, it shows what actually happened because of your marketing.
The treatment window runs a minimum of four weeks (up to 24 for slower-moving categories), followed by a post-treatment period of at least two weeks so lagged conversions land before the result is finalized. Setup takes minutes: pick a test design, channel, country, and campaigns, and Dema suggests statistically optimized geo-splits with the power analysis done.
Yes. You scope the test to a channel, country, and storefront, then include or exclude specific funnels and campaigns. That's how you keep brand always-on out of a prospecting test, or test one funnel without touching the rest of the account.
Every experiment reads on six outcome metrics from the same regions and window: gross sales (ROAS), net sales (eROAS), net gross profit (epROAS), new-customer profit, returning-customer profit, and new customer count (CAC). A channel can look incremental on gross sales and still fail on profit or new customers — so you check all of them.
Attribution divides credit among touchpoints that were already recorded, so it can only describe conversions it observed and it inherits each platform's reporting bias. Incrementality testing withholds or adds spend and measures the difference in actual outcomes, which answers whether the spend caused anything at all. In Dema the two meet: test results calibrate attribution through causal factor attribution, so day-to-day numbers reflect experimental evidence.
They're stronger together. Incrementality tests answer specific causal questions ('Does Meta in Germany work?') but can't cover every channel every month. MMM gives continuous, full-mix measurement — and every finalized test can be fed into the model as a ground-truth calibration anchor, so it stays accurate between experiments.
Connect your ad platforms (Google, Meta, TikTok), your e-commerce platform (Shopify), POS or store systems, and your cost data. Dema builds the unified data model automatically. Most teams are set up within a day.
Bring one commercial job