Geo-experiments
We run geo-experiments: we split regions into test and control, turn on/boost advertising in some and leave others as is, and measure the true causal effect on sales from the difference. It is a way to assess returns without user tracking, at the geography level. Honestly upfront: this is an experiment — it is closer to causation, but gives a probabilistic result with a margin of error for the test conditions, not an exact eternal guarantee.
Geo-experiments — overview
Geo-experiments is an experimental project: we pick comparable regions, assign them to test and control, set the intervention (turn on/boost/turn off a channel), define metrics and the required scale/duration for statistical power, run it and analyze — estimating the incremental effect with a confidence interval and significance. The geo approach is good where there is no user-level data or click attribution is powerless (offline, brand, privacy). Honestly about the nature, this is key: even a correct geo-test gives a PROBABILISTIC result with an interval, not an exact number, and is valid for the participating regions and period — in another context the effect may differ; it is not an eternal guarantee. Honestly about requirements: enough regions and volume for significance and comparable markets are needed; on few geos or very different markets the conclusion is unreliable. Honestly about the method's cost: control regions are deliberately 'forgone' intervention there for measurement cleanliness. Honestly about discipline: the design and duration are set in advance, you cannot 'peek' and stop on a random spike. Honestly about data: correct regional sales/conversions are needed (garbage in, garbage out); external events in individual regions can add noise. Honestly about the essence: this is effect measurement, NOT running ads and NOT a growth guarantee — the test says whether the intervention works, while decisions and scaling are done by you. An important boundary: these are geo-experiments, not an audience holdout (at the user level — separate), not attribution/MMM (correlational — separate, complement) and not running ads. If there are few regions or markets are incomparable, the geo-test is premature. Picture this: instead of 'we boosted ads everywhere and argue whether it helped' you see the test/control difference by region as a causal estimate. The base price starts from 35,000 ₽ per experiment; it depends on the number of regions and duration.
Problems we solve
- You do not know the real ad effect without user tracking.
- Offline and brand are not assessed by click attribution.
- You boost ads everywhere and do not understand if it helped.
- Regional budget decisions are by gut feel.
What's included in the Geo-experiments service
- Selecting comparable regions (test/control)
- Designing the intervention and metrics
- Computing the needed scale and duration for power
- Running and controlling experiment cleanliness
- Analyzing incremental effect with an interval and significance
- A report with the result, limitations and validity conditions
- Reviewing results with you
- Recommendations for next tests
What you get
- A causal estimate of effect by region (with an interval)
- You understand whether the intervention actually works
- Region/budget decisions on effect, not gut feel
- A base for next tests (growth and running — separately)
How the work goes: steps
- We clarify the hypothesis, regions, metrics, scale; plan the design
- We run it, control cleanliness, do not 'peek'
- We analyze effect and significance, compile a report, review with you
Why PDV Expert
- Fixed price and timeline — no surprises on the invoice.
- Report and recommendations in plain language — clear without a technical background.
- In touch at every step and answering questions about the result.
FAQ
Will the geo-test give an exact and eternal answer?
No. Even a correct experiment gives a probabilistic result with an interval, valid for the participating regions and period. In another context the effect may differ — it is the best causal estimate available today, not an eternal guarantee.
Are control regions a loss?
Partly: in control you deliberately do not boost the intervention for measurement cleanliness. It is the price of a reliable answer; in return you stop spending where there is no effect. Enough regions and volume are needed for significance.
How is it different from a user holdout and from MMM?
A geo-test splits geography; an audience holdout splits users (separate); MMM is a correlational model on aggregates (separate). Experiments give a causal estimate, MMM/attribution scale the picture — better combined.
About the provider
The «Geo-experiments» service is provided by PDV Expert — a team specialising in «Conversion & analytics». We work under contract and deliver a written report with recommendations.