Measuring true marketing impact: How we measure campaign lift with geo analysis
You’ve just launched a major marketing campaign across multiple markets. Three weeks later, sales are up 15%. Success, right?
Not so fast.
This isn’t academic curiosity—it’s the difference between scaling winners and throwing good money after bad campaigns.
How much of that growth would have happened anyway?
How do you separate genuine campaign impact from seasonal trends, competitor missteps, or natural market fluctuations?
Challenge
Traditional measurement approaches fall short:
Before-and-after comparisons: These can be misleading due to external factors that may influence sales.
Simple A/B tests don’t account for market spillover effects
Correlation analysis confuses coincidence with causation
Solution
To solve this, we turned to Geographical Lift modeling, a powerful causal inference technique. The core idea is to identify similar geographical areas (like Designated Market Areas, or DMAs) that did not receive the campaign treatment and use them as a “control” group. By comparing the performance of treated DMAs to these carefully selected control DMAs, we can isolate the true lift generated by the campaign.
Here’s how we approach it using a Bayesian framework with PyMC:
- Data preparation: We gather data on sales, marketing spend, and other relevant variables for both treated and control DMAs. This includes historical sales data, promotional calendars, and any other factors that might influence sales. We start with a rich dataset containing:
dma_id: Unique identifiers for each geographical market.
week_start_date: Temporal information to track performance over time at weekly level.
is_treated: A flag indicating if a DMA was exposed to a campaign during a specific week.
campaign_id: Crucially, identifying which specific campaign was active.
gmv_local_currency: Our key performance indicator – the Gross Merchandise Volume.
- Building the Bayesian model in PyMC with Synthetic Control Groups
To accurately estimate the incremental impact of each marketing campaign, we combine Bayesian inference with a synthetic control strategy. This hybrid approach improves robustness by constructing control outcomes that closely mirror the treated regions’ pre-campaign behavior.
Our model aims to decompose the observed GMV into distinct components:
Baseline GMV: This captures the natural sales volume in each DMA through fixed effects — including:
Campaign Lift: True incremental impact attributable to your marketing
Market dynamics: External factors affecting all regions.
By leveraging synthetic control groups, we construct credible counterfactuals for treated DMAs, strengthening our causal claims. And by embedding this in a Bayesian framework, we quantify uncertainty and deliver posterior distributions for each campaign’s effect — not just point estimates.
3. Causal impact quantification
We deliver precise estimates of incremental revenue with confidence intervals—not just point estimates, but the full range of probable outcomes.
Results
Immediate Value
ROI Clarity: Know exactly which campaigns drive real growth vs. borrowed future sales
Budget optimization: Reallocate spend from underperforming to high-impact initiatives
Stakeholder confidence: Present marketing results with statistical rigor that satisfies CFOs
Strategic advantages
Competitive intelligence: Understand how market dynamics affect your performance
Regional insights: Identify which markets respond best to specific campaign types
Predictive power: Build models that forecast campaign performance before launch
Ready to move beyond marketing guesswork?
If you’re tired of making major budget decisions based on incomplete data, let’s talk. We specialize in bringing scientific rigor to marketing measurement for companies serious about growth.
What we deliver:
Comprehensive campaign lift analysis
Statistical confidence intervals for all estimates
Strategic recommendations for budget reallocation
Custom dashboards for ongoing performance monitoring
Industries we serve:
- E-commerce and retail
- Financial services
- SaaS and technology
- Consumer packaged goods
Get Started
Free Consultation: 30-minute discussion of your measurement challenges
Pilot Analysis: Proof-of-concept study on one recent campaign
Full Partnership: Ongoing measurement and optimization support
Contact us to schedule your consultation and discover what your campaigns are really delivering.