WHY EPPO

By Role

Data Scientists Engineers Product Managers Product Managers

Resources

Customers Outperform Updates White Papers White Papers

FEATURED CASE STUDY

Coinbase Saves Millions, Reduces Experiment Analysis Time by 40%, and Restores Trust in Experimentation with Eppo

Learn more

About

Marketing

April 8, 2025

Better Results, Faster: Eppo Geolift vs. CausalImpact and Matched Markets

Greg Dale

Before Eppo, Greg was the CEO of Tech for Campaigns where he led large consumer advertising campaigns and experimentation programs

With higher standards for marketing ROI, measuring with high rigor is critical – but often painful. When it comes to geotesting, the emerging “gold standard” way to attribute marketing initiatives to the business results they generate, teams have been stuck choosing between quick but wrong approaches or complex statistical methods that few understand and take forever to launch. Eppo Geolift changes this.

Three Big Ways to Geotest: What Are the Pros and Cons?

Geotesting aims to answer the question: "What would have happened without the campaign, creative, or tactic I'm testing?" We all know A/B tests, where groups of individual users are randomly and evenly split up, so that any differences observed can be attributable to the intervention being tested. Geotesting comes into play when one can’t randomize at a user level, like in many advertising and marketing situations. It enables you to measure the true causal contribution of your marketing to business results, supplementing methods like multi-touch/click/view attribution or vanity metrics like visits/views.

Before a new wave of geotesting tooling arrived, teams were limited to two imperfect approaches, namely manually-created matched market tests and open source packages like CausalImpact:

Matched Markets: Manually pairing treatment with control regions (Chicago with Milwaukee) using simple metrics like similar sales numbers.

Quick but inaccurate
Results rarely hold up when campaigns scale
Misses seasonal patterns and regional differences

CausalImpact: Data science teams with expertise in casual inference often use this open-source package from Google, but it falls short in key areas when used at scale:

Requires statistical experts to implement
Maintains dependencies on those experts to update, reanalyze, and socialize - analyses often remain invisible to other teams
Misses out on the opportunity to increase accuracy by accounting for advertising-specific factors like burn-in and saturation

(Plus, CausalImpact hasn’t been substantially updated since 2022!)

How Eppo Geolift Compares

Eppo Geolift is a new-wave geotesting tool that guides you through the process, brings next-generation statistics to the problem, and makes results visible and readable.

Planning: Suggests optimal test regions and their statistical power
Data integrity: Direct warehouse connections, no CSV exports or imports
Analysis: During the experiment, results update nightly, and after the experiment concludes, continue updating through a post-treatment “cooldown” period
Results: Visual explanations cross-functional partners can understand

Improving workflows is not very valuable unless the statistics and data behind them are trusted. For that, Eppo Geolift’s methods build upon approaches like CausalImpact by:

Building region-specific "what-if" scenarios (counterfactuals) using data from control markets
Improving accuracy by accounting for real advertising dynamics (delays, declining ROI at more saturated levels of spend)
Eliminating data-related discrepancies by using your internal data warehouse data to compute results

	Matched Markets	CausalImpact	Eppo Geolift
Accuracy	Low	Medium	High
Method	Correlation	Bayesian structural time-series model	Bayesian generalized linear model
Test Planning	Manual	Manual	Automated power analysis and region selection
Region-Specific Models	None	None	Yes
Adstock Modeling	None	None	Yes
Saturation Modeling	None	None	Yes
Setup Time	1–2 days	2–3 weeks	1–2 days
Required Expertise	Basic stats	Advanced stats knowledge and R	None
Reporting	Spreadsheet/Notebook	Notebook, data scientist-oriented	Data scientists, marketers, and executive stakeholders

For statistics fans:

Eppo Geolift uses Bayesian hierarchical modeling, explicit time series components, and estimates marketing-specific quantities like adstock and saturation. This contrasts with CausalImpact's structural time series and Bayesian state space models, which lack marketing-specific components and region-specific parameterization.

Eppo Geolift also properly handles post-treatment analysis periods with varied effect durations, allowing you to measure persistent effects after campaigns end – critical for calculating true ROI and understanding consideration cycles.

How Improved Systems Drive Better Decision Making

Eppo Geolift introduces enormous workflow improvements that help teams work together and a number of advanced statistics features that drive more reliable results.

From Frustrating Ambiguity to Confident Action

An omnichannel retailer faces measurement, coordination, and governance challenges across teams and channels – how can the performance team, brand team, and data science get on the same page? As examples:

The performance marketing team wants to test conversion-optimized YouTube ads. Data science advised running in the southwest US, and then ran CausalImpact, showing a potential 1-10% lift – too wide a range to confidently scale nationwide.
Meanwhile, their brand team is ready to cancel the next flight of a new billboard campaign after seeing minimal first-week impact on web visits.

Eppo Geolift solves both teams’ problems with its flexible statistical model. For the YouTube campaign, the specific advertising improvements delivered a tighter 5-10% range for the YouTube campaign. For the billboard campaign, incorporating the delayed effects of advertising revealed the campaign was working as expected. Many digital-native teams are used to the immediate feedback provided by last-click attribution, but that advertising effects, especially for out-of-home and linear television, can take up to 2-3 weeks to show up fully, and can make the difference between scaling or canceling a campaign.

Forget It! We'll Just Go With What We Have

A DTC brand's marketing team wanted to test different TikTok creatives. Past experience working on geotesting left them hesitant: "Last time we worked with data science on a geotesting plan, it took them three weeks to turn something around with us spending a lot of time explaining basic things. When we needed to change the plan due to changes in OOH activation plans, they said we'd have to wait another two weeks, and by that time, we ran out of time to do the test before the holidays, and we had to just keep rolling with our attributed results.”

With Eppo Geolift, they set up their test in one day. Eppo guided them through region selection with clear visuals and tradeoffs everyone could understand, and got the test launched. They discovered their UGC creative outperformed the big-agency “safe” version with a 37% decrease in signup cost – a finding confirmed when rolled out nationally.

The Sweet Spot: Diminishing Returns

An eCommerce company increases the budget for Google Shopping. There is a rise in platform-attributed purchases and visits, but numbers from transaction platforms look about the same as before. By deploying Eppo Geolift's saturation, the team pinpoints exactly where efficiency drops for Google Shopping:

Under \$50K/week: Strong returns
\$50K-\$75K/week: Moderate additional lift
Over \$75K/week: Minimal extra results

As they were already at \$60K/week, they shifted \$25K weekly to untapped channels, boosting overall purchase performance by 23%. Beyond marketing, Eppo Geolift’s flexibility shines in scenarios like testing regional price changes or in-store promotions.

For marketing teams making multi-million dollar decisions, it delivers the statistical rigor of academia with a pragmatic focus on relevance and ease, across the entire business.