Skip to main content

Data Model

Overview​

Eppo Geolift leverages Eppo's metric primitives but has additional considerations when compared to A/B testing, detailed in this document.

Geolift focuses on detecting attributable differences in the Metrics analyzed from ground up, at aggregated levels.

Planning New Tests​

To plan a new test, consider:

  • The channel, creative, or campaign you'd like to test
  • The business question or hypothesis you'd like to answer
  • What units of randomization might fit your data (see below)
  • Any units/regions that must be included in treatment or excluded from the synthetic control
  • Eligible testing dates (blackout periods, promotional periods, other tests in-flight)

Analyzing Existing Tests​

To analyze an existing test, consider:

  • The baseline test design
  • The start and end dates of the intervention
  • The unit of randomization (see below)
  • The units/regions included in treatment and included in control

Geolift Units​

In A/B testing, the unit of analysis is often a user, account, order, etc. In Geolift, the units of analysis are usually geographies or a set of non-geographic units that are more aggregated than individal actions. For example, in a regional geotest, the treatment and control would vary across US States, DMAs, or postal code clusters.

The Geolift unit of analysis is created as an Entity in Eppo. The provided units are read as strings and must be consistent across the entire pipeline, including KPI modeling, spend data, and results.

Eppo supports multiple entities across Geolift, A/B testing, and contextual bandits. Read more on how to create entities that match your business here.

Geographic Units​

Popular geographic levels include:

  • US Regions (DMAs, MSAs)
  • Commuting Zones
  • ZIP/Postal Code Clusters
  • States
  • Countries

Non-Geographic Units​

Popular non-geographic levels include:

  • Retail stores
  • Fufillment centers
  • Web pages
tip

The unit should be consistently available across:

  • Targeting (e.g. ad platform location inclusion/exclusion controls)
  • Spend reporting (e.g. ad platform reporting breakdowns)
  • Metric reporting (e.g. transaction-level dimensions)

Metrics and Spend​

Primary and secondary metrics and ad spend data are made available to the Eppo Geolift system by creating Metrics. Generally, a Metric should be aligned with the source of truth in your warehouse. For Geolift, the most important factors are:

  • The Entity to be analyzed is mapped to the column that contains its values
  • The Entity value (e.g. location) is attached to the Metric

Geolift supports SUM and COUNT aggregations. The Metric layer is also where filtering can be performed on values that are supplied as Fact Properties.

Spend​

Ad spend data is supplied through the same process as your measurement metrics -- creating a SUM metric that maps to the appropriate level of aggregation and matches the intervention.

Historical Data Needs​

For your Primary Metric, we require at least three months of historical data to exist in the system. We support up to 18 months of data.