Mortar MMM’s platform
When using the platform, you'll upload a dataset and identify the columns that represent Time, Location, and Output. You'll also need to select a KPI Type from the available options: Revenue, Conversions, New Customers, or Orders.
The platform offers 5 data cleaning options:
Trim Mostly Zero Locations: Removes locations that recorded no conversions or sales in over 80% of the dataset. These locations won't contribute meaningfully to our synthetic control due to their very low conversion volume.
Clip Outliers: Improves model accuracy by adjusting extreme values when a location experiences an unusual spike (10× typical output). This is helpful because when such huge spikes exist in only a few locations, it becomes difficult to find donors for the synthetic control that can match these spikes, which would inflate the treatment error.
Trim to Last Year: When your dataset spans 2+ years with significantly different revenue scales between early and recent periods, this option retains only the most recent year for more relevant analysis. This helps because the algorithm would otherwise focus on finding good fits for historical trends, potentially compromising accuracy in more recent periods.
Check Problematic Seasonality: Identifies locations with trends that significantly differ from other locations in the region. These outliers can complicate modeling when included in treatment groups. For example, imagine a location that spikes when other locations fall, and falls when other locations spike—if this location ends up in the treatment group, building an effective synthetic control becomes very difficult.
Remove Locations: Allows complete removal of specific locations from analysis. These locations will not be considered in either the treatment or the control. This is useful when your dataset contains irrelevant regions—for example, when you have data from both the US and Latin America but only want to analyze US markets.