One of the advantages of Mortar MMM over other attribution methodologies is it takes into consideration marketing spend across all your channels. Since customer journeys may be influenced by multiple marketing efforts, having a holistic overview of all spend is essential in order to measure true incrementality of marketing spend. In order to do this Mortar MMM needs a comprehensive and complete view of your marketing program.
Comprehensive
Mortar MMM needs data for all sources of marketing spend. If you are spending money in a channel and Mortar MMM builds a model without accounting for that channel the amount of revenue being driven by that channel is either:
Going to be credited to other channels or the organic component of the model
Going to be credited to the baseline
Not going to be credited at all in the forecasting
This results in bias where we either overstate the effectiveness of some other component of the model or underestimate the overall number of sales/conversions generated.
Short term experiments are okay to not include, but once a channel becomes a significant portion of your overall spend (Typically >2%) and is going to be a long-term channel, we need to account for it in the model.
Complete
Another problem that can introduce bias is if the data Mortar MMM receives is not complete. By complete, we mean that if a day appears in your dataset all marketing channels and revenue/conversion numbers must be accurate for that day. A variety of causes can cause data on a given day to be incomplete:
The reporting for a given channel is behind, so we don’t have any data for the last week in a channel even though all the other channels have data.
Spend (or revenue) numbers “trickle in” over time so we might have some spend recorded for days in the last week, but those numbers will be revised upward as time goes on.
Any incomplete data will cause bias in the model, although the severity depends on the magnitude of the difference between the reported numbers and the actual numbers.
To solve this, in general we recommend providing only complete data even if that means your model isn’t completely real-time (ie. it’s July 7th but the Mortar MMM dashboard only uses data through June 30th). This is because we believe a slight delay is going to be less severe than bias introduced by using incorrect data.
If there is going to be a consistent, substantial delay in the ability to get complete data on a particular channel (e.g. your TV provider can only provide spend numbers twice a month), we can work with you on a strategy to use projected spend as opposed to actual spend in order to alleviate the bias while still being able to provide close to real-time insights. Please reach out to your Mortar MMM CSM contact if you think that is necessary.