Non-Revenue Models Flexibility

[rajeshpaleti999]

A nice capability to have is to allow revenue_per_kpi to vary by channels for non-revenue models. Often it is not the case that all channels generate equal 'quality' KPI. Some channels may bring in more high-yield/long-term customers whereas other channels bring in low-spending/high-churn customers (based on other sources like last touch attribution).

[cpulavarthi]

Hello @rajeshpaleti999,

Thank you for contacting us and detailing your use case!

In Meridian, revenue_per_kpi is a conversion factor that has the same dimensions as the target KPI itself. This is used in revenue impact and budget optimization calculations when the chosen target KPI is a non-revenue metric. Having a channel-wise varying revenue_per_kpi will not be useful when we have to convert a single target KPI that is the result of base demand and all the marketing activities, which doesn’t itself have channel-level granularity.

Your concerns about the “quality” of KPI varying with channel are valid. However, in this case I’d recommend you choose a KPI for your MMM study that doesn’t have these variations because the goal of an MMM study is to determine the causal relationships of media execution on the target KPI. The difference in contributions to the target KPI and the causal relationships identified via modeling are used to determine budget allocations and essentially to identify the differences between increasing marketing efforts in different channels. To determine the differences between channels, using a KPI that also varies with channels isn’t ideal.

You may check our documentation on Causal estimands and estimation and the pages that follow to better understand the focus of Meridian in particular and an MMM study in general. While these sections don’t directly guide you regarding the choice of a target KPI for your study, understanding the modeling process and assumptions may help you better decide what metric is better suited to be used as the KPI for your use case. The documentation on About MMM as a causal inference methodology may be especially insightful for your understanding.

Some actionable steps can include:

1. Create a new value-weighted KPI and use that as your target KPI.
2. You may model using a KPI that is further down the funnel which is more indicative of value, such as "New High-Value Customer Sign-ups" or "First-time Purchases Over $X".
3. Using segmented models and building separate MMMs for different customer segments may also be helpful. Such as building a model for acquiring high-LTV customers and another one for low-LTV customers

Do reach out if you have any further questions or suggestions for us!

Google Meridian Support Team