MIM Model Summary | How can I use ChannelMix Monitor to QA my ChannelMix AI model performance?

ChannelMix AI model performance has to do with the time frame and the variety of different data sources. Afterall, you can’t predict across platforms if you only have one platform in your data. Likewise, the machine can’t learn if you only have data for a short period of time. 

ChannelMix Monitor allows you to see deeper into your data to help find areas within your data that may be causing a low model performance. 

Dates for Datasets

For a well-performing model, you must have data for at minimum of 90 days for multiple data sources. In the Dashboard Details section of the QA monitor, you can see all of your data sources with the earliest (min) report date and the latest (max) report date. If you have a min date that is less than 90 days ago, we recommend you complete a backfill if possible (see ChannelMix Platform | How to backfill data).

Gaps in Data

The next thing to look for is gaps in the data. If the data is spotty and not consistent, it will result in poor conditions in which the machine is learning.

Looking at al the timelines at once can be overwhelming, so to filter down to just a specific source, click one of the  metrics above. Below is an example of Instagram filtered by cost for the last year. As you can see, there are a few gaps in the data in August of 2022 and November of 2022. If these are not expected, I can now either submit a ticket to have our team look into any issues or complete backfills for the dates for which I am missing data.