When should I split a channel into more granular breakouts?

For major channels (like Facebook and Google) we almost always split the channels into tactics (e.g. prospecting and retargeting). For other channels, there’s no right or wrong answer, but the factors we consider are:

1. Will there be enough spend in the smaller components for the model to get a reasonable signal? There are no hard guidelines here, but if the split you’re considering is more imbalanced than 80/20, it probably does not make sense to split the channel any further.
2. Can you take action based on the model’s recommendations? If you’re not able to divert additional spend into the specific subcomponents, then it does not make sense to report the incrementality at that level of aggregation. The channels should reflect your decision making process.
3. The broken out channels cannot be highly correlated. If the spend in 2 channels always happens on the same days, and the proportion of the spend in each channel is similar day-to-day, the model has to distinguish between many possibilities: Channel 1 had great ROI and Channel 2 had horrible ROI, Channel 2 had great ROI and Channel 1 had horrible ROI, or anywhere in between. The problem is that each of those hypotheses are equally valid given the spend patterns, and the model has no way of knowing which is more likely. Unless there is some way of breaking the correlation (like going dark in one of the channels or raising spend in one while lowering the other) the model will not be able to distinguish between the effects of the two channels.

For more information on how to determine your channel breakdown visit the Channel Breakdown section under Onboarding.