Accuracy vs. Reach

All Lookalike Models try to balance a trade-off between accuracy and reach, which are two of the most important indicators of how your model will perform.

• Accuracy: the precision of a Lookalike Model’s predictions.
• Reach: the relative size of a Lookalike Model’s addressable audience.

As a general principle, you cannot optimize for both accuracy and reach. Deciding which one to focus on will depend on your marketing use case.

Optimizing your Lookalike Model for accuracy is typically used for targeting later stages of your funnel. This enables you to be more precise, with the trade-off of reaching fewer users. By identifying users who are most likely to convert, you can optimize their high-touch experiences to drive engagement, improve conversion rates, and increase customer lifetime value.

In the example above, the model has a high accuracy score of 9 and a low reach score of 1. The shape of the model predictions graph has little overlap between the source and target audience, which indicates less similarity between the users of those audiences. However, for the select users that fall into the area of overlap, they have a higher likelihood of converting.

Optimizing your Lookalike Model for reach is most applicable for targeting users in earlier stages of your funnel. This will allow you to reach more users, with the trade-off of being less precise. You can think of this as "casting an intelligently wide net". By identifying users who are least likely to convert, you can focus your marketing resources on the users who are likely to convert, improving conversion rates and maximizing your budget spend.

In the example above, the model has a low accuracy score of 2 and a high reach score of 8. The shape of the model predictions graph has a good amount of overlap between the source and target audience, which indicates more similarity between the users of those audiences. Therefore, you will be able to reach more users in the source audience, but they have a lower likelihood of converting compared to a model with higher accuracy.

When balancing the trade-off between accuracy and reach, consider the sum of accuracy and reach to determine a model’s fitness to be used. See the table below for a quick estimation of your model's fitness to be used.

In the first two screenshots shared, each model had a sum score of 10 for accuracy and reach (9 and 1, 2 and 8 respectively). Therefore, both models would be considered "good" but they are optimized for different use cases. For a comparison, see the model below that has a moderate accuracy score of 5 and a moderate reach score of 5.