Infer’s predictive models score and stack-rank prospective buyers on a 0-100 point scale based on their fit for a specific company’s product. For example, here is a typical distribution across an entire business:
Sometimes a model will score a certain subset of leads (e.g. leads from a specific geography, industry, product, etc.) lower than average because a company hasn’t historically focused on those types of leads. For example, prospects in EMEA might just score lower because you may have started selling to EMEA customers only recently. This can cause problems if left as-is because your sales reps (or marketers) who focus on EMEA would only see Infer C and D-Leads, giving them limited insight with which to prioritize those leads.
There are two ways to address this challenge. First, if you have enough data volume to support it, we can build a second model dedicated to that particular subset of leads—for example, we can build a dedicated EMEA model. However, oftentimes there isn’t enough data because the subset of leads may represent a newer or smaller part of your business, so in that case, we can use another method to address this, which we call Infer Lenses. If your primary Infer model can effectively stack rank prospects within the particular subset you want to focus on, an Infer Lens is often a good solution. It is less overhead to manage, and your model will benefit from having a larger set of data to train over.
To implement a Lens, we effectively tell the model to rate leads using an appropriately adjusted, or Normalized, scale. Where we might have told the model to rate only leads scoring 90 to 100 as an A-Lead, we might now tell the model to rate EMEA-only leads scoring 56 to 60 as A-Leads instead. We’re adjusting the rating buckets for EMEA leads only in a way that better represents the potential customer fit of those leads.
With this approach, you could even create two or three lenses over the same core Infer model to adjust for different geographies, industries or product lines. And inside the Infer application, you’d have the ability to visualize the performance of the Model Normalizations using the same tools as you would for visualizing the performance of the core model.
So when does a second model make sense?
In some cases, a new subset of customers you’re selling to is so different from your core customer base that normalizations aren’t the answer. If there is enough data, we can build a second model and compare the results to the Lens approach. Ultimately, when you let the data talk, the right answer will become obvious.
Infer Lenses are just one of the many tricks we’ve learned while building predictive models for hundreds of companies. If you’ve got questions, contact us and we’d be happy to help out.