Infer’s Model Build Process
A lot of people have asked us to pull back the curtain on how we build our predictive models. This two minute video provides a simple, step-by-step explanation. If you have question or want to learn more contact us, and we’d be happy to help.
First 3 Pages
How Predictive Models Are Built
Once you’ve recognized the value of heading down the predictive path, it’s important to have an understanding of the process. Building accurate predictive models requires seven key steps.
Start with your existing data
To create a predictive model, you first want to pull in your existing data and look at historical outcomes. This might include converted leads vs. archived leads, wins and losses, purchase history, and additional data for each record like company name, email address, and opportunity amount. That’s why at Infer, we’ve built easy connectors for Salesforce, Eloqua, Marketo, and Pardot. The more data you have the better, and it’s worth noting that systems like ours are designed to work in environments where data is sparse or even dirty.
Add thousands of external signals
The next step is to expand the data by adding external signals about your leads and customers. For example, with nothing more than a company name or email address, Infer can uncover all kinds of information about the individuals and the organization they work for. Thousands of factors like their relevant job postings,
employee count, patent filings, social presence, website traffic, and even the technology vendors they use might indicate their likelihood to buy.
Determine which signals are predictive
Once you have all the available data in place, the next step is to use machine learning to determine which signals are predictive. In some cases they’ll be positively correlated with conversion, and in some cases they’re negative signals. You can even set up your model to
weight signals towards large deals, so you can prioritize your efforts where they’ll have the biggest revenue impact.
Create the optimal formal
Advanced machine learning is used to test millions of scenarios and produce the optimal model for your business. It will determine the precise cutoff points (e.g. optimal sized customer has between 124 - 207 employees) and the proper weighting (e.g. employee count accounts for 1.7% of the overall score).