Lime Energy helps small businesses across the country become more energy efficient with LED lighting and other solutions through the implementation of utility direct-installation programs. Their diverse client base represents a multitude of industries and locations, from insurance brokers in California to coffee shops in Massachusetts. As part of the Lime Energy sales process, leads are given a free energy audit that outlines a unique efficiency plan and identifies how quickly their investment will be paid back.
While the sales audits generate a considerable amount of data, analysis on these opportunities was occurring on a one-off basis. Lime’s VP of Business Intelligence, Jesse Fallick, realized that there was a need for a more robust, machine learning-driven set of algorithms. However, the team at Lime lacked the internal resources to do a comprehensive dive into the available data, and didn’t have the budget to hire a full-time Data Scientist or the time to fully scope a project for a traditional firm.
Defining the Solution
When Sustainabilist began their work with Lime, a lot was still undefined. The data was there, but the initial scoping work that traditional data companies would require hadn’t yet been tackled. Sustainabilist’s Dr. Jason Trager took charge, embedding himself within Lime Energy’s team, connecting and working with a diverse team to find answers and insights that informed the data explorations necessary to grow the idea into a real model.
Jesse particularly appreciated the difference of Sustainabilist’s integrated approach. “Working with Jason gave us the flexibility of a contractor, but he felt like a member of our own team.”
Once the exploration was complete, Sustainabilist built a supervised machine learning classifier with over 300 inputs to predict which deals would close based on past performance of similar deals. To test the tool, Sustainabilist took a sample of 40,000 leads and split it in two. Half of the sample would be used to teach the classifier, the other to test it. The tool Sustainabilist built was able to predict whether a lead would close with 88% accuracy.
Making Efficiency More Efficient
A model now constructed, Lime Energy is applying their learnings. They’ve launched a pilot to put the model to work, laser-targeting sales efforts to leads the model predicts are more likely to close. The model has also opened up other possibilities, including data-driven salesperson assignments. In the process of this, the model has been integrated into the Lime Energy data structure and is now incorporated into their production database operations. Sustainabilist handled a significant portion of this integration, including defining the model’s database tables, automation, logging and testing.
The classifier can predict which salesperson is most likely to close that particular deal. This necessitated experimentation with a new commissioning model where team members are rewarded for bringing in a lead that closes, whether or not they were the sales person to close it. The collaborative versus competitive model rewards sales staff for passing on their leads to the team member who is most likely to close.
An Increasingly Perfect Model
While the initial modeling is now implemented, the model’s work has just begun. With each new lead, sale, or missed opportunity, the classifier gets more refined, getting closer and closer to perfection. The beauty of machine learning: the more data you feed it, the better it gets. Likewise, Sustainabilist continues to work with Lime on sales process improvements, from the pilot results and beyond.