11 August 2016


How to Use Data Science to Predict Qualified Leads

We eat what we make

It’s true. We use our data science platform for the benefit of RapidMiner. For this particular project we revolutionized our internal marketing and sales processes using only our product suite and within a short time window.

It all started two and half months ago when we turned a critical eye to all our sales and marketing processes  The goal was to shorten sales cycles, better qualify leads, and learn how users turn into customers.

We wanted to use data science to predict qualified leads for our sales teams. But this meant moving away from the standard MQL driven sales model in favor of the new PQL one.

So how did we do it? How did we use data science to predict qualified leads?

We used RapidMiner Studio to analyze data from many disparate sources – sales, marketing, product and finance – to paint a comprehensive picture of users who buy our products or don’t. The majority of time looking into Salesforce data for historical wins and losses, Pardot data, and the typical usage patterns inside RapidMiner Studio.

We wanted to answer the important question, “are there any significant usage patterns that affect the propensity to buy.” The short answer is yes.

Predicting Qualified Leads

Just like with any data science project, we needed to form a cross functional team. We formed the “PQL team” early on which included key members from the sales, marketing, and data warehouse admins groups.

We held routine meetings to discuss progress, results, and provide any insight and guidance to keep the model iterating forward. In short, the success of the project hinged on active and fast communication.

The best part about this entire process? We implemented the entire analysis and solution from soup to nuts using nothing but our products. It started with RapidMiner Studio and ended with RapidMiner AI Hub (formerly RapidMiner Server), as we offloaded training and scoring processes. That gave us time to iterate over several models and select he best ones.

Aside from some SQL embedded in our Read Database operator and some “if-then” statements in our Generate Attributes operator, there wasn’t a shred of code that needed to written to put this process into production, it was all done inside Studio’s visual interface.

Sure I had to learn some SQL but fact that I could attack the business problem head on without having to rewrite data-frames or read 1000’s of lines of code to find some buried statement just felt more productive to me.

Time to Production

So, how long did it take to go from analysis to full production? Two months.

Yep, that’s two months of extracting and prepping data, generating and optimizing features, testing, modeling, evaluation and putting it into production. Just two months to revolutionize the entire sales and marketing process at RapidMiner. Now that’s cool!

Getting started with a data science project can be overwhelming. Request a free assessment and RapidMiner will help by creating and analyzing a portfolio of use cases in terms of technical feasibility and business impact.

Related Resources