Machine Learning for lead scoring

Different data points can provide insight into a customer’s fit and their interest in your product. A lead scoring model is key for qualifying potential customers as part of the pipeline generation process. Starting with Marketing responses and incorporating firmographics and intent signals, historical purchasing patterns with similar characteristics should drive the qualification of a given lead. Machine learning for lead scoring is the name of the game. But keep an eye out for these hidden pitfalls to maximize success.

CORRECT INCENTIVE

A machine learning model can provide an accurate prediction of the likelihood of a lead becoming a customer, which can translate into a valuable qualification score. What happens when a Marketing team is incentivized on conversion to a stage before closed won? They will have a higher interest in the success of leads reaching the early stage, despite how that could impact the Sales team later in the funnel. Resist building models with the wrong incentives. Variables will be present that are important in an early stage but ultimately have nothing to do with closing deals, like a phone number that’s not null. Optimizing the business objective is paramount.

CORRECT THRESHOLD

The model should be accurate in showing a relationship with historical win rates when back-tested. Depending on the structure of the Sales team and the quantity of leads available, you may be asked to determine a score threshold to increase or reduce volume. Because of that relationship with win-rates, a threshold could critically impact business outcomes. Perhaps the best threshold is no threshold, with ranked leads reaching the Sales team to work down the list and prioritize amongst themselves. Otherwise it’s a no-win situation when questionable leads add doubt to the model, or high quality leads are removed arbitrarily. Regardless, constant evaluation of lead conversion to closed won is needed for two things: 1) for leads qualified for the pipeline, 2) for leads non-qualified for the pipeline eventually found through outbound efforts.

As is often the case when using data to drive decision making, the devil is in the details. Be thoughtful about the correct incentives and thresholds up front, and the revenue generating power of the right lead scoring model will become apparent.