The challenge
The company sells to healthcare providers through a field team of more than 50 reps. Its prospect list was short and its sales history thin, and a rep's time was too valuable to spend on weak leads.
The approach
- Built the provider universe. Every US provider by National Provider Identifier, individuals and organizations, joined with 18 government datasets and a proprietary dataset of each provider's purchases in dollars, updated every quarter.
- Trained the models on real purchases. Four machine learning models (XGBoost) learned from those purchases, for individual providers and for organizations: one for whether a provider looks like the providers who buy, and one for the volume it would buy.
- Scored every prospect twice. A propensity score and a volume score, each a rank from 1 to 100 against every provider. A volume score of 90 means 90 percent of providers would buy less.
- Put the scores into everyday work. Both scores in Nockpoint for analysis, and in Salesforce on every prospect, where a rep sees them while working the account.
- Kept the data current. Each quarter, Nockpoint brings in the newest purchases as a delta and the scores refresh, well ahead of the government's yearly data.
- Loaded deep, assigned narrow. The top fifth of scored prospects went into Salesforce. A territory assignment gave each rep a focused list, and any prospect a rep met was already there, with its scores.
The results
- About ten times as many qualified leads in front of the sales team as before.
- 8.5 million providers screened: 6.7 million individuals and 1.8 million organizations.
- The top fifth of scored prospects in Salesforce, with both scores on every record.
- The best leads shared fairly across more than 50 reps, by geography first.
- Scores refreshed every quarter from the newest purchase data, well ahead of the government's yearly release.