Sturdy Statistics Case Studies: Medicare Supplement Insurance Provider

Takeaway:  Running unstructured transcripts through our model transformed it into structured data that immediately revealed powerful insights. Schedule a demo to unlock your unstructured data today.

A Medicare supplement insurance provider noticed an unexpected pattern in its sales calls. Sales for this company typically entail a brief (<15 minutes) discussion about premiums and coverage, followed by a long (>45 minutes) enrollment by an insurance agent. The company expected to see a mix of short sales calls which didn’t “close,” alongside a fraction of long sales calls which resulted in new customer enrollments. However, the company noticed a troubling third tranche of calls: long conversations that did not close. The company was at a loss to understand why callers would spend 45 minutes or more going through the enrollment process before deciding to drop out.

These long calls incurred a hefty cost: not only did they consume thousands of sales agents’ hours per month, but since these calls came in through marketing campaigns, the insurance company was paying hundreds of thousands of dollars per month in targeted advertising to drive these time-wasting calls. The company assigned a team of data analysts to find a way to understand and reduce these fruitless calls. The analysts checked all of the metadata surrounding these inbound calls, from the landing page url, to the marketing campaign, to geographic location, to browser cookies and other personal information about the callers. However, they were not able to find any meaningful predictor of the undesired calls.

Sturdy Statistics ingested one year’s worth of calls, along with all the metadata associated with those calls. After automatically structuring each conversation into a discrete set of topics, even a very simple quantitative analysis can recover remarkable insights. We ran a basic regression analysis to identify which topics most predict the undesired calls. Our analysis immediately surfaced a handful of topics related to union and veteran’s benefits. These topics were quite rare — they appeared on only a minority of calls, and when they appeared, they were discussed for only a minute or so. Due to this rarity, these topics would have been difficult for a human to detect in a cursory review of the transcripts. However rare, these topics were significant: the call ended after any of these topics were discussed, usually abruptly, and always without a sale.

When we brought this result to our contact at the insurance company, he immediately recognized the problem: callers with certain benefits were ineligible for some forms of supplement insurance. He checked with the sales team and learned that they sometimes only find out about this ineligibility late in the process. The company revised their sales workflow to discover disqualifying details earlier in the conversation. Moreover, through continued use of the Sturdy Statistics toolchain, the analytics team has been able to monitor these topics over time, and thereby to study the effects of different changes to the sales script. This result highlighted an inefficiency in the company’s sales process and gave them a path to optimize and improve the process.