With over 80 million customers passing through their pharmacies every day, CVS Health is always striving to provide more meaningful interactions that put customers on a path to better health. In 2018, they embarked on a journey to personalize experiences through machine learning on Hadoop environment, but the complexity and scale of the diverse data set prohibited understanding the behaviors of a large number of micro-segments of customers. With Databricks, CVS Health was able to analyze their customer data to implement different experiences, experiments, and a variety of segments for personalization at scale.
In 2018, CVS Health was ready to focus on personalization, but with 10,000 stores across the United States and a large number of microsegments full of unpredictable behaviors, personalization was easier said than done.
“We’re not dealing with a typical grocery store customer,” explained Raghu Nakka, Sr. Analytics Advisor at CVS Health. “It’s really hard to predict behavior because a customer could go to a convenience store just because he forgot the milk, or he could go to a convenience store like CVS to pick up his prescription and stop to grab some candy. So the predictor of the behavior could be unpredictable, which is why our data dimensionality leads to overfitting issues for any kind of machine learning model.”
CVS Health kicked off its personalization journey in Hadoop. They built out an environment and launched their first personalized campaign to 1% of customers within a few months, but ran into roadblocks when they tried to scale from 1% to 5% because of a lack of processing power and physical data storage. “There’s an actual constraint around building additional hardware to support the scale that we wanted to get to,” said Michelle Un, Director of Enterprise Analytics at CVS Health.
In need of a more robust platform to achieve the level of personalization they wanted, the CVS Health team once again began setting goals and exploring their options. “We wanted to understand which channel the customer would respond to better, whether that’s a text message or a phone call, or an in-store offer, so optimizing our process was a really important goal,” added Raghu. “And the timing was just as important — like let’s say someone is up for a prescription renewal. That could be the right time to send an offer to the customer, rather than the customer who has already filled a prescription.”
CVS Health ultimately decided to transition to a cloud-based environment using Azure Databricks, which allowed them to expand the number of use cases that they were able to support and eventually scale to offer personalized offers to pharmacy customers.
“Through Azure Databricks we have the flexibility to spin up clusters that meet our unique business and various business use cases,” said Michelle. “We’re also not restricted by any more physical hardware constraints.”
Databricks also lent the teams some much-needed agility, which enabled them to focus on building and testing experimentation frameworks, rapidly iterate on different experiences, and expand their ability to utilize machine learning. And with centralized assets and interactive notebooks, any issues related to disparate data sources, such as limited data access and resource-intensive pipelines, were a thing of the past.
In their new, collaborative environment, all teams were finally able to work together: data engineering on building faster data pipelines, data scientists on training ML models more efficiently, and data analysts on visualizing financial and operational metrics for smarter internal decision making through Tableau.
With Databricks, CVS Health has a much better idea of who their customers are and what their needs are in the moment. They can measure the probability of buying a particular product, or when to remind a patient to fill or pick up their medication, and even identify potential side effects the patient may be experiencing.
“That’s really the core of our strategy,” said Raghu. “We predict the behavior of the customer and then make our offers more relevant and increasingly personal.”
The results of boosting personalization at CVS Health include a 1.6% improvement in medication adherence, meaning an increasing number of patients are now taking their medication on time and as directed.
With Databricks as their data analytics foundation, CVS Health is able to further patient-centric healthcare through actionable data insights and innovative models focused on improving long-term customer health and well being.
Databricks helps us to predict customer behavior and then make our offers more relevant and increasingly personal.”
– Raghu Nakka, Sr. Analytics Advisor at CVS Health