I did some work with Silverlink some time ago and was really impressed by their use of decision management to improve the communication of health plans and others in the healthcare space with members and patients. I got a chance to catch up with them recently to discuss progress and their use of adaptive control – also known as champion/challenger, test-and-learn or A/B testing.
Silverlink provides communications services (on behalf of health plans, population health companies, and pharmacies) to educate and support consumers on important health decisions. Their programs span health, economic and administrative behavior change. Silverlink has grown to 100 plus employees from about 50 in the last couple of years, with an increasing focus on analytics as part of their solution. They had 47% growth in 2008 and strong (though as yet unreported) growth in 2009. despite the recession. They have completed over 250M interactions with healthcare consumers and have worked with all of the top 20 health plans on a variety of programs in health behavior change from flu shots to drug adherence programs to health risk assessments and alternative coverage identification. Perhaps half the plans they work with are regulars, organizations that are really thinking about how analytically-driven outreach can help with their business overall. Many of these started smaller, with one-off projects, but are now rolling these out more broadly. They are also doing more work with Pharmacy Benefit Managers like CVS or Walgreens and these folks tend to be more analytically sophisticated – used to the idea of control groups etc.
Often the programs being managed combine outcome benefits for members and financial benefits for the plans. Moving to by-mail prescriptions, for instance, decreases costs for the PBM and improves health outcomes too.
Typically Silverlink develops an initial segmentation based on demographics and health data. Increasingly Silverlink is trying to use analytics and the experience it has built up to help companies make good initial selections. Once the initial intervention is launched they develop a champion and then challenge it regularly and rapidly, using their experimentation strategy and randomized control groups before integrating the results for ongoing improvement. As you might expect in such a complex and regulated world, data integration remains an issue, but the plans have become much more willing to use claims data as part of their analytics and this is key as the claims data represent the core behavior data for members of these plans.
It is critical to analyze how different segments respond to different approaches even when you make good selections, though, so they always measure and track. Even though Silverlink has some best practices the specific program, population, targeted behavior change all impact what will work best. Often too the challengers are better for a sub segment (e.g. a male voice works better for certain ethnicities and young males). Analyzing how and why a challenger strategy is better is a key element of deciding what changes to make when. Silverlink puts an explicit focus on finding barriers to response, engagement and ultimately ‘taking action’, and can address them dynamically during a call – real time decision-making based on the data being collected during the current conversation is really important in driving behavior. In addition, Silverlink is running programs that study the impact of multiple-action/sequenced activities.
To illustrate, various changes can be considered to an outreach program as part of a challenger strategy. These examples are for certain populations and specific applications, as no two populations are the same, but they are illustrative. The approach and the results will vary by the behavior you are trying to motivate.
- The number of attempts is one lever. For instance in one program going from 1 to 3 attempts increased response by 50%
- Using different voices for the recorded messages is another and changing the voice used also had a 50% improvement in one program.
- In one program, Silverlink is using last names to predict ethnicity/language to help plans do targeting in regions where many ethnicities live and many languages are spoken but where this data is not available for all members. Colorectal screening was an interesting test and an example where a male voice outperformed female voices. Interestingly for Hispanic groups this was particularly strong. Within the voice testing there was also an ethnic v standard messaging variation by group.
- Day of week and hour of day is another. Monday performed really well with Weds and Thurs next and Tuesday and Friday last. But Monday is a call center challenge because most call centers are so busy with in bound calls. Nevertheless the data suggested that organizations could not leave Monday out of their outbound plans.
They had one great example of a program to help engage members in disease management and lifestyle programs. They started with a 6.6% response rate, changed the message (up to 8%), the script twice (12.2% then 12.8%), the voice (13.1%) and finally the policy (16.3%). Using adaptive control they improved the response rate by 150% over 6 weeks. A great illustration of the power of adaptive control and learning through champion/challenger.