In August we wrapped up our initial pilot with the Dexcom G6 sensor system. We implemented a new product development process enabling us to leverage much more of the team and take on more projects at once. We experimented across key areas in our guided journey and social experiences as well as had key launches for scaling, like automated identity verification.
We also crossed 1 million meals logged in our beta program. We believe this is the largest dataset of non-diabetic glucose logs paired with food and activity data.
Sample of most common foods logged over the last 6 months prepared by Helena from our data science team.
In August, we wrapped up our initial pilot of the Dexcom offering. We shipped Dexcom units to several hundred new and existing members and collected detailed feedback on the experience along the way. Unlike the Freestyle Libre, Dexcom G6 sensors stream glucose data to smartphones without having to be manually scanned. Dexcom also has an API with which we can integrate at scale. On the downside, the API is currently degraded to a three hour delay, and the dual sensor + transmitter CGM package can be cumbersome.
The headline result from the pilot is that users do prefer the Dexcom to FreeStyle Libre, and given the hypothetical of real time rather than delayed data, do so by a very large margin.
With this result we will now move forward towards an Institutional Review Board (IRB) study and partnership with Dexcom, which will allow for real time data integration for Levels members.
One of our top objectives is to continue exploring to improve the guided journey with Levels. In August, we iterated through several positive scoring concepts with the intent of encouraging engagement and improvement. We launched the first version of Meal Insights and designed the second version: tagging. Design made progress on the next version of our program, and we also launched a pilot effort for in-app video to test whether video formats will be a more effective way to guide members through the program.
We ran several experiments testing different metrics that count up throughout the day instead of decrease, like our current Metabolic Score. We're still experimenting, but so far have learned that glucose should remain the star of the show, and that other metrics should have supporting roles. We're focused on a more positive metabolic score model that will be more explainable as a next iteration.
We launched the Meal Insights feature that we started developing in July, and we started working on the next iteration of it that will leverage our 🏷 Tagging system.