Home care is often singled out for being slow to embrace and implement technology, but as the demand for care services grows, providers are forced to think outside of the box when it comes curbing caregiver turnover.
San Francisco-based home care startup Honor understands this all too well, according to CEO Seth Sternberg. The company is using insights gleaned from machine learning to examine and address turnover internally and with its network of home care partners.
Honor, which has raised $115 million since launching in 2014, teams up with independently owned and operated agencies by taking over caregiver recruiting, onboarding and training, in addition to day-to-day logistics.
Currently, the company operates in Arizona, California, New Mexico and Texas.
Home Health Care News recently caught up with Honor’s Sternberg to discuss the company’s evolving technology platform, some of its lessons learned and why it’s important for home care providers to “professionalize” caregiving.
Below are highlights from HHCN’s conversation with Sternberg, edited for length and clarity.
Q: Turnover in the home care world is extremely high — 82% in 2018. How is Honor using predictive analytics and machine learning to glean insights into Honor Care Network turnover?
One of our own machine learning models is a predictive model on any given care pro’s likelihood to churn in 30 days.
What that model does is it determines the reasons why people are most likely to churn. Then you can score everybody on there based on where they are on all those different variables.
For example, there is a high correlation with turnover and the number of clients a care pro serves. It turns out that two is the optimal number of clients. If a care pro has one client, they’re more likely to turnover, and they’re also likely to turnover if they have three or more.
That would be another thing that would go into someone’s score, an aggregate score on someone’s likelihood to turnover. We have a whole set of those kinds of variables that the machine learning algorithm determines. This is important because I think people don’t quite understand the way machine learning works.
Machine learning tells you which variables matter, and then it tells you how much they matter in a given circumstance.
Q: In the past, you’ve explained to HHCN how Honor was using massive amounts of data to learn about chronically late care pros and clients’ attitudes toward them. How else has Honor leveraged data?
What machine learning is really doing is taking large amounts of data from the past and finding patterns in order to try to predict the future. There are a lot of ways we use that to end up with better care pro and customer experiences.
Honor watches to make sure care pros are going to be on time. We can see the GPS chips in phones as they move toward the appointment, and that can help determine if it looks like a care pro is going to be late or on time.
But about 20% of the time, you can’t see a GPS chip. There is also a machine learning algorithm that just tells our people when to call a care pro because it cannot see a chip. A lot of the time, the machine learning algorithm can still tell you if a care pro is going to arrive on time even if they can’t see the chip.
All of this is to say, we’re using the past to predict the future. That’s what we’re trying to do.
Q: Industry-wide, there’s been this view that weekends are harder to staff because caregivers want to spend that time with family, going to church or attending other social functions. At the HI2 conference in Chicago this summer, you mentioned how Honor has kind of learned there’s another layer to that story, correct?
Another way you can use machine learning is you can isolate. You can basically create an algorithm that says, “I want to look at every single variable around how easy or hard it is to staff a specific shift. I want to understand the impact of that variable on that shift.”
For example, a variable you could isolate would be all shifts in the town of San Francisco or all shifts that are three hours in length.
The reason why a weekend is hard to staff is that customers who are more difficult to serve are disproportionately likely to use you on the weekend. That’s just another example of where machine learning can get you an insight that then lets you solve around it.
Since we now know what the real challenge is, we can better target the care pros who are more skilled or better trained to work with more challenging customers.
We also now vary pay rates for if there is a challenging customer.
Q: Being a care pro or caregiver often requires lots of travel, driving back and forth between clients. What has Honor learned about caregiver commuting and their feelings around that?
There are two different kinds of things you can use. You can either use implicit data or explicit data. Implicit data is where you literally can just run an analysis on the care pro’s behavior without talking to them. The other way is explicit data, where you call a care pro and you talk to them.
We do a lot of both.
Our team is constantly working to improve the care pro experience. They talk to care pros and get the stories, and the stories are often the “why” behind the data.
The distance a care pro has to drive from where they live to a customer is highly correlated with turnover. And the distance is completely variable by city. For instance, in the [San Francisco] Bay area, for every 18 miles a care pro travels, that care pro doubles his or her likelihood of churning in 30 days.
In Los Angeles County, it’s every 25 miles. The whole point of machine learning is that it can be very nuanced about the actual underlying data. You have to recognize it’s distance correlated to the geography because care pro behavior and expectations are different in each geography.
The takeaway: If you have a care pro who’s driving a distance that would make them a risk factor for turnover, you want to find them work that’s closer to where they live.
Q: What kind of infrastructure does Honor need to have in place to gather all this data? Can an independent home care agency take steps to do something similar all on its own?
We are able to take all of this data across all of our partners and aggregate it into a big model that helps predict how to make care pros happier and how you make customers happier.
The problem for a truly independent agency — that is not part of Honor — is they don’t have enough data to be able to come up with these insights. I’d say an average agency probably has 20 homes and 30 caregivers. Even with five years of data, you won’t get enough data points at that scale to have such insights.
We’re very transparent. We publish stuff on our blog, And as we share insights, independent agencies can use them.
Q: In the past, you’ve also mentioned how it’s important that providers “professionalize” the caregiving job. What does that mean exactly — professionalizing the caregiver position?
At a very base level, it’s training a care pro well.
Then it’s telling care pros what is expected of them. Then it’s giving them the tools to be able to do a good job. The final piece of it is, for those who perform extremely well, what’s the path upward? Does that mean, over time, you earn more training to go into higher acuity homes where there are higher pay rates? Does it mean a promotion into HQ? Does that mean you become a trainer? What does it mean for my career?
That’s what I mean by professionalizing.
Q: You founded Meebo, which was acquired by Google in 2012. How did your experience there shape Honor?
The number one thing you learn when you build a company is that the only thing that matters is who works at the company. How good are your people? How much do they care? How much do they have empathy? How hard do they push to solve really difficult problems?
Q: What’s next for Honor?
The most important thing is to create a high quality, super consistent nationwide network of home care.
We are always looking for those really great partners. We have these amazing partners who have spent decades-plus on home care, and they make us better.
The post Honor CEO Seth Sternberg: ‘We’re Using the Past to Predict the Future’ appeared first on Home Health Care News.