At the well-funded startup Omada Health, coaches teach patients to prevent diabetes by eating better and exercising. They don’t meet face-to-face, but communicate over the internet — and the coaches are increasingly aided by a machine-learning-powered software that provides cues for interacting with the patients.
Since the San Francisco company was founded in 2011, these coaches were a mix of full- and part-time staffers. But in November, Omada let go of all the part-timers and instructed the remaining coaches to rely more on the software, the company told BuzzFeed News.
CEO Sean Duffy insists that his long-term goal isn’t to replace people with software. “The thesis is that we don’t think we’ll ever be at a point in Omada’s trajectory where we’ll ever take people or coaches out of the equation,” he told BuzzFeed News. “But they’ve got really smart systems to help them.”
Like many other tech-enabled health care services that connect patients with experts — coaches, therapists, nurses, doctors — over video chat, email, and text, Omada is trying to navigate a fundamental shift in labor. People are expensive — at least compared to automated, data-driven chatbots that could give advice and diagnose diseases without needing a salary or a college degree. But bots aren’t nearly as good at holding conversations, perceiving emotions and subtext, and delivering sensitive information like, say, a cancer diagnosis. If they want to grow, startups will have to figure out whether their patients and businesses alike will be best served by man, machine, or some blend of the two.
“There’s a spectrum of totally autonomous machine learning and the other side is totally human-driven,” said Mike McCormick, principal at Comet Labs, a venture capital firm that invests in artificial intelligence startups. “And then there’s every shade of gray in the middle of that.”
The previously unreported cuts at Omada last fall were small, and affected 10 to 12 part-time coaches, according to a spokesperson. It has about 60 to 70 full-time coaches and 250 employees overall. In another set of layoffs that Duffy said were unrelated, the startup also laid off roughly 20 workers last week, saying it “had to focus on Omada’s core business and expertise, while orienting the company for long-term success.” Omada has raised $125 million in venture capital, including $50 million this month in a round led by the health insurer Cigna.
“There’s a spectrum of totally autonomous machine learning and the other side is totally human-driven.”
Duffy said that as Omada has treated more patients and collected more of their data, it’s trained an algorithm to detect important behavioral changes. For example, if a person weighed in on a digital scale every day consistently, then stopped weighing in for three days, the system would flag the coach. It’d then “suggest messages they might send to a participant that might result in an outcome” — in this case, to find out how a person is doing and why they’re skipping weigh-ins, Duffy said.
The CEO was quick to note that the machine isn’t prewriting messages down to the word, but rather suggesting a gist to convey. He said that users could tell when a nominally human-written message is computer-generated, and that this makes them lose trust in the system.
Coaches can also say no. “If we get enough coaches declining these suggestions and saying, ‘That violated my intuition as a human being,’ it trains the system to get better and better and give better and better suggestions,” he said.
The part-timers had access to this technology, but Duffy said that the company benefited more from having full-timers who are constantly involved and invested in improving it.
Omada isn’t the only company exploring how to use AI to improve health care. Startups like Babylon Health, HealthTap, and Remedy are developing chatbots to assess patients’ symptoms. Big Health has an entirely automated program called Sleepio, which stars a cartoon professor and is designed to help people with severe insomnia. But these nascent technologies are too new to definitively prove that machines can improve health more than humans can.
To survive, any kind of virtual health service will have to prove that it can get people to sign up, stay involved, and actually improve their health, said Liz Rockett, director of Kaiser Permanente Ventures, which has invested in both Omada and Big Health. “Doing the work of proving efficacy and reach is the best way to define that line of what works and what doesn’t – including on the question of using coaches in the delivery or having an all-virtual offering,” she wrote in an email to BuzzFeed News.
For now, there are way more people-to-people telemedicine services. Ginger.io initially tried to infer behavioral patterns and mental health problems from passively tracked smartphone data, but switched to a text and video-chat model with human therapists.
So in 5 to 10 years, will patients be more likely to interact with a human or a chatbot when they open up a health app? It’ll largely depend on how high stakes the situation is, McCormick said. You’d want to hear that you have cancer from a trained expert with an extremely high degree of accuracy and emotional sensitivity. But for, say, nutrition coaching, he said, “maybe people are ready now ... It has to be nuanced.”