Chatbots hit a hot streak in 2016. It seemed like every week there was a new bot. Many of these bots faded as fast as they came on the scene. Most likely, they weren’t backed up with a proper business model, and many in the industry overestimated the capacity of chatbot and AI to tackle problems.
We’re going on almost two years working with this technology in the healthcare space. Early on, we realized that a pure AI chatbot would never be able to replace a physician or benefits expert. Our solution to the problem was the “human-augmented chatbot.” Our hybrid approach combines the best parts of both man and machine. We’ve had to develop quite of bit of technology on the front and back end to make it work. Service is not all in real time, but we set the right expectations with our users and it’s much faster than the alternatives. We’re saving companies hundreds of dollars per employee with our technology, so we’re seeing a lot of traction in the marketplace.
Why the human-augmented chatbot?
AI-powered chatbots have obvious advantages over humans, but their limitations are just as important to understand. They are better than people at computational tasks, crunching huge data sets, looking for patterns, performing repetitive tasks, and viewing millions of records in seconds. All of that sounds good, but there needs to be a human on the other end who tells the AI how to use those skills. Take our planned “population health automation” technology, for example. Our medical team can determine risk factors for an illness, and the chatbot can crunch the CDC’s influenza location data, the age of an employee, the employee’s insurance information, and pharmacy pricing data into our system. Our medical team can then review the output to make sure we are identifying the high-risk individuals. Combining medical knowledge with a computer’s speed, we can send notifications to all employees who fit a specific criteria and need a flu shot.
Our bot can then hold a conversation with employees about getting a shot at the closest location that’s covered by their insurance. This chat interaction — the information intake and conversation with our member — would be cost-prohibitive to do with a human staff but only costs a few dollars to run on our servers and reaches millions of people. This is a simple example of the potential of this technology. Just getting employees immunized in November instead of December can save an employer between $63 and $95 per person.
Our chatbot is designed to take these lower-level tasks and leave our staff to pursue higher-valued interactions, such as those involving human emotion. The bots never read between the lines, so they cannot handle the complexity of every healthcare interaction. We also never want our chatbot to handle specific subjects nor to give medical advice (a big legal no-no.) This model of the human-augmented chatbot allows companies to provide a greater level of service at a lower cost.
Designing your next chatbot
If you’re trying to create a chatbot, take time to think about whether it’s the best approach for specific types of interactions. Don’t be afraid to add real people into the mix; use them for what they are best at, communicating with other people. Start your automating with low-end tasks, and move your way up the value chain. With this approach, we’ve been able to use automation to increase utilization of telemedicine and our healthcare concierge to over ten times the industry average. It’s given us a huge advantage over our competition, and we expect to see big growth in 2017 because of it.
Chatbot technology is here to stay. It’s the perfect technology and interface for many industries, including one as complicated as healthcare. And it’s user friendly. In user testing, I’ve seen 80-year-old grandmothers enjoy interacting with our chatbot. Even for people who have trouble using other tech, it’s the perfect human interface.