Here at the American Health Insurers Digital Experience Conference (AHIP), Jeff Cribs of Gartner and Ang Sun of Cambia Solutions gave an interesting presentation on AI trends to Watch in 2018.
Example: the imageNet large scale challenge. Over the last 7 years it’s improved. in 2010, there was a 30% error rate. Today, Microsoft and Google have an error rate of around 5%. Humans do it slightly worse
Example: Google Photos does facial recognition. He showed a photo where a loved one was in a picture he didn’t even realize. His memory of his last years with this person is helped by the ability to get all those photos together.
If you want to consider what trends to follow in 2018, get used to the whole story. Think personal story. Both examples above are the same trend. One is inside out. The personal story is outside in. Both form the whole story.
Quote: Someday everything digital will improve its own performance without recoding
Good UX Means Good Business
In a world where technology is rapidly advancing and user expectations are rising, it’s no longer enough to have an average user experience; to delight your users and surpass your competition you must strive for the exceptional.
What is digital? Over 50% of payer industry users have done something digitally. Gartner likes to think of this in terms of ease of use vs trust. Members trust insurance companies to use AI for health insurance things like check benefits and copays. They don’t trust as much for online tools, diagnosis, etc.
Advice: Jump in with both feet………into AI Strategy. Understand that AI is still maturing and you can’t jump into a whole bunch of machine network solutions.
- How will the enterprise define AI. What is AI?
- The the enterprise views the impact of AI on the payer industry overall.
- Where the payer organization sees its most valuable use cases for AI
- What criteria will be used to determine how to acquire AI capabilities
Ang Sun is the Chief AI Officer at Cambia. He was the senior data scientist at Expedia before that. He’s a “strategic” data scientist. His personal background is natural language processing.
Cambia uses AI for three things:
- HealthSparq: empowers consumers with curated choice on services options. Also allows for comparison of cost
- MedSavvy: Enables better medication choices. Simplifies the jargon. Informs the conversation with physicians.
- Personalized Care Management Tool. It’s a data driven view of the member and allows you to enable an optimized personalized care approach. It makes predictions on where they will be.
The power behind the solutions above is AI. At Cambia AI is less about AI and more about augmented intelligence. They are building AI with humans. AI algorithms know their limits and when to ask a human for help. The AI learns from feedback.
Cambia applies augmented intelligence to front and back end use cases. One model prioritizes the list of claims to be reviewed. This resulted in a 15X reduction in claims reviewed with 3X more savings. In this model, the AI detects issues before they actually pay the claim. The bill to the member is correct the first time.
On the front end, Cambia uses a smart bot. They use a bot to answer a number of questions. The member feedback is great with comments like: “It’s simple and easy to use”, “I like the bot’s personality.” The bot reacts within seconds and starts answering questions quickly.
Trends to Watch
- Machine learning
- The most mature area
- Amazon, MSFT, and Google are investing heavily in this infrastructure
- Can help in care management, risk adjustment, and underwriting.
- Computer vision
- Think early diagnosis of charts, scans, etc.
- Relatively mature area but you need a lot of training data
- Natural language processing
- Not very mature
- Think Amazon voice service
- Entry level is lower now because of open source package like Open NLP
- Can do this with your developers without hiring data scientists
- Conversational User Interface
- chat bot
- Least mature area even though it’s one of the hottest areas
- Most chat platforms are not chatbot platforms. They are rule based vs learning based
- Need to help the bots build knowledge and needs to have the ability to learn over time
Best Practice: Back end of your AI platform MUST be service oriented.