If you want to get a sense of the art of the possible, go to China. Look at WeChat and how it is integrated into people’s daily lives and bustling ecosystem of eCommerce players.
Then take a look at ZhongAn. This is an online-only, tech company that knows how to sell insurance to Millennials!
They’ve only been around for 3 years or so and yet have managed to sell over 5 billion policies. Impressive traction for an InsurTech with only 1500 employees, no salespeople and over half its workforce are engineers and developers.
Their ability to scale so quickly is made possible because 99% of all underwriting and claims operations are automated!
This is artificial intelligence in action and hardly makes #3 in my InsurTech 2017 predictions look very profound!
AI will dominate mobile customer engagement
To get a better understanding of the AI customer engagement landscape, I turned to an InsurTech startup who are building customer-centric automated experiences using an in-house hybrid bot.
If you don’t know of SPIXII, where have you been? I first featured them a year ago, around the time that Facebook opened up the Messenger platform to chatbots. There are now over 35,000 chatbots running on Messenger with access to a user base of 1 billion.
SPIXII are pure InsurTech. They went through the first Startupbootcamp InsurTech program and have just finished at MassChallenge and Allianz Accelerator. SPIXII is available in 6 languages and is currently live on a major insurer’s website in France.
To get an update on their progress, I caught up with Alberto Chierici, one of the three founders of SPIXII.
“This is quite a jump for insurers,” Alberto told me. “What we have done is to implement bots into a chat window. To start with, it’s not a full customer journey on the bot. At some stage, we hand over to the insurer’s standard workflow.”
If you think about it, this ‘one-step at a time’ approach makes sense. There is a lot of hype and heightened expectations about how far the tech can go today with machine to human engagement. Conservative estimates claim that the best generative bots out there today are achieving between 60% and 80% success.
And on the flip side, us humans are not yet universally fully conversant with engaging with a machine, certainly not in the West anyway.
It’s a different story if you go East, where adoption of mobile and digital engagement has been much greater. You only have to look at WeChat and the phenomenal success of their eCommerce extensions. This has enabled AI-driven chatbots to handle around 70% of all online sales through WeChat.
It’s all about Data and Convenience
“The thing with any AI solution,” continued Alberto, “is that it needs data to build up the capability. This is why we are starting the initial customer engagement through a web-based chatbot. SPIXII is a non-invasive solution allowing carriers to implement it with no effort into their existing distribution channels. The next pilots we are setting up will expand to another channel, Facebook’s Messenger.”
What I like about this is the convenience this offers the customer.
Imagine this scenario. You’re at the dentist’s in the waiting room. You’ve just been told you need some work doing and want to get it booked in. Obviously, you don’t have your dental plan with you and don’t know what you’re covered for. You jump onto Facebook Messenger, open a chat with your insurer, confirm your ID and the bot comes back with your policy details. Data-enabled convenience!
And the upside for the insurer is they don’t have to handle the call in the call centre. I’ve seen estimates that as many as 80% of all calls are routine and “trivial” and can be better handled by an automated system. (“What’s going on?” is a subject I covered previously.)
Intelligent Automation is the sweet spot for customer engagement
I picked up this subject recently in a call with Donna Peeples, Chief Customer Officer with Pypestream. They have developed a secure messaging platform that serves multiple industries. This is an industrial scale platform that uses intelligent automation for customer engagement.
I was interested in how Pypestream has been used in the insurance space. As the former CCO for AIG, Donna was a great person to ask.
“With insurance being in a low interest, low involvement category for consumer attention, we had to find new ways to engage with customers,” Donna told me. “Texting and chatting has taken over as the primary method of communication, and it’s expanding across all demographics, not just for the Millennials.”
“The problem is more difficult because Insurers tend to think of customers as a policy or as a claimant. But the world has changed and consumers expect so much more from their providers. Namely, convenience and instant response and anytime, anywhere communications that doesn’t need to involve a phone call.”
From a customer’s perspective, this all sounds very reasonable. We are in an age when we have an app to tell us where our taxi cab or pizza delivery is, but if we want to know the status of a claim, it’s not so easy. Hence the rise in “self-service” as the way to go for personal lines claims.
One of Pypestream’s customers is Lynx Services, a subsidiary of Solera Insurance Services. Donna explained what Pypestream have done with Lynx. “We have developed an automated chat capability in the auto claims space. This serves both the end customer and the third party service/repair shops who are able to do everything from initiate through to complete transactions in real-time.
“With every company, there are only a finite number of questions customers ask. Pypestream has developed a guided decision tree model with dynamic routing for chatbots where customers are given options within the message stream. With the same type of queries being asked, bots can be intelligently designed to answer them in real-time.”
Not All Chatbots are the same: Natural Language versus Workflow
In the massive spectrum of AI, there are many variations and differences in definitions. And there is no exception within the specific part of the AI world that includes the Chatbot.
Alberto explained it to me; “broadly speaking there are two approaches to chatbots. The first is to use a natural language approach using AI to interpret human language, your intent, your context, your meaning.
“Or, there is a structured workflow approach using automation to map a dialogue with a customer to an existing business process. In this approach, the Chatbot will use mostly closed questions or respond with a direction like “now upload a photo of the claim”.
“At SPIXII, we have developed a platform that is a hybrid, somewhere in the middle of these two approaches.”
This is a great perspective on the full spectrum of AI capability when it comes to customer engagement.
Hype or Reality: The art of the possible
It is usually the case with all evolving technologies that we commentators can get a little fanciful. We’re seeing some of that from the headline writers for stories about insurers writing new skills into AI Alexa on the Amazon Echo.
From Liberty Mutual and Nationwide in the US to Aviva in the UK, they have been the first to add an insurance “skill” to the 3 million units sold to date. Personally, I’m not convinced that homeowners are going to turn to AI Alexa to access an insurance glossary of terms. But they might once the skill is developed to answer a personalized question about their own policy.
Another story that has gained a lot of headlines comes from Lemonade. AI Jim, Lemonade’s automated claims bot that just paid out a claim in 3 seconds!
The whole process actually took around 6 minutes for the customer to log the claim, submit a video testimony, get approved and get paid.
But you can see why Lemonade are proud of their 3-second achievement. Because, in just 3 seconds, immediately after the customer touched Submit on their phone, AI Jim reviewed the claim, cross-referenced it against the policy, ran 18 anti-fraud algorithms and issued wiring instructions for the full amount less a $250 deductible.
Hype or Reality: Google’s Zero Shot Transition
In this AI space of human: machine communication it is impossible to ignore that is coming out of Google. Without making much fuss about it, Google changed the way Google Translate works last September.
In a nutshell, Google Translate used to be like a giant phrasebook, matching your search request to its library of words. But, it was limited to only the words it knew and it couldn’t second guess any meaning.
Then came GNMT (“Google Neural Machine Transition”). It is smart enough to learn from those that use it. It can determine meaning and context and, most significantly, it can fill in the missing words. It’s fascinating and well worth the 5-minute distraction to read this blog by Gil Fewster.
The slippery slope of singularity?
It depends on whom you listen to. If it’s Elon Musk, then apparently yes. He says AI is “the most serious threat to the survival of the human race” and talks of a world where humans are treated “like pets” and machines are connected to humans by a “neural lace”.
However, since I don’t know the CEO of Tesla, I turned to Alberto from SPIXII.
“History tells us that humans always adapt to the machine, not the other way around. Look at driving a car, we all adapt to how the machine works. The same goes for search engines. Our brain has adapted to only type keywords, not whole sentences. We have changed the way we communicate with the machine. We have adapted to the machine.”
I have no idea if we should worry about singularity. Somehow I don’t think so. From my view of the world, I still see a need for human involvement. Even in insurance!
If you’ve just been in an auto accident or have a serious health claim, you’re going to want to talk with a compassionate and understanding human being, not an AI powered Chatbot through a mobile app!
Help or hindrance: the role of the regulator
And of course, there’s the role of the regulator. In the European Union, the reform of the Data Protection Act comes into force in May 2018 in the form of the GDPR. This has major consequences for the use of AI in the insurance industry.
For example, Article 22 states that individuals “shall have the right not to be subject to a decision based solely on automated processing. That will impact both underwriting and claims!
And Article 13 states that individuals have the right for a meaningful explanation of how an AI based decision is made. “What do you mean, the computer said “No”?”
Are insurers going to want to exclude segments of customers who withhold their consent to be AI’d? Are InsurTechs going to share the inner workings of algorithms when they are their competitive advantage? The answer is most certainly going to be a No to both questions.
There will be no avoiding this either; GDPR will be law and not a directive (as the DPA was). And it will carry a penalty of up to 4% of annual turnover for any infringements. This isn’t going to a regulation to be taken lightly!
Engaging with a robot; Hugh Terry style
On a lighter note, and to wrap up this month’s edition of InsurTech Insights, here’s what happened when Hugh Terry interviewed Sophia, the robot from Hanson Robotics, at the Digital Insurer Asia conference in Hong Kong last November.
The author, Rick Huckstep, is an InsurTech thought leader, public speaker and Chairman, The Digital Insurer.