≡ Menu

Internet of Things: A few speculative thoughts #ficoworld

Share

Larry Rosenberger, FICO Research fellow, ex-FICO CEO and generally fascinating analytic thinker was up next to discuss analytics and the internet of things. For all the “speculative” in the title, Larry things the whole area of the internet of things is becoming more real and less speculative. Larry began by defining the internet of things. Today, he says, information technology and the internet are almost completely reliant on human entered data. As a result they have more information on ideas than on things! This is changing as more things are able to be tracked and counted automatically, generating electronic data that describes the real world. The potential is to reduce waste, improve efficiency in supply chains, better understand the world around us and more. The history of the Internet of things begins around the turn of the century with RFID and exploding through the growth of smart phones and other devices with GPS and internet connectivity with much more to come as everything is instrumented. Some examples include urban planning, remote healthcare, smart meters and home automation and more were identified in a very cute Internet of Things Comic Book. While some examples are a little esoteric, others are clearly going to have an impact on our day to day lives. For organizations that are consumer-facing or involved in regulating consumer services there are a wide range of these applications that relate to what Larry called the Internet of People.

Before getting into some examples, Larry spent some time discussing the way traits, inclinations and situations contribute to your behavior. Whether you are responsible or not, inclined to be organized or not combine with a situation like having a new baby to drive behavior about, say, booking medical check ups for your child. He uses a tree as an analogy – you can see the leaves  (your behavior) on the  branches caused by each new situation but you cannot directly see the roots of your inclinations and definitely not the deep roots of your core traits. Plus of course you are not looking at one tree, but thousands or millions representing your customers.

For his first example he focused on drivers. Surveys regularly show that 90% of drivers consider themselves “above average” but this simply tells us that people are over-confident! FICO is doing some research with a company called drivefactor to collect “Big Data” from sensors and devices that might be in a vehicle. The drivefactor infrastructural uses the cell network to make it easy to collect this data and stream it. Many devices can be used from professionally or self-installed boxes to cell phones with their accelerometers etc. The first step would simply to show this data to someone to show how well they are driving. But this is messy complex data so how can analytics help? FICO has been developing a safe driving score based on analyzing the raw sensor data. This simplifies consumption of the data and makes it easier for people to see how and why to improve. Insurance data suggests that there is a 3x range of claims between the best and worst drivers and Progressive, for instance, has over 1M drivers hooked into a system that offers discounts based on safe driving.

Now it is an interesting fact that credit data can predict the risk the claims – better credit scores are strongly correlated with fewer accidents and claims. Clearly this is not because having better credit MAKES you a better driver. Larry’s hypothesis is that some of the same hidden traits and inclinations that drive you to manage your credit well are the same as those that make you want to drive safely. These two unrelated things are correlated because there is a common underlying trait. And of course this implies that data about how you drive might be usable to predict how you might handle credit!

A second example includes health sensor data. Lots of personal devices are now available to monitor activity and health – everything from speed of walking, heart rate and blood pressure, to EKG (BodyKom, an old favorite of mine). Even food diary applications are being targeted with new apps that allow things like taking a photo of your meal to get a calorie estimate. Now, once again, there are studies showing that credit data can predict who will stick to an exercise program. This leads inexorably to the same question as the first example – could health behavior data be used to predict credit risk? Again all because of shared, hidden, traits that drive both.

Finally he used an example of one on one conversations with customers. One path towards better dialogues is to consider their interests, wants, priorities and then use this information to drive longer, deeper customer relationships. The classic metaphor here is the comparison of mass marketing/customer service with the corner store (discussed on my blog previously here). Can we use analytics and decision management to enable a personalized next best action interaction at scale – being relevant and helpful through automation. Of course one of the problems in this approach is the lack of detail in feedback – we use surveys and market research which are not very fine grained. Also we are rooting our analytic approach in our experience in risk where we focus on very concrete data. Going forward we can do this differently, making dialogues that are genuine interactions including simple questions and preferences and collecting data from every aspect of the conversation. This data can be used to develop consumer utility functions as well as sequential analytics (how to sequence things) and so drive much more productive dialogues. To get started with this companies can begin with adding simple questions to the available actions – so the next best action might be a simple preferences question instead of an offer. Over time this can evolve to more conversational questions, analyzed using text analytics. In the end real-time responses, using geolocation data to provide context, will be central to these dialogues.

Some issues:

  • Privacy – zealots will not participate but others will, some are indifferent and will provide everything but most are pragmatists who will consider each question to see if it is worth answering, worth sharing their data
  • Security for selective sharing and confidence in storing data
Share

Comments on this entry are closed.