My good friend Vinnie participates in an automobile insurance program that rewards him for good driving behaviors; the better driving behaviors he exhibits, the more money he saves on insurance. You stick a device into the vehicle’s diagnostic port (usually under the steering wheel in most vehicles manufactured after 1996), and the automobile insurance company tracks your driving behaviors and offers you automobile insurance discounts based upon the quality of your driving behaviors. The program actually “grades” driving behaviors including acceleration, turning, speed and braking, and once a month sends a report card on the past month’s driving performance (see Report Card in Figure 1).
And for my friend Vinnie, as a result of sharing his detailed driving data, he saved $1.49 over the past 6 months. That’s right! Vinnie is saving $2.98 per year by sharing his detailed driving data with his auto insurance company. That’s not even enough for a Starbucks’ Venti Chai Latte!
Now maybe if Vinnie had a better report card he might get enough of a discount to afford two Venti Chai Lattes every 12 months (he’s got about a C+ average…).
The objective of this blog is to highlight how combining big data with behavioral economics to provide timely, detailed and actionable recommendations can influence behaviors and drive desired outcomes.
The purpose behind the insurance company’s program and the resulting report card is to make drivers better insurance risks by improving their policyholders’ driving behaviors – to get their policyholders to conform to whatever this particular insurance company has determined as the best practices for turns, acceleration, braking, speeds and time of day driving. And incenting people to change their behaviors through monetary and other rewards is the heart of Behavioral Economics.
Behavioral economics seeks to quantify the effects of psychological, social, cognitive, and emotional factors on the economic decisions of both individuals and institutions including purchase, pricing, consumption, financial and resource allocation decisions.
Influencing behaviors and habits requires an awareness and feedback with respect to one’s performance as compared to industry norms, benchmarks and/or segments of similar others. However, in many situations, the timeliness of this feedback greatly influences the degree to which we can influence the behaviors; that is, the most effective outcomes are achieved when the latency of the performance feedback is minimized.
For example, imagine that you are a baseball player whose batting average has slowly been declining over the past several months. Getting a “Batting” report card once a month or even daily does little to help the player understand specifically and in detail what behaviors to fix and what behaviors to continue.
Without nearly instantaneous feedback on the batter’s behaviors and performance, there is no learning.
In order to fix the batting problem, the player would want feedback on each pitch regarding hitting variables such as hitting approach, pitch selection, stance, swing mechanics, balance in the batter’s box, weight transfer thru the swing, etc. The key point is that if one wants to change behaviors then the feedback on that behavior (e.g., batting, driving, smoking, diet, exercise, performing arts, finishing cement, installing electrical) must be specific, detailed and nearly instantaneous (see Figure 2).
With the Internet of Things, new devices and sensors are providing new metrics that can be used to change behaviors. Continuing our baseball example, there are two new metrics that leading baseball organizations are exploiting to try to improve batting behaviors and outcomes:
- Launch Angle measures the vertical direction of the ball coming off the bat; a launch angle of zero degrees would be a flat line, with positive numbers indicating an upward ball flight and negative ones indicating a ball driven into the ground. Hitters with high launch angles tend to be sluggers who produce lots of fly balls.
- Exit Velocity represents the speed at which a ball leaves the bat; a detailed measure of how hard each ball was hit. Unsurprisingly, at the top of last season’s exit velocity leaderboards you’ll find the game’s greatest sluggers — such as Giancarlo Stanton (99.1 mph), Miguel Cabrera (95.1) and Jose Bautista (94.3) — with average exit velocities greater than 90 miles per hour.
If the hitter knows these two metrics – plus other traditional hitting metrics – immediately after each swing, then the batter will be able to constantly tweak or refine their batting mechanics in order to achieve a desirable result (like a higher batting average and slugging percentage).
Supervised Machine Learning Learnings
The similarities between these Behavioral Economics concepts and Supervised Machine Learning are striking. Let me explain.
Supervised Machine Learning draws inferences from labeled outcomes or responses such as fraud, customer attrition, purchase transaction, part failure, social media engagement, or web click (versus Unsupervised Machine Learning which draws inferences from data sets without labeled responses).
If a baseball batter knows the desired results (e.g., exit velocity, launch angle) after every swing (labeled outcome), then the human mind and the hitting coach (like a supervised machine learning algorithm) can process the different hitting mechanic variables to determine which of those variables – hitting stance, swing mechanics, bat grip, weight transfer, hitting approach – might have impacted the outcome. And over a large enough detailed data set of swings, the batter – and their hitting coach – would be able to detect patterns or behaviors that are indicative of a good (versus bad) swing mechanics (see Figure 3).
Insurance Company Economic Benefits
One last point before wrapping up another overly long blog, this insurance “driver data” program enables the insurance company to collect detailed driving data that can be mined for new insights that can ultimately be used to create new monetization opportunities across a number of dimensions including:
- Traffic conditions by geo-locations, time of day, day of week, holidays, etc. that could be valuable to logistics and delivery companies
- Performance benchmarking of cars that could be valuable to car manufacturers, car dealers, finance companies and consumers
- Road conditions (potholes, bumpy road sections) that could be valuable to the Department of Transportation, construction companies as well as logistics and delivery companies
- Age and condition of cars which could be valuable to automobile manufacturers, car dealers and service stations
- Driver behavioral segments which could be valuable to marketing companies and maybe even law enforcement (watch out Vinnie!)
I’m not sure how customers are being rewarded for sharing their detailed driving data that enables these new monetization opportunities, but I suspect it’s worth more than an annual free Venti Chai Latte!
So like how a good hitting coach can make a batter aware of good and bad hitting behaviors with specific, detailed and nearly instantaneous feedback (the heart of effective behavioral economics), organizations that want to incentivize their customers, partners and employees (and other humans) to “improve” their behaviors, need to provide specific, detailed and nearly instantaneous feedback. That’s just like what an effective supervised machine learning algorithm would do.
Figure 3: “Swing the Bat”