Irrational Product Design!
Designing Products that Marry Behavioral Science and Data Science for better user outcomes
Data Science and Behavioral science both deal with decision making. At the outset, that may seem to be the only thing common between the two fields because data science deals with building computer-based models, machine learning, statistics, etc. to optimize for economic outcomes while behavioral science deals with applying insights from human psychology to ensure people act in accordance with their long term goals.
Two Complementary Approaches to Decision making
However, on taking a closer look, there is much in common between the two fields and much to gain by complementing these two disciplines to influence user behavior and decision making.
Behavioral Science has well established that people tend to make irrational decisions because of mental biases (link). Notable works on these have been made by Daniel Kahneman, Richard Thaler, and many more. Mental biases tend to range over a wide variety from drawing different conclusions depending on how information is presented (Framing) to favoring preexisting beliefs while arriving at a decision (Confirming). Behavioral science tries to design people’s choice space and the way it is presented, to ensure decision-makers choose the options best suited for their long term well being despite the biases.
Data science uses a neutral, statistical approach and tries to be a corrective aid to the biases discussed above when people make decisions. It however ignores real-world effects of mental bias and thus falls short of being an accurate description of reality. Though a predictive model provides one with an unbiased, optimal solution, it does not consider psychological factors that come into play during decision making that make people choose sub-optimally.
Thus, even though one can get a list of all possible choices from a data science approach, making the choice is again a human problem where psychology and bias come into play. This has implications on designing product journeys and user flows as we will see below, especially about strategic goals that a product team is focussed on like creating new user behaviors
Much of Nudge economics has focussed on helping people make the right decision that suits their long term well being without confounding them with choices. Data science is mostly used here as a means to identify the correct path or decision one must choose based on available data.
The role of data science is limited to identifying all the choices available for a given problem. Post which, one uses behavioral science to identify how those choices are presented.
Surprisingly, you can also reverse the roles of data science and behavioral science to influence decision making in a scalable manner. You can build data science products to deliver ‘nudges’ and interventions to users based upon insights from behavioral science research and influence their behavior.
This becomes all the more relevant since a product team is concerned about creating and sustaining certain user behaviors among the users. Using behavioral and data science together would help one to systematically scale such new behavior adoption
The kind of behaviors that a product team wishes to build amongst users can also have long term strategic value, if not short term ones. The behaviors that can be built around in this manner can center around :
- Activation,
- Onboarding,
- Hand holding users along the ladder of engagement
- Re-engaging users
- Building social proof for the product
- Driving users towards discovering product value
- Engaging users in engagement loops
- Driving users towards value exchange
Delivering Nudges through Data Science
One example of this approach is in Insurance and Telematics.
Many Insurance companies have successfully built and deployed AI and ML-based data science products that monitor a user’s driving behavior to understand risk and update premiums accordingly. The data is collected in real-time from a device on the vehicle. Usually, parameters like speed, acceleration, collision, etc are stored and processed. The data is also shared periodically to data centers where the user’s behavior is scored, benchmarked, and compared with multiple other users.
The insurer can also choose to use this data to decide on updating someone’s premium or not in addition to multiple data science use cases
This is the traditional approach adopted by insurers.
However, the same data can be used to influence the vehicle owner’s behavior. It would be in both the insurer and the insurance companies' interest if a customer in the high-risk bucket can be brought to a low-risk bucket. It would mean premiums paid for a longer duration, fewer claims and an increase in customer loyalty due to personalization of services
The insurance company can try to influence the user behavior towards the above goal though ‘ behavioral nudges’ in the form of curated information. This can be delivered in a timely manner to change the user’s behavior.
The Magic of Contextually Relevant Bytes
The crucial thing to note here is that for the information to make an impact on the vehicle owner, it becomes very important to decide what information is shared and in what context.
Sharing judiciously selected and contextually relevant information would have the greatest impact. The guidance of data choice and delivery would come from Behavioral Science insights.
For example, one can choose to periodically send information on the relative rank of the user on key parameters to compare with the general population. This will reduce overconfidence bias amongst users and at the same time nudge them to take based on social proof.
Test, Test, Test
The use of behavioral and data science to design and deliver nudges would increase the effectiveness of such steps dramatically compared to using only one of the approaches. Post Design and implementation, measuring results become equally important. One must also make sure that there are enough behavioral actions in the product to differentiate between someone who displays the intended behavior after the nudge. Constantly designing AB tests to see if the interventions are having the desired effects.
References:
