Personalization is the future of Applied BeSci
A maturity curve to get there
Behavioral Science Maturity Curve
Richard Thaler’s canonical story of the power of default in altering organ donation began the movement of modern-day applied behavioral science. Finding applications across industries, nudges appear as a panacea where minor environmental tweaks can yield a significant impact. While useful, the greater truth is that nudges are seldom the one-size-fits-all solution choice architects once believed them to be, and behavioral change is still as complex as ever.
At scale, a choice architect can rarely release a single nudge believing it will have a universally positive impact. Instead, when diving into details beyond the average effect size, nudges might not affect specific individuals, and some people may suffer adverse outcomes. In any organization that collects data on the usage and behavior of their product or service, there is potential to find the actual value of nudge. But first, it requires getting on the path to the personalization of interventions.
Understanding the need to personalize interventions begins with discussing whether we have overvalued nudge potential, especially when considering them as the go-to solution. Companies have yet to emerge as highly successful behavior-changing products, allowing them to position themselves as experts in personalizing interventions. Lirio, led by Amy Bucher, is emerging as one of the more progressive organizations encouraging improved decision-making through tailored nudges. This article will cover a maturity curve for teams hoping to develop the capabilities required to begin personalizing interventions.
Defining the personalization of interventions
Personalization of content has been going on for years, with Netflix being one of the most advanced. Their example is they target content to users based on various characteristics and identify the best thumbnail for the show to get you to pause and choose it for your viewing (an intervention in its own right).
Personalization for our purpose is targeting specific features, notifications, and interventions to get users to follow through with their intended behaviors. Looking at Lirio, this looks like targeting a reminder to attend a doctor’s visit. In this example, an individual has indicated that they want to begin a behavior that will benefit their health. By targeting an email reminder that invokes a particular call to action, they intervene to help that person follow through with their intentions.
Building on a framework laid out by Stuart Mills, I define personalization across three main factors:
- The first is the delivery channel, which is how the team will deploy the intervention in an individual’s life. Within technology, this can look like emails, notifications, IM, text messages, etc., but it can also take the form of policy.
- The second is the specific design choice the designer suggests. In e-commerce, this could be the most recommended flavor of jam or the bike, which will most likely satisfy the customer’s needs.
- The third is the behavioral technique that will most effectively lead them to the decision. For example, one message could focus on the potential for a loss that inaction will cause (loss aversion), and another could focus on the number of other individuals displaying the specific behavior (social norms).
Eight components for personalization of interventions
Advancing a team’s competency in tailoring interventions requires eight resources. It is not the responsibility of the behavioral science team to create them. They must orchestrate this process, ensuring other teams build them for future unification.
A quick note on resource requirements:
Creating the components required to personalize interventions demands an investment across multiple teams and tools. Anyone pursuing this will likely need a solid business case to support investment. I plan to write separately about the growth opportunity of developing products that create behavioral change. For now, I will make the case that many successful startups and rapidly growing businesses have positioned themselves around creating behavioral-changing products. These focused services only appear to be growing. For a detailed list of examples, check out this great resource on the behavioral institute.
Behavioral profile framework: The foundational framework outlines the types of data connected with the product or service that the company is creating. It is a future-looking document that should include both the data that one currently has and what to collect in the future. The team will fill this framework with baseline values after establishing archetypes.
User outcome metrics: At its core, every product has an outcome it tries to drive for the customer. Sometimes these are easily measurable, such as whether or not a weight loss app is helping an individual decrease their calorie intake. Other times, it is complex, such as whether an at-home fitness device helps build confidence. The user outcome connection starts with the outcome that a product desires to drive and connects to specific behaviors which will alter that outcome. It then ends with proxy measurements for that behavior.
Experimental platform: Generally owned by a data science or research team, this is the collection of tools that allows for the flighting of different versions or interventions of a product or service and the collection of the difference in the outcome. A human resource should handle inferential statistics within this experimentation platform to determine and validate hypotheses.
Passively collected data: A system, most commonly digital, that measures a customer’s actions throughout a product or service journey. Anything at the scale requires a pipeline between products and analytics tools to query and examine data quickly.
Archetype: These are user groups built by examining users’ different personas or classifications. Archetypes can provide the basis of personalization and quickly classify users after beginning usage. Cohorts will have established baseline values for each data category in the behavioral profile framework.
User feedback systems: A user sentimental and emotional data collection system that passive data systems cannot collect through the observation of usage alone. This user feedback system can range from interviews to product surveys.
Beta pool users: A collection of users identified that they are willing to participate in research. Starting with these users significantly decreases the risk of upsetting customers with tests. However, they should be one of many groups tested as they inherit bias.
Interventions: Any feature or change to a product intended to convert behavior in some fashion. Ideally, we build on user research, which suggests that the users have some intention to bring to the service that is currently not fulfilling.
Introducing the maturity curve toward personalization
Now that we have covered all the components needed to create personalization of interventions, I will introduce a maturity curve describing the stages to get there. This graphic should be considered a guide where one can advance after meeting specific conditions, although everything must be in a different order for a team to go to the final stage. Each step represents a significant milestone in a team’s ability to weave behavioral science into the product and develop the interventions that the designer will ultimately tailor to users.
Stage 1 — Evangelization
The first stage is behavioral science Evangelism. A behavioral science team at this phase is anywhere from brand new to spending most of its time introducing other groups to the power of thinking through a behavioral science lens. Teams at this stage still need to create proofs of concepts, introduce people to the basics of BeSci knowledge and other foundational tenets of the field, and generate buy-in toward the vision of applying behavioral science. These teams are usually a single individual starting a behavioral science practice or a collection of individuals interested in creating a more extensive unit but have yet to get the support to develop their team.
At this phase, the team should develop foundational frameworks on the types of data they want to collect to understand their users more deeply. As this will be a primary artifact used throughout this journey, the team should consider what is currently possible and develop a vision of areas to expand. Building this out will require thinking about the outcomes the team is trying to achieve; for example, getting someone to work out will be affected by everything from how they feel in the moment to their identity as a fitness person. This framework will be fundamental as the team advances to the next phase and needs to hone its scope on what behaviors they desire to change. Kameleoon provides a great example of this type of framework where they include: time spent on site, purchase history, location, and weather at the primary location, among many others.
Example: The new behavioral scientist at Savings Co. focuses on introducing their various partner teams to the basic concepts of Behavioral Science. Sometimes they do this in live presentations, sometimes by recording videos or writing articles.
Required Components
- Underlying framework
Stage 2 — Application
The next phase is Application which occurs after buying has happened and the team is ready to incorporate behavioral science insights into their product or service. Teams at this stage no longer have to explain the basics nor justify their existence; instead, they focus on getting features into the product. In this phase, groups of behavioral scientists begin to form, with each individual specializing in an area where they can contribute to seeing their applications brought to life. A single behavioral scientist can carry a team to this maturity but must do so by connecting with other researchers. Intervention development is necessary if it has yet to occur, with various examples here.
The main workstream to help advance a team toward the Stage of Application is the development of user outcome connections. These connections are foundational to building behavior-changing products by describing the relationship between the user’s desired outcomes, from usage to the specific, measurable behaviors that change them. Ideally, the product developers will also connect these user behaviors to business outcomes such that users will return and share positive feedback regarding the product when fulfilled. An example of a user outcome connection would be a finance app that helps users save money for retirement. The user outcome is a desire to increase cash in a savings account, correlating with user retention. The connected behaviors would include setting aside money at every paycheck, rounding out purchases, and directing the additional change towards one’s savings. Both behaviors are measurable through the data collected on usage and the user outcome by the amount of money in the account.
Powerful user outcome connections have measurable behaviors with data collected through usage because it allows for scalability. Although as a fallback, we can also use user surveys or interviews that measure a specific behavior. Ideally, as a team grows, they will utilize passively collected data generated through any digital interaction that a user has with a product or service. Advancing the passively collected data pipeline means not just deploying and organizing the inflow of user actions; it means deploying it in a way that allows teams to query and analyze it quickly.
Example: Savings Co.’s BeSci team researches the most significant barriers to users allocating parts of their paycheck to their savings accounts. They recognize that during the time between when an individual wants to start a savings account and when they receive their pay, they lose motivation for their long-term goal. Thus, they develop an intervention where users can direct a future amount of their compensation into savings when exploring savings options.
Required Components
- User outcome connections
- Interventions
- Passively collected data
Stage 3 — Experimentation
The third phase, Experimentation, is when exponential value begins to occur. Teams take the previously developed interventions at this phase and test them for effectiveness. They also utilize the user outcome connections they have constructed to see what tweaks to the user’s environment may have measurable outcomes or user behavior. A single individual can still be the leader of this thinking but, if so, must have a strong partnership with either a data science or experimentation group. Ideally, at this stage, a team will have formed with a qualitative researcher who can focus on developing interventions based on what users share they would like to change or improve and a quantitative researcher who can assist with testing and validation. Interventions at this stage will begin to be ranked by their overall effectiveness, dropping ineffective ones.
Rolling out experiments requires expertise; doing it at scale requires the proper systems to be in place. Organizations that are already personalizing content or using models to maximize user engagement should have the infrastructure in place, but those who do not could benefit by first investigating simple A/B tests with in-app notifications (with a platform like OneSignal) or emails (example). After achieving wins, the team can further enhance the systems in a way that tests more than communications; instead, experimenting with the flighting of features to see which combination creates the best user experience (such as Optimizely or LaunchDarkly). If all of this sounds too complicated and you don’t yet have the necessary buy-in, the behavioral scientist can start by rolling out a simple “paper prototype.” Show the value experimentation can bring a firm and make a case for investing in the proper infrastructure.
While ideally, at this point, the team will have established a system of collecting passive data from usage that offers a picture of whether an intervention is successful, attitudinal user-generated feedback is necessary to complete that picture. Thus, a user feedback collection system is the next essential element for this phase. These can be as simple as user interviews but ideally are deployed as an in-product survey. Taking the ladder approach allows for timely feedback and rapid collection of attitudinal feedback. The most cutting-edge way to pursue this is using bot technology that can slip questions into the user flow in a helpful way where the user feels they are getting value for answering the inquiry.
When deploying these services and experimenting at scale, the best path is to start with a select group of users. The team should do this with users who have indicated they are open to participating in experimentation, which I call a beta user group. One of the best examples of this is Microsoft Windows Insider. Users who are excited about access to early Microsoft feature releases will be able to access them in exchange for feedback. By participating in the program, the Windows team can rapidly experiment without the risk of upsetting unaware customers. This beta group can be one of many sources of experimentation as they hear significant bias by self-selecting into their participation. Still, they serve as a great starting point before expanding the testing to the entire audience.
Example: On top of the previous intervention the team created, they developed another, which begins to round up a user’s purchases and deposit the money into a savings account. Utilizing its newly created experimentation platform, the Behavioral Science firm tests the features to see the most effective.
Required Components
- Beta pool users
- User feedback engine
- Experimental platform
Stage 4 — Personalization
The final stage, a never-ending iterative process, is actual Personalization. Here, teams begin to craft the ideal change environment for each subgroup or individual because this level of capability requires a team. Each archetype will estimate how much change content they can handle before disengagement, their best channel for communication, and the behavioral techniques which have the most effect. Each intervention can predict how effective it will be for each subgroup while tracking those with potential adverse outcomes. At this phase, the behavioral science team working in concert with the product team can begin to realize the vision of a tailored experience based on user characteristics.
The only additional component required to succeed at this stage is the development of archetypes using the data collected as part of the foundational theoretical framework. Archetypes should be developed based on clusters of users identified through an unsupervised learning model or similar processes. A great example is what Cedar has done with their customers, creating four main types of users. These groups serve as the baseline values for each primary data type and can be updated according to specific user behaviors. In other words, a new user can receive a tailored experience immediately based on their group membership. Then, as they continue using the service, the experience will become further personalized as they generate more data that updates their unique profile.
Example: The Behavioral Science team, in collaboration with data science and user experience teams, create two archetypes. First is the Forgetter, who wants to save more but forgets about their plans, and the second is the Reluctant, who only saves because someone told them it is a good idea. The team takes the two previously mentioned features and begins testing them to see which is best for each archetype so that when future users join, they can advertise the most compelling feature for them.
Required Components
- Archetypes
Where to begin
Getting to the final personalization stage is lengthy and detailed, but this maturity curve intends to lay out a path to success. For individuals trying to get their team on this path, start by aligning stakeholders around each component while building a vision deck that connects each effort to a larger strategy. One individual can ultimately choreograph this entire process, but it requires many people’s effort to get all sections running.
As with all proposals to interweave science into a product, it requires a strong business case. The investment in aligning a team around this vision is significant. Still, the payoff is ultimately building products that help achieve the intentions that users bring to them and, in turn, which they will retain. Personalization of products can increase revenue through retention and further sales of services to consumers who believe your brand yields success. It also can decrease costs by lowering customer acquisition costs as viral spread occurs through consumers sharing the gains they have achieved through usage. Regardless, if you utilize these or other connections to the bottom line, start with a strong case around the benefits that a vision of personalization will bring both your customers and the business.
Please clap 👏👏 if you find this post helpful. Thanks!
