Tenets of digital behavioral change and the renaissance of Generative AI
Behavioral change at scale requires the personalization of interventions; this has only recently become realistic with the acceleration of Generative AI.
The Tenets of Personalized Interventions
Creating new habits using digital products provides a world-changing opportunity. Yet, no team has seen global-scale adoption of a tool that yields measurable positive behavioral change fulfillment. Two decades of work have advanced the field tremendously, leading us to an age-old realization that humans are complex and unique. To change, we require tailored solutions that match who we are, what we are trying to do, in what environment we act, and our social interactions. Personalized interventions are the answer; below are six tenets to follow.
1) Thou shalt not consider one intervention a panacea
One size fits all nudges only work on rare occasions. Creating behavioral change at scale requires different techniques for different people, times, and contexts. When looking at the average effectiveness of a solution, one will be misled to either believe the influence is present or absent from all users. This blunder overlooks the nuance that interventions may help specific people while harming others. Teams must experiment with different techniques to determine who they fit best with and when they should be applied.
2) Thou shalt not measure success only by usage
Determining the success of experiments requires a collection of metrics, which don’t just tell whether a user interacted with a feature but what happened because of it. Utilizing basic product analytics such as usage and time on task does not paint a complete picture. Behavioral analytics only tell what happened; understanding why it happened requires organizations also to collect attitudinal data. At a minimum, teams must strive to collect rudimentary behavioral and attitudinal data to determine if a launch was successful. Truly determining if a feature is having the intended impact requires additional outcome-focused metrics. Only when a team is comfortable collecting and utilizing data on the change post-usage can they know if their feature is successful.
3) Thou shalt intervene at the best time and place
Even the most beneficial intervention is only practical when triggered at the right time or place. To promote a behavior, the user must have the right mix of motivation and ability to act. Many characteristics impact these factors, but the two most important are whether the notification occurs where the user can exhibit the action in a straightforward fashion, place, and when there are no other pressing factors, time. Teams must experiment to understand when are ideal times and places and then further tailor based on the individual.
4) Thou shalt trigger and reward to build sustained habits
Sustained behavioral change is the north star we must strive towards, yet, it is also the most challenging. While habit formation occurs based on a series of triggers, cravings, responses, and rewards, the bookends of this process are the most essential. A trigger must be strong enough to cut through the noise of an individual’s daily life yet, not overwhelm or upset them. A reward must matter, but its rapid delivery after an action is even more paramount to forming sustained behavior. The entire trigger and the reward need not be individualized, yet they must be worded to connect with the individual user.
5) Thou shalt ensure interventions are achievable, attractive, and meaningful
Missed or intentionally forgone simple design changes have plagued even the finest interventions from ever having a chance. Whenever an intervention is developed, regardless of the scope, it must undergo a final review to avoid common pitfalls. While it will be up to the team to determine the factors which are most important in their context, the three that appear universal for ensuring that the intervention is successful are:
- Challenging enough to engage but not too difficult.
- Visually captivating and straightforward.
- Valuable enough to pay attention and commit.
These are in addition to the intervention’s timeliness, as shared in Tenet 2
6) Thou shalt work as a team to conquer behavioral change.
Individual behavioral scientists who falsely believe they could do it alone have led many failed attempts. Additionally, many failed efforts originated with teams without behavioral science expertise and lead purely by intuition. All successful efforts share the commonality that they are interdisciplinary and have a project leader focused on developing impactful interventions. Choose to embark on the journey to personalized interventions if the organization is prepared for the investment and alignment required.
Proclamation on AI Accelerations
Once one has committed to The Tenets of Personalized Interventions, they may soon discover their ambitions have passed what is feasible. For most who have reached a point of readiness to embark on the journey, the overwhelming effort leads to half-committed efforts and failure. All hope is not lost, as the generative AI revolution has upended established thinking.
AI accelerations have four proclamations:
1) Unlimited arrangements
The single most costly aspect of tailored interventions has been creating content. If one must create 100 versions of a message, each somewhat changed, depending on a set of characteristics, it might require months of effort. This endeavor is now reduced to less than a day and costs little more than the salary of the person using GPT.
The creation of text is only the beginning. As these systems advance, the ability to rapidly create permutations of design and code will closely follow. One day, entire web pages will be loaded uniquely, containing distinct elements tailored for the individual. The new limiting factor will be data collection to make these tailoring decisions.
2) In the flow data collection
While advances in product analytics have allowed for behavioral data collection at scale, attitudinal data still requires significant human investment. This pivotal data is a bottleneck to truly measuring outcomes, but no more. All forms of user feedback will soon be accessible through AI processing that can create insights from the raw data.
New systems can go out into the world when new data is required and gather data more effortlessly. Before, an individual may have needed to conduct an interview or send a survey; now, a chatbot will reach out to an individual in their flow and ask them timely questions in exchange for further helping them along with their work. The new limiting factor will be collocating and determining what to do with all the data.
3) Informed and accessible teams
Transitioning insights from intriguing to actionable is a time-intensive process requiring an experienced individual. This arduous effort will be made much more straightforward as documents are summarized and combined into digestible formats for all to access within an organization (Dropbox is already doing this). Breaking the silos of information among colleagues will enable a more functioning and collaborative workplace. While the limiting factor of internal politics and structural deficits will persist, data will become more accessible, and with it, more potential for collaboration shall ensue.
4) Experimentation and segmentation
Emboldened by the revolution of generative AI, all SaaS tools will be more effective. Experimentation tools will not just deploy the tests but analyze the results and share them with stakeholders. Chatbots and other conversational agents will conduct studies in a way that feels less like solicitation and more like assistance.
Analytics tools will leverage all these data to begin creating customer clusters. These segments will further the ability to deploy tailored interventions that work on the first or second attempt. As these systems grow more powerful, a feedback loop will follow, increasing the effectiveness of these models while reducing the cost of experimentation.
Here, the limiting factor is our imagination; what would the world look like if teams created interventions for each specific user? What if a system could understand a person's needs to achieve the most desired but unattainable goals? The effort to make this future a reality only recently went from insurmountable to possible.
