Five tips for building data-fueled personalized experiences.
Personalization is a vague, wide-reaching concept that’s often thrown about causally at industry conferences and executive meetings — typically in the company of AI and Machine Learning. Everyone looks for ways to create personal experiences that cater to users while helping the bottom line. Here are five tips for building data-fueled personalized experiences to consider before kicking off your personalization project.
1. Make sure you’re solving the right problem
Look beyond the symptom presenting itself and try to understand the cause of whatever it is you’re trying to improve, whether that’s conversion, repeat visits, account creation or something else. Look at industry research, customer feedback and perform user research to identify the problems — and there will be multiple problems — that need solving. Be sure to include key stakeholders during the problem framing process as well. This ensures a shared understanding of the problem and makes it easier to gain alignment across the organization when you’re ready to start making changes. An excellent way to begin the process is to hold a design thinking workshop. Pull together a cross-functional team, perform user interviews and start the process of framing problems and understanding your users’ needs and motivations.
2. Data Science, Business Analysis & User Research teams work better together
Form a group that includes Data Scientist, Business Analyst and UX professionals to get a holistic view of the problem. User journey maps and customer interviews provide much-needed context to Data Science teams before they jump into building models. Understanding how Data Science works and where recommendations or another relevancy tactic might impact the customer journey enables designers to create better experiences. Having the business analyst in the room means both teams gain a deeper understanding of how success is currently — or can be — measured. Plus you can ensure that the solutions you build can be appropriately measured when it’s time to perform tests. Finally having everyone in the room means everyone is working toward the same objective and you are more likely to have a single holistic view and insight in the end. For example, when I first began working on personalization, I discovered that less than half of our users realized the content on our app homepage was personalized. If UX and Data Science had been collaborating it would have been apparent that the labels we assigned to the content areas on the app homepage didn’t let the user know why we were showing it to them. A simple change of text improved engagement and conversion before we even got started with making material changes.
3. Data Science isn’t an “Easy Button”
Personalization covers a wide range of activities. How you put it into practice is highly dependent on context, where the user is in their journey and the specific need you’re trying to assist within that given moment. That means there isn’t going to be one model that’s going to solve all your problems. I’m almost sure that you’re not going to have perfect data for every user in every situation, so assuming you have the data to provide the best possible recommendation at all times is nearly impossible. It’s also beside the point. Instead, try to identify what the user needs or wants and build a model to help them at that moment in the customer journey. For example, if the user is browsing for inspiration, create a model that emphasizes serendipitous recommendations to surprise and delight. If the user is on a journey where relevancy and ease of use are more important, make sure your model emphasizes recommendations or design elements that prioritize past behavior.
4. Understand what the algorithm is doing
As the product manager, you don’t need to understand how the math works in your model or all of the granular technical nuances, but you do need to know conceptually how it works. Are you using a supervised or unsupervised machine learning model? Do you know the difference between the two? If you’re doing item classification, understand how your particular model makes decisions. If you don’t know the benefits and limitations of your model, you won’t be able to match the right model effectively to the problem you’re trying to solve. It’s also crucial that you be able to explain to your stakeholders what the model is doing — in layman’s terms — so they can be confident and support the process. Invest some time in understanding all the tools available to you in the Data Scientist’s toolkit. You’ll win major points with your data teams, and be a more effective partner to the business when it comes to leveraging ML and AI to create value for your customers.
5. Leverage UX research throughout the process as you iterate
After a release do another round of user research to understand if you’re impacting the customer journey in the way you intended. (Yes even if the data looks good!) It’s important to know if the change you made helped the user move through that moment in the customer journey. There are so many variables that can impact the performance of your experiments, so I like to conduct an additional round of user interviews with users in both the test and control experiences as another data point to validate the results I’m seeing. This user research is meant to supplement, not replace the actual A/B test analysis and provides another opportunity to refine the experience or provide insight into the next problem area where you should focus your efforts. You have to know before you begin making changes how you’re going to measure success. Make sure you’ve identified your primary KPI and secondary KPIs and any health metrics you should keep an eye on to make sure you’re not harming the overall experience.
Personalized experiences that go awry at best have little impact or meaning to your customers or your bottom line, at worst they might frustrate users or distract them from completing the task at hand. By assembling the right team and taking the time to understand how personalization addresses your customers’ needs and then aligning your design and machine learning models to work together cohesively across critical moments in the user journey, you’ll create relevant experiences that both your users and executives will appreciate.
