In Fall 2023, I collaborated with Salesforce's Einstein AI Builders team on a sponsored course project to enhance the user experience of Einstein Prediction Builder. We improved the configuration experience by making crucial information that was previously missing or overlooked visible and easily accessible to users.
September 2022 - December 2022, 4 months
1 Data Analyst, 1 Product Manager, 3 Designers
Figma, FigJam, Miro
UX/UI Design, Heuristic Walkthrough, Wireframe, Prototyping
Einstein Prediction Builder(EPB) is an ai-based, low-code tool that builds predictors to help Salesforce users analyze customer data and make business decisions. It is embedded within the Salesforce software ecosystem, and used by users from different user groups.
I know what is the business problem.
I want to see how this product can help me.
I know how EPB functions.
I want to know how to use it to answer business problems.
I know about EPB very well.
I want to be able to use it both effectively and efficiently.
A significant user drop-off rate of 87% happens within the configuration process of EPB, which is the second phase of its flow.
In this project, we wanted to know why users stop using EPB during this phase and try to reduce the drop-off rate through our design.
In order to understand our users and the problems they face during the configuration process, we conducted different types of user research. This research gave us the chance to emphasize with users to see what their actions, feeling and experiences are.
Based on user research, we identified the user flow of EPB’s configuration process with key steps. Surprisingly, the flow consisted of only four major steps and was very simple and straightforward. Ideally, users could use the guidance and tools within EPB to build the predictor for their business questions without difficulty.
After synthesizing our research results, we found users stop using EPB because its guidance or tools are not able to help users.
“...how might we help users using EPB by providing clear guidance and useful, responsive tools that provide proper feedback and notify errors timely?”
We immersed ourselves into the flow again to discover design opportunities that can improve the performance of the original guidance and tools in EPB. We then prioritized our ideas on their value and feasibility together with the Salesforce design team.
The previous Data Checker in EPB was supposed to help users understand system status clearly, and provide shortcuts and suggestions on each step of configuration. It failed because it was too small and can be easily ignored, and was text-heavy that users can not intuitively understand the system status.
In our final concept, I proposed the idea of visualizing the Data Checker to clearly show users the status of the system as well as notify them of errors.
Previously, EPB has a navigation bar, instructions on each operation, and also a data checker to guide low ai-maturity users to understand and use it. Although these guidance are helpful, they were not well-organized and users sometimes feel lost in this information.
Therefore, we analyzed information on the current EPB interface and explored different ways of showing them through wireframes.
We showed our prototype in Figma to the Salesforce design team and gained useful feedback from them.
Several iterations were made based on their feedback.
Feedback
“What if users do not know how to use this new Data Checker?”
Iteration Design
We add a “help” with quick guidance to help users understand how to use Data Checker to help them configure their data.
Feedback
“I can not understand these red color at first glance. I though they all means error, which is confusing.”
Iteration Design
Use orange color to represent error status to distinguish from red color which means “No” example records.
Feedback
“Users may still do not know how to deal with the error.”
Iteration Design
Provide suggestions and shortcuts to the error page to help users recover from configuration errors.
We conducted A/B testing with 20 participants who do not have much experience with ai-based tools. They were asked to use EPB to build a predictor of their own, and solve the error along the way. They were then asked to rate which version of the design they prefer, and we also recorded how often could they notice and recover from the error by themselves, and if they drop-off half-way.
Collaborating with product managers and data scientists posed a challenge due to “language” and expertise differences. To address this, I communicated and check-in with them frequently, sharing my research results and updated concepts. This helped me overcome barriers and receive valuable feedback for a successful project outcome.
Due to project restrictions as a sponsored project by Salesforce, I was unable to interview real users of EPB, which was a departure from previous school projects. Despite this, I utilized stakeholder interviews and user testing with data science students to empathize with users. However, I would welcome the opportunity to interview with real users to understand the obstacles they face directly.
