At-A-Glance
Over last fall, I worked with the AI Platform team at Salesforce to truly enhance the experience of Einstein Prediction Builder, as well as undertaking UX Research. The main goal for my project was to understand the inefficiencies and simplify the existing prediction experience (navigation, data visualization, user flow, and more).
Project Background
What is Einstein Prediction Builder?
Einstein Prediction Builder is the first comprehensive AI for CRM(Customer Relationship Management) - a tool that allows users to create custom predictions for their data by selecting features, training a machine learning model, and setting up actions based on the prediction results.
What is the Problem?
EPB(Einstein Prediction Builder) is a low code tool that abstracts all data and allows admins to quickly create machine learning models, but it’s not easy setting up the models. Through Salesforce’s past research, they’ve found that:
The adoption to EPB is low.
EPB is not serving customers’ needs and its main features haven’t been utilized well.
Models created with Einstein Prediction Builder are often never deployed.
Customers need more help understanding the process of building predictions.
To truly understand how EPB works and what’s causing these problems. I started my journey by researching in the form of heuristic evaluations, unmoderated usability testing, user interviews, and competitive analysis. Based on all the informative research findings, I then came up with design recommendations and implemented them through sketching, designing high-fidelity mockups, and prototyping.
Goals
Streamline prediction building’s process for accessibility and intuitiveness
Increase adoption, utilization, and user acquisition
Provide support and additional resources to enhance customer education
Some Of My Designs
Before
Name & Type
Click to Enlarge
Name & Type
Click to Enlarge
Before
Configure Data
After
Configure Data (Object + Segment)
Click to Enlarge
Click to Enlarge
Before
Review & Build
After
Review & Build
Screen 1. Object
Screen 2. Segment
Click to Enlarge
Click to Enlarge
After
The Process
USER STORIES
What are the use cases of the Prediction Builder?
To understand what actions and decisions EPB is helping users make, I read past research documents from Salesforce and distilled users’ feedback to compose two user stories to have an empathetic understanding of the use cases:
👨💼
Jim is a finance manager using EPB to predict future invoices. By training a predictive model on historical invoice data, he can predict the expected amount of invoices in the future and when they will be paid.
With this information, he can better manage cash flow, make informed financial decisions, and ensure that the organization has enough funds to cover upcoming expenses.
For Numeric Prediction:
For Yes/No(Binary) Prediction:
👩💼
Jenny is a marketing manager using EPB to predict the likelihood of a customer buying one of her products so that she can create targeted marketing campaigns that increase sales. By accessing the predictions in each record of the customer object, she can provide her team with insights that enable them to focus their efforts on the most promising leads.
Having read the research documents, I then conducted a heuristic analysis and ranked all the problems found based on the priority to fix. I also had the chance to present my findings to the PM and engineers.
Heuristic analysis
What are the usability issues existing in the system?
Scoring Methodology
Findings
COMPETITOR ANALYSIS
Discover space for improvements
I’ve also looked at our direct and indirect competitors to find inspiration. The insights that stood out the most were:
Simple and clear user interfaces, accompanied by essential guidelines, are crucial for low-code users to understand the process of setting up a prediction.
Intuitive visualizations of prediction outcomes can help users understand the analytics better and take the necessary actions accordingly.
Help documents and Support services at every stage are essential for users to get the necessary guidance they need.
Products that I conducted competitive analysis with
USER INTERVIEWS
Evaluate hypothesis from research
I drafted a screener survey and recruited participants with data analytics background across channels of 200+. 5 participants were selected from the survey pool to cover a variety of demographics, coding capability, and familiarity with AI prediction tools. I conducted and took notes for all these interviews.
FINDINGS
Strike a balance between optimizing the process and offering resources
A more intuitive and clear progress bar is needed.
The interviews confirmed my assumptions that the side progress bar is unintuitive. When asked which step they were at in the process of setting up predictions, the interviewees had difficulty locating the progress bar on the left-hand side. Moreover, three out of five interviewees mistook the sidebar for a navigation bar that connects to other Salesforce products outside of the Prediction Builder.
More details coming soon!👩💻
A more streamlined workspace is needed for users to better focus on the main workflow.
The Onboarding/Help content is useful for users who want to learn how to build predictions. However, its location is too prominent and causes a disruption in the workflow of setting up a prediction. To make the prediction set-up process smoother, there is a need to separate the support section from the prediction setup section.
Click to Enlarge
Click to Enlarge
Provide error notifications that are easy to find and noticeable.
The current placement of the error reminder in EPB is not intuitive, leading to inefficiencies in workflow and inaccurate predictions. To improve efficiency, a more user-friendly and prominent error reminder is necessary for early detection and correction of errors.
Click to Enlarge
DESIGN
01
Progress bar with built-in error notification
02
Designated area for Help and Support
03
Error notifications that are easy to find and noticeable.
Click to Enlarge
Click to Enlarge
USABILITY TESTING
Evaluating the design with users
To evaluate this prototype, I conducted a moderated task-based usability study with 5 participants spanning data analysts, first-time EPB users, and frequent EPB users. This was followed by an interview where I ask participants to describe their thought processes while performing the tasks. Here are a few highlights from the results:
NEXT STEPS
If I had more time, I would…
🧾 Redesign the Prediction Result page in a clearer and easy-to-understand manner.
📝 Conduct Usability Tests with larger scales of participants of differing experience levels to get more insights into the effectiveness of the design.