Einstein Article Recommendations is a Salesforce feature that saves agents time and improves accuracy with intelligent article recommendations for cases based on the case information. I worked on developing the onboarding experience and dashboard an admin would set up and manage the feature in their organization. 

What problem is EAR solving?

When agents assist customers with issues, there's often existing solutions or articles that can aid them. This feature utilizes AI to surface articles by proactively identifying key words, phrases and numeric fields in agent-customer communications to provide the agent with additional support. The admin can manage the AI model to test accuracy and refine the settings as needed. 

2019 Internship Project

Based on insights gathered from user persona interviews, I identified pain points and collected user needs. Understanding the priorities and goals of the admin, I designed an onboarding experience for setting up an AI model to recommend articles to agents. Additionally, I developed a dashboard that enables the admin to evaluate the model's effectiveness and adjust suggestion sensitivity.

Setting Up the Feature

When designing the onboarding flow, I focused on setting clear user expectations right from the start. This included outlining the time required, the necessary information from admins, and addressing potential points of hesitation directly on-screen with actionable feedback. 

Features from Insights

Based on insights from admin interviews, I developed a feature that addresses their desire for a zero-consequences testing environment. This allows the admin to thoroughly test the AI model without any real-world impact, ensuring it is properly set up before deploying it organization-wide.

Design Process

Outcome

This is a live feature today that is utlized by hundreds of companies to increase the productivity of customer service agents by streamlining solving problems for their customers. My main addition to this feature was the option for different testing environments that admins can use to ensure the effective

implementation of this feature. This was informed by user interviews I held in partnership with our User Research Intern. This project showed me how vauble the design process is in revealing unknown user needs and informing the designs to reflect those findings.