SALESFORCE × HCI/D STUDIO · 2024 · 9-person team · 4 months

Einstein Chatbot

We turned Salesforce's chatbot from a sales funnel in a costume into a product advisor with a personality: one that asks before answering, reads hesitation, and never forces a handoff.

Sponsor: Salesforce16 personality interviews3 archetypes tested4 features shipped to hi-fiMy lane: research + interaction design
01 · The Problem

Einstein wasn't a generative AI chatbot. It just looked like one.

When we started, Salesforce's chatbot could produce simple responses and push users to contact sales, and that was about it. No memory, no personality, no product discovery. Four failures kept showing up.

Navigation

Endless scrolling with no way to revisit earlier parts of the conversation. Users lost their place and gave up.

The sales push

After most responses, the chatbot pushed users to contact sales before they had enough information.

Tone

Generic, repetitive phrases made the experience feel robotic. No distinctive personality or human-like tone.

Discovery

Product suggestions were vague and generic. No links, no comparisons, no way to make progress on a decision.

"How does a little window like a chatbot get people to be excited about this experience?"

SALESFORCE SPONSOR · THE BRIEF IN ONE QUESTION
02 · The Research

One study asked how it should talk. The other asked what it should actually do.

Personality research and context research ran in parallel: 16 interviews with three chatbot personas on one side, contextual inquiries across ChatGPT, Amazon Rufus, and Einstein on the other. Open either one for what it found.

We tested three personality archetypes (playful, formal, empathetic) across 16 user interviews using ChatGPT. Participants prioritized content relevance over personality in product browsing, but wanted empathy in troubleshooting and opinions during discovery. The right tone isn't one style. It's knowing when to switch.

Playful won for its salesperson-like connection. Formal gave the most structured answers. Empathetic won troubleshooting. The finding wasn't a winner, it was a schedule.

We ran contextual inquiries across ChatGPT, Amazon Rufus, and Salesforce Einstein to understand how people navigate chatbot conversations. Five pain points kept surfacing: threads and journeys, prompts, reliability and transparency, personal information handling, and length of entries.

Users struggled to find previously discussed products, couldn't start new threads intuitively, and felt uncertain whether the chatbot understood short entries. They needed navigation, not just answers.

03 · The Character

The research collapsed into a dolphin named Fin.

The personality research told us how the chatbot should talk. The context research told us what it needed to do. They merged into Fin, a curious dolphin with a schedule for its tone, and three principles that settled every feature argument after.

Principle 01

Assist, don't redirect.

Connecting to sales is a feature, not a fallback. It appears only after repeated failed attempts.

Principle 02

Adapt the tone to the moment.

Product browsing gets opinions. Troubleshooting gets empathy and brevity.

Principle 03

Make the information findable.

Every response should be locatable, comparable, and saveable without scrolling the entire thread.

The character sheet

Fin is curious, empathetic, informative, and engaging. They use human-like language, provide structured responses, take initiative with suggestions, and adapt their tone to the situation: warm and approachable, but always professional. Playful in product browsing, brief and empathetic in troubleshooting, exactly on the schedule the research found.

04 · The System

Every feature either finds the product or reads the room.

Half the system helps you find and judge the right product. The other half watches how the conversation is going and adapts. Together the four features make one discovery arc. Click through them.

The conversation gets a map

A timeline alongside the chatbox generates clickable headings for each section of the conversation: user needs, product suggestions, comparisons. Clicking a heading jumps directly to that part. We tested two designs: a dark tab (hard to find without onboarding) and a skeleton timeline with a clock icon. The skeleton won, but users clicked instead of hovered, so hover-to-expand became click-to-expand.

05 · The Proof

Nothing came out of testing unchanged.

We put the mid-fidelity wireframes in front of four business professionals: timeline, triggers, and comparison flow. Every finding below forced a change. Open one for what we did about it.

Users' first instinct was to click, not hover. The hover-to-expand was accidentally triggered repeatedly. We switched to click-to-expand.

Information was too tightly packed. Users wanted more filter options, customizable categories, and clearer headings.

Labels like 'Message' confused users. Did the chatbot receive a message, or is this a messaging feature? We rewrote every label.

Some users found unsolicited reassurance unnecessary. The final design triggers it only after sustained backspacing, not brief edits.

06 · What I Took From It

What four months with Salesforce taught a team of nine.

Presenting to Salesforce sponsors every few weeks forced us to articulate design rationale clearly. The ability to defend a decision without getting defensive.

Our instinct was to make the chatbot more proactive. The research kept telling us the opposite. Users wanted to explore independently, with the chatbot as a resource they control rather than a guide that steers.

It's still nascent. Deeper research, media integration, returning customer flows, and input from actual sales agents would all refine the experience further.