®2024

AI Chatbot : Context, Personality and Tone

The vision is to design a GenAI-powered chatbot for Salesforce that delivers smooth, conversational interactions while ensuring quick, accurate results. By enabling natural, context-aware dialogue, the chatbot reduces friction. It makes users feel supported and understood, while its focus on speed ensures instant access to the right information, solutions, or workflows.

Client

SALESFORCE

DELIVERABLES

RESEARCH
PROTOTYPES

Year

2024

Role

RESEARCH DESIGN

Chatbots have become redundant

Including the Einstein Chatbot on the Salesforce website. All it can do is connect you to an Agent, which you probably could've done yourself.

No Contextual Awareness
Most chatbots can't remember context , so they give generic answers or repeat things. It almost feels like you’re talking to a wall.
They lack personality.
Most bots feel robotic or overly scripted, which makes the experience dull and transactional rather than engaging.
They fail at multi-step tasks.
Great at answering a single question, but terrible at carrying a conversation across goals (e.g., browsing, comparing, deciding).

But how did it reach this point?

No Contextual Awareness
Most chatbots can't remember context , so they give generic answers or repeat things. It almost feels like you’re talking to a wall.
They lack personality.
Most bots feel robotic or overly scripted, which makes the experience dull and transactional rather than engaging.
They fail at multi-step tasks.
Great at answering a single question, but terrible at carrying a conversation across goals (e.g., browsing, comparing, deciding).

Also…

With AI taking over the most automated tasks , it just didn't make sense for Salesforce to stick to a script based Chatbot anymore. Hence we decided to plug in AI but we had to do it meaningfully.

Focus of this
Project

Exploring and designing chatbot experiences that leverage GenAI, specifically looking at how these bots can adapt to user needs, provide meaningful interactions

Research
Design
Prototype
Contextual inquiries

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

Salesforce sponsor

What Salesforce asked of us

Losing track of
conversations →

Scrolling endlessly through lengthy chats without having anyway to look up old information.

No Visual
Hierarchy ↑

Hard to scan and find information or products when responses are long blocks of plain text.

Expert
Assistance

Frequently escalates to an agent instead of resolving the query directly or offering in chat alternatives.

Insufficient
Information

Struggles with one-word queries due to limited context. The chatbot cannot interpret one word queries and will reply with a long text of information

Issues with the Salesforce Chatbot

Our take on these Issues

Chat
Timeline
A timline that lets you revisit old conversations.
Users can keep track of chatbot conversation based on topics of entry.
Through this they are able to jump over to specific sections to find the relevant responses they need.
Visual Clarity
for text
Improved visual hierarchy for a smoother experience.

Simplifying product discovery through the use of CTA's like 'compare' .

Summarizing key product features into scanable points.
Clarifying
Questions
The new mechanism focuses on having a natural conversation.
Asking clarifying questions to guide them to provide elaborate entries.
No forced redirection to support anymore.
Suggestive
Prompts
Users often struggle to structure potential quests in chat.
It helps the user form their queries quicker without breaking their flow.
Makes the user feel heard and gives them a sense of continuity.

Protocols and Insights from probes/interviews etc.
Insights, sketches and notes!
Meet the team
Solving issues related to inactivity

Lots of
User testing!

Multiple bouts of user testing using Chat-GPT and Amazon's Rufus as our primary tool.

20+
Contextual
Inquiries

These helped us identify the gaps in current tech. and understand user expectations in context of a chatbot in todays age.

4+ Feedback
Sessions

The feedback and critiques we received helped us iterate, clarify our direction, and develop the final solution.

10+ Probes

As part of our user research, we leveraged ChatGPT to explore the impact of tone and personality on user interactions.

How we got here

Reflections

What could've been
done better

Deeper research on both personality of the chatbot as well as testing concepts with a larger sample.

Future Scope

Develop the timeline concept further to enhance the experience for customers and integrate media elements for richer engagement.

What we learned

Communicating design ideas to stakeholders. Being able to maintain a healthy group dynamic in a large group