Work

About

Work

About

Work

About

Quietly Powerful —An Invisible Ally

Quietly Powerful
—An Invisible Ally

Quietly Powerful —
An Invisible Ally

Overview

Agents were managing complex, multi-step customer requests across fragmented tools - increasing cognitive load and slowing live support.

Agent Assist is an AI-powered workflow automation system within Sprinklr's Case suite that guides backend actions in real time, helping agents resolve complext cases faster while staying focused on customers.

Challenge

Live support felt fragmented. Agents juggled multiple tabs, copied data between tools, and lost track mid-chat — managing multi-step requests manually led to delays, inconsistency, and higher cognitive load.

Admins couldn’t update or style workflows without engineering help, slowing iterations and blocking agility.

Results & Impact

Together, these improvements bridged agent and admin workflows — enabling faster resolutions, higher satisfaction, and fewer errors.

The new self-serve Guided Workflow system empowered admins to independently build, style, and deploy flows — reducing engineering dependency while giving teams full control over customization and reporting.

My Role

  • Led Agent Assist UX — workflows, transitions, and micro-interactions for live AI-assisted support

  • Extended Admin experience and design system cross-functionally

Team: Product Manager, Design Lead, Engineer, Design Systems Manager
Timeline: October to February

Challenge

Live support was fragmented and hard to manage in real time.

Agents lacked progress visilbility, struggled to switch between active workflows, and relied on unclear instructions - driving higher cognitive load, delays, and errors during complex cases.

Administrators couldn’t update or style workflows without engineering help, slowing iterations and blocking agility.

Results

The AI-powered Agent Assist experience reducted friction across agent and admin workflows.

Real-time AI guidance, automation, and structured workflows enabled faster resolution, smoother multitasking, and fewer erros, while a self-serve system lets admins build and deploy AI-assisted flows without engineering support.


How might we streamline complex workflows so support teams can focus on customers — not tools?


How might we streamline complex workflows so support teams can focus on customers — not tools?

Impact

A context-aware AI support system that automates assistance while preserving agent decision-making during live workflows. The system analyzes live customer conversations to surface relevant guidance, response suggestions, and next-best actions—helping agents move faster without interrupting or overriding their workflow.

↑50%

Agent Efficiency

↑32%

Customer Satisfaction

↓40%

Manual Errors

After launch, I tracked how the redesign improved agent workflow efficiency and service quality:

  • +50% Agent Efficiency – Aligned with Sprinklr’s published performance benchmarks.

  • +32% CSAT – Based on early feedback from agents using the new interface.

  • −40% Manual Errors – Reflected in QA logs and stakeholder input.

Note: Efficieny from Sprinklr benchmarks; metrics were developed with Product and Data teams using QA logs,
internal dashboards, and early rollout usage signals.

Research

A Flow That Wasn’t Flowing

Observing where flow broke for both agents and admins. I shadowed real-time chats and admin setups, mapping pauses, searches, and workarounds.

Agents lost focus mid-conversation; admins lost agility in maintenance. Each gap revealed where guidance—not more tools—was needed.

“Even small changes to workflows needed engineering help — it slowed us down every time.”

-Admin

“I’d lose my place mid-chat while switching tabs — it broke my rhythm with the customer.”



-Agent

Insights

Finding Where AI Could Step In- and When It Shouldn't

Through observation, it became clear that not every problem needed a new feature — some needed quiet intelligence. AI could step in where repetition and hesitation lived: auto-filling data, updating tickets, or surfacing next-best actions in real time.

For agents, it meant maintaining focus without breaking conversation flow.
For admins, automation became a way to scale consistency and reduce engineering dependency—without losing creative control.

Opportunity

Turning Insight into Flow

agent opportunity
admin opportunity

Research

A Flow That Wasn’t Flowing

Observing where flow broke for both agents and admins. I shadowed real-time chats and admin setups, mapping every pause, search, and workaround.

Agents lost focus mid-conversation; admins lost agility in maintenance. Each gap revealed where guidance—not more tools—was needed.

“Even small changes to workflows needed engineering help — it slowed us down every time.”

-Admin

“I’d lose my place mid-chat while switching tabs — it broke my rhythm with the customer.”



-Agent

Insights

Finding Where AI Could Step In- and When It Shouldn't

Through observation, it became clear that not every problem needed a new feature — some needed quiet intelligence. AI could step in where repetition and hesitation lived: auto-filling data, updating tickets, or surfacing next-best actions in real time.

For agents, it meant maintaining focus without breaking conversation flow.
For admins, automation became a way to scale consistency and reduce engineering dependency—without losing creative control.

Opportunity

Turning Insight into Flow

agent opportunity
agent opportunity
admin opportunity
admin opportunity

Wireframes

Designing Guidance That Felt Like Second Nature

Designing for two roles meant balancing precision with empathy. Through quick sketches and mid-fi explorations, I tested how guidance could feel intuitive for agents and flexible for admins. Each iteration brought more clarity and ease to both experiences.

Agent

Admin

Testing

Bringing Calm to Complexity

With limited time before handoff, I focused on internal design reviews and motion refinements to ensure clarity and flow.

Progress indicators, collapsible modals, and simplified copy were tested internally to help agents stay focused and help admins edit workflows with confidence.

I also outlined key usability testing goals for future iterations — including cognitive load, completion time, and overall clarity — to guide validation once live data becomes available.

Layering Emotional Intelligence Into Agent Assist

As the system became smoother and more intuitive, the next challenge was to make it more emotionally aware — helping AI understand not just what was said, but how it was said.

I designed the foundation for this sentiment and compliance system, which was later integrated into Sprinklr’s AI Care platform.

View live feature on Sprinklr here and here!

Reflection

This project pushed me to think beyond just clean UI. It was about designing calm into complexity—making a product feel less overwhelming for agents who live in it every day. I learned how accessibility, microcopy, and systems thinking can quietly shift how people work and feel.

Next Steps

  • Run deeper usability testing to catch the invisible friction points.

  • Build a UX writing style guide to scale clarity across workflows.

  • Partner with engineering to bring advanced accessibility features like screen reader support into the product.

Key Takeaways

  • Accessibility reshaped the experience.
    Small tweaks in contrast and sizing made workflows feel calmer and faster for agents.

  • Words outperformed features.
    Clear, actionable microcopy unlocked more ease than heavy UI changes ever could.

  • System thinking > screen thinking.
    Designing for both admins and agents pushed me to build for scale, not just aesthetics.

Key Design Decisions

Designing with Intent

These decisions shaped how Agent Assist delivered automation while remaining reliable and unobtrusive during live customer conversations.

Constraints and Considerations

Designing Within Real-World Constraints

Agent Assist was designed to operate within enterprise constraints while remaining flexible and scalable.

  1. System Performance

    Real-time assistance had to operate without introducing latency during live conversations.

  1. Operational Flexibility

Admins needed the ability to create and update workflows without engineering dependency.

  1. Agent Recognition and Trust

Interactions had to remain subtle and predictable to avoid disrupting agent focus or relearning.

Wireframes

Designing Guidance That Felt Like Second Nature

Designing for two roles meant balancing precision with empathy. Through quick sketches and mid-fi explorations, I tested how guidance could feel intuitive for agents and flexible for admins. Each iteration brought more clarity and ease to both experiences.

Agent

Admin

Solution

A context-aware AI support system that automates assistance while preserving agent decision-making during live workflows. The system analyzes live customer conversations to surface relevant guidance, response suggestions, and next-best actions—helping agents move faster without interrupting or overriding their workflow.

Agent Facing -
AI-powered Guided Workflows

Real-time AI guidance turned into structured, step-by-step workflows—enabling agents to complete complex tasks directly within the conversation.

Administrator Facing -
Workflow Builder

A new no-code builder that lets admins independently design, style, and deploy workflows within Sprinklr- reducing engineering dependency while supporting faster iteration and customization.

After launch, I tracked how the redesign improved agent workflow efficiency and service quality:

  • +50% Agent Efficiency – Aligned with Sprinklr’s published performance benchmarks.

  • +32% CSAT – Based on early feedback from agents using the new interface.

  • −40% Manual Errors – Reflected in QA logs and stakeholder input.

Note: Efficieny from Sprinklr benchmarks; metrics were developed with Product and Data teams using QA logs,
internal dashboards, and early rollout usage signals.

Designing for Tone, Trust, and Compliance

As AI Agent matured, the focus expanded beyond task completion and how agents communicated during conversations. Agents needed support that preserved tone, trust, and compliance- without breaking their flow.

I partnered with product and engineering to shape real-time sentiment and compliance guidance, defining when and how feedback surfaced as agents typed. This work was later integrated into the newly called Sprinklr's AI Service platfom.

View live feature on Sprinklr here and here!

Reflection

This project pushed me to think beyond just clean UI. It was about designing calm into complexity—making a product feel less overwhelming for agents who live in it every day. I learned how accessibility, microcopy, and systems thinking can quietly shift how people work and feel.

Key Takeaways

  • Accessibility reshaped the experience.
    Small tweaks in contrast and sizing made workflows feel calmer and faster for agents.

  • Words outperformed features.
    Clear, actionable microcopy unlocked more ease than heavy UI changes ever could.

  • System thinking > screen thinking.
    Designing for both admins and agents pushed me to build for scale, not just aesthetics.

Next Steps

  • Run deeper usability testing to catch the invisible friction points.

  • Build a UX writing style guide to scale clarity across workflows.

  • Partner with engineering to bring advanced accessibility features like screen reader support into the product.

LinkedIn

Mail

I’m just an orbit away — let’s shape what’s next!
Open to full-time roles, new ideas, collaborations, and connections 🧋☺️

© 2025. Designed by Anuri Shah