All

Resume

Richa Srivastava

All

Resume

Richa Srivastava

Contents

Agentic AI workflow | B2B SaaS

Feature | 0 to 1

Challenge 🎯

HR, IT, and finance teams spend most of their day resolving repetitive tickets that don't need their expertise.

Goal 📔

Design a system where HR admins can define their workflow in instructions, and an agent executes it — freeing up HR admins' time for the complex work that actually needs them.

Results 💹

70% of action tickets auto-resolved.

Let's dive in

Overview

Leena AI

Leena AI is a pioneer in Agentic AI Colleagues for enterprises - voice-enabled, personified AI coworkers prebuilt for IT, HR, Finance, and Procurement. They automate complex workflows across existing systems like Workday, ServiceNow, and SAP.

Agentic AI workflows

Designed both halves of an Agentic HR platform: a no-code builder that lets HR admins describe a process in plain language, watch it decompose into a structured workflow, test it against real data, and launch it live, along with some pre-built AI colleagues — no engineering required ( Part 1 )
and an assistant that executes multi-system requests for employees with full transparency and reversibility. ( Part 2 )

My role

· I led the design end-to-end
· Research - qualitative
· Information architecture
· Interface & interaction design

Timeline

12 weeks ( with design system )

NDA

In compliance with NDA agreements.

Team

· Collaborated with the PM and fellow designers to understand requirements and brainstorm ideas
· Worked with engineers to check feasibility
· Ran a QA review pass before release

Understanding the Problem - part 1

The problem

The back office work is manual and repetitive
HR, IT, and finance spend most of their day on work that doesn’t need their expertise
Eg. HR is spending their day answering “how many leaves do I have left”, “can you send me my payslip”

The back office work is manual and repetitive
HR, IT, and finance spend most of their day on work that doesn’t need their expertise
Eg. HR is spending their day answering “how many leaves do I have left”, “can you send me my payslip”

The problem depth 1 - systems

A single employee touches multiple systems
One request from one employee requires HR to go into Workday then Service Now, then email, then SAP and then back to email
A human being is manually carrying information between systems that don't talk to each other

A single employee touches multiple systems
One request from one employee requires HR to go into Workday then Service Now, then email, then SAP and then back to email
A human being is manually carrying information between systems that don't talk to each other

The problem depth 2 - workload

Tickets pile up and never fully resolve
70% of tickets are resolved by humans doing work that didn't need a human

Tickets pile up and never fully resolve
70% of tickets are resolved by humans doing work that didn't need a human

The problem depth 3 - how chatbot works currently

Existing chatbot only answer questions
They don't do anything
A chatbot could tell about a policy, holiday list etc.
Chatbot couldn't apply the leave and update the workday

The problem depth 4

Building automation requires engineering, which takes too long
By the time it was built something had already changed

Business value and why it matters

Expensive people doing cheap work

The company is paying specialist rates for answers a system could give.
That is a cost problem.

Headcount grows with the company

More tickets means more HR, IT and finance hires
Leadership wants to grow without adding support staff at the same rate

Slow answers cost employee time

Every ticket that sits in a queue blocks someone.
A new joiner without a laptop, a manager waiting on an approval, an employee waiting for a payslip for a loan.
Across thousands of employees, that adds up to a lot of lost hours, and it hurts employee experience.

Automation is too slow to build

If every workflow needs engineering, IT ends up with a backlog and the business can’t adapt when a policy or process changes.
The cost isn’t only the build. It’s also the months of manual work while people wait.

Expensive people doing cheap work

The company is paying specialist rates for answers a system could give.
That is a cost problem.

Headcount grows with the company

More tickets means more HR, IT and finance hires
Leadership wants to grow without adding support staff at the same rate.

Slow answers cost employee time

Every ticket that sits in a queue blocks someone.
A new joiner without a laptop, a manager waiting on an approval, an employee waiting for a payslip for a loan.
Across thousands of employees, that adds up to a lot of lost hours, and it hurts employee experience.

Automation is too slow to build

If every workflow needs engineering, IT ends up with a backlog and the business can’t adapt when a policy or process changes.
The cost isn’t only the build. It’s also the months of manual work while people wait.

About users

Two users: the HR ops person clearing tickets all day, and the HR leader who has to prove ROI.

Research

User interview insights

What an HR ops manager told me, sorted into six themes, from her daily queue to how she wants to edit.

Point of view

One statement per user: who they are, what they need, and the insight behind it.

The idea that hit both users most directly

Four ideas that served both users: guided capture, preview, gradual trust and live proof.

Solution

Prototyped in code to test ideas quickly ( HTML in Claude )

Approach/direction 1

A chat where HR teaches the AI their process.

Dropped: admins didn't want to build the AI agents themselves.

Approach/direction 2

A welcome page showing what the AI can do.

Dropped: it still asked HR to do too much up front.

A welcome page showing what the AI can do.

Dropped: it still asked HR to do too much up front.

Approach/direction 3

Ready-made workflows to review and use.

Kept: it shows value before any setup.

Ready-made processes to review and use.

Kept: it shows value before any setup.

Iterated on approach 3

Six cards became three, then a list and detail split so the admin makes one decision at a time.

To click on their relevant agent workflow card.

User flow - high level

Four steps: land on agents, open one, edit it, make it live.

Structuring & visualising the landing screen

Layout options for the HR, IT and Finance pre-built agents page. The thumbs up marks the one took forward.

Prebuilt workflows agents - User flow 1 - lo-fidelity

The admin picks a ready-made workflow, changes one step by chatting with the AI, and takes it live.

Monitor the agents - User flow 2 - lo-fidelity

The admin monitors the live agent workflows: see what needs attention, open the request, act.

Prebuilt workflows agents - User flow 1 - hi-fi - iteration 1

First full pass: processes, sub-processes, edit with AI, then live. The notes show what I changed.

Con: Below the cards, Create gets pushed out of view as process grow.

User flows

Went back to canvas and created all different possible scenarios on a high level

Prebuilt workflows agents - User flow 1 - hi-fi - final iteration

First visit with nothing connected, so the admin sees what to connect before anything can run.

Decision 1 : Connectors come first, because agents can’t go live without them.

Decision 2 : Renamed “process” to “colleagues” to build trust in the agents.

First visit with nothing connected, so the admin sees what to connect before anything can run.

Decision 1 : Connectors come first, because agents can’t go live without them.

Decision 2 : Renamed “process” to “colleagues” to build trust in the agents.

Business impact

-70%

Reduction in manual HR agent tickets for action-based tasks

10 mins

From plain language description to a live running process

0 code

Required to configure and launch a process

My Learnings

Designing for AI requires communication at every step to build trust
When AI executes consequential actions, transparency isn't optional — every step needs to be visible to the employee so they can trust what's happening and why.

Every interaction matters — even the tiniest ones

In an AI interface, interactions carry the experience. What happens between steps is just as important as the steps themselves.

Every design decision is high stakes — I traced each one back to the employee

This assistant touches payroll, tax, and insurance for thousands of employees. Every decision carried real consequences, so I kept coming back to one question: what does the employee actually need to feel confident here?

Designing for AI requires communication at every step to build trust
When AI executes consequential actions, transparency isn't optional — every step needs to be visible to the employee so they can trust what's happening and why.

Every interaction matters — even the tiniest ones

In an AI interface, interactions carry the experience. What happens between steps is just as important as the steps themselves.

Every design decision is high stakes — I traced each one back to the employee

Interested in
working together?

Let's bring your vision to life

Get in contact with me

Mail

Interested in working together?