AI Product Design | Enterprise | Conversational UX
iPDA
Designed an AI-powered digital assistant for Microsoft Teams, partnering with a product manager and developers to simplify IT support for 22,000+ employees at The Ohio State University Wexner Medical Center.

Overview
I worked on this project as a UX Designer within the Business Technology Solutions unit of The Ohio State University Wexner Medical Center's Information Technology department (OSUWMC IT), which develops internal tools to improve employee productivity and workflows.
The existing IT support process was fragmented across multiple channels and relied on inefficient ticket submission forms, contributing to employee frustration, incomplete ticket data, and longer submission times. The goal was to leverage AI to create a centralized, guided support experience that simplifies how employees navigate IT requests and submit tickets across an organization of 22,000+ employees.
My Duties
UX research
Conversational UX
Interaction design
Wireframing
Prototyping
Design handoff
Duration
11 months (ongoing)
Team
Product Manager
4 Developers
UX Research Intern
Me (Lead UX/UI Designer)
Tools
Figma & FigJam
ChatGPT
Microsoft Suite (Copilot, Adaptive Card Designer, Visio, Power BI)
Why AI?
The business was interested in leveraging AI to improve how employees access IT support
The organization had identified AI as an opportunity to modernize its employee support experience and reduce the friction associated with its existing ServiceNow and call-based process.

A portion of OSUWMC's medical campus that is based in Columbus, Ohio.
The Problem
The existing IT support process was fragmented, complex, and lacked guidance for users
There are several pain points associated with the IT support process that have been identified by staff and business stakeholders over several years. For the scope of this work, we honed in on the current process being fragmented, complex, and lacking guidance.

Lack of guidance

The employee has to enter information that may not be readily accessible.
Employees are often clinical, and may not have the technical knowledge to diagnosis certain IT issues.
There is a lot of required information that tends to not be pertinent to the request at hand.
The Solution
AI-powered assistant within Microsoft Teams to simplify IT support through conversational interactions
Designed in collaboration with a product manager and developers, the AI assistant integrates into Microsoft Teams, leveraging employee context and AI inference to reduce manual input, ask clarifying questions, and guide employees through the IT support process.


Understanding employee support needs and existing workflows
Research methods
Heuristic evaluation of an existing support agent, interviews with staff, and analysis of Microsoft 365 usage data and analytics
I led discovery research across three areas: evaluating an existing OSUWMC AI support agent, conducting 18 of 21 semi-structured interviews with clinical and non-clinical employees, and analyzing Microsoft 365 (M365) usage data. Since Teams was being considered as a potential integration point for the new experience, I examined the usage data to better understand how employees accessed M365 tools and whether Teams implementation could support their workflows.
Participant profiles for semi-structured user interviews
All 21 employees were OSUWMC employees
14 clinical, and 7 non-clinical staff, of varying domains spanning support, administrative, medicine, nursing, lab, and imaging.
What did the Power BI display?

41% of Teams users…
access it via mobile (iOS, Android). Usage on iOS was significantly more dominant.

58% of Teams users…
are actively using it to either call, exchange chat messages, or participate in meetings.

Challenge & Resolution
Teams was already being considered as a potential integration point for the new experience, creating the risk that existing business assumptions could bias the discovery process. To ensure I was not leading participants towards a predetermined solution, I asked broadly where they would expect to access the new experience. For those who did call out Teams, I asked follow-up questions to better understand their reasoning and expectations.
Primary research findings

Research Finding 01
Employees struggled to determine which ticket category applied to their issue.

Reseach Finding 02
Existing forms required employees to enter information that was readily available in organizational systems

Research Finding 03
Clinical employees needed accessible support across demanding, mobile workflows; Teams was already a significant part of employees daily toolset.

Reseach Finding 04
The existing support agent offered limited flexibility in interpreting requests and recovering from errors.

Designing the AI conversation and interaction model
Why a conversational interface?
It could allow employees to use natural language in a guided process
A conversational interface offered an alternative to the current form experience in ServiceNow. Rather than requiring employees to understand IT support categories and navigate busy forms, the new system could leverage natural language, asking clarifying questions and guiding users through the process.
Approach
Focused on four key interaction considerations in the flow
Using the findings from research and conversational UX principles, I mapped out an end-to-end interaction model in Microsoft Visio to define how employees and the AI experience would interact through the ticket submission process. It was also at this time that, in having discussions alongside our developers, I learned about the opportunities that AI inference could provide to streamline the experience for users.
Employees struggled to determine which ticket category applied to their issue

Interaction opportunity
Rather than requiring employees to identify the appropriate ticket category themselves, I proposed having the AI infer the ticket type based on their initial description of the issue.


Rationale
This would reduce the cognitive effort required to submit a request, allowing employees to describe their issue naturally without having to understand the organization’s IT support categories.
Existing forms required employees to enter information that was readily available in organizational systems

Interaction opportunity
I designed the interaction to leverage available employee context, such as their registered building location, rather than asking them to provide those details.


Rationale
Reducing manual data entry would make ticket submission more efficient, while allowing employees to review and correct information inferred by the AI.
Clinical employees needed accessible support across demanding, mobile workflows; Teams was already a significant part of employees' daily toolset

Interaction opportunity
I incorporated an edge case where the AI’s inferred building location might not reflect where an employee is currently having an issue. The workflow allows employees to provide a correction before moving forward.


Rationale
This accounts for the dynamic nature of clinical and operational workflows, helping ensure that support requests contain accurate location information even when employees are not in their usual work location.
The existing support agent offered limited flexibility in interpreting requests and recovering from errors

Interaction opportunity
I accounted for variations in how employees might phrase their requests, incorporating different terminology and descriptions of the same underlying issue into the proposed workflow. I also considered how the AI should respond when it couldn't interpret a request or needed additional information, incorporating opportunities for clarification and correction.


Rationale
Since employees may not always be technical, they should not need to know specific terms or phrase their request in a particular way to successfully interact with the AI. Allowing natural variations in language and clear guidance when the AI is unsure would make the experience more flexible and unforgiving.

Translating the conversational flows into interface wireframes
Approach
Leveraged the Teams SDK to see how AI interactions
With the conversational interaction model established, I translated the proposed workflows into wireframes in Figma, exploring how the AI’s responses and user actions could be presented within Microsoft Teams. This would involve balancing conversational messages with structured, interactive components to make the experience easy to follow and navigate.
Microsoft Teams UI zoomed-out

iPDA would hypothetically be accessed from Teams' side panel.
The chat section, where messages and Adaptive Cards between the user, and AI, would appear.
Balancing conversational messages and interactive cards
I examined when information was best communicated through conversational text as opposed to structured Microsoft “Adaptive Cards” (basically reusable, bite-sized blocks of interactive content).
Conversational messages
I opted to use chat messages for standard conversational text (e.g., acknowledgements, and follow-up questions), adhering to Microsoft Teams SDK best practices and to keep things familiar for Teams users.
Several examples of conversational messages I mocked up that would be later translated to Teams
Adaptive cards
For interactions involving inferred user information, I explored how the AI could communicate its assumptions while giving the user control over the final input. I leveraged Microsoft’s Adaptive Card components to structure the information in an easy to review manner and provide clear options for confirming or correcting details.

"Human-in-the-loop" ticket summary adaptive card example
I made the information to confirm more concise, ensuring seamless review.
I employed what Microsoft refers to as "quick actions" as an alternative to typing for users.

"Welcome" or "first-run adaptive cards" that convey the capabilities of the AI for users' awareness
In this example, I proposed a phrase the user could type to get started
I leveraged Microsoft's "quick action" CTAs as a way for the user to quickly execute certain functions
I explored using multimedia in the card to communicate the AI's functionality in a digestible format.

Bridging design and development through prototyping with AI
Why prototype with AI?
To make the design-dev handoff more seamless and upskill in AI
As the sole designer working alongside several developers, I explored how I could make the design-development handoff more seamless by translating my Figma wireframes into higher fidelity Adaptive Card prototypes with the help of AI. This also presented an opportunity to upskill in incorporating AI meaningfully into my design workflow.
From Figma to Adaptive Card prototypes
I began my refining my wireframes into higher-fidelity designs, using components from the Microsoft 365 UI kit, aligning them with the established typography, spacing, and color patterns.
Wireframe and JSON (code) block

Then, I uploaded the designs into Microsoft Copilot and iteratively prompted it to generate corresponding Adaptive Card JSON schemes. I then brought the schema into Microsoft’s Adaptive Card Designer to recreate the components and test them at various screen sizes.

Concepts created through the Adaptive Card Designer tested at different screen sizes
AI "tool call" portrayal challenge
Developers initially proposed displaying the technical details to employees before the action was performed, referencing security concerns around AI. While I understood the need for this, I was concerned that exposing the technical details could overwhelm employees, making the experience more complex.

Developers' initial vision, with technical details being conveyed to users for review
To address concerns about information overload in the existing form, I proposed hiding unnecessary details from the user interface.
Since users may often not be tech savvy, I was concerned with the CTA copy being too technical
The “Confirm” and “Cancel” button colors did not align with Microsoft’s Figma design guidelines.
Resulting change
Through iterative discussions with the developers and the broader team, we aligned on an Adaptive Card that presented the relevant ticket information in a structured, easier to read format.

Resuting adaptive card after implementing several changes
Simplified the ticket review to essential details (phone, email, building, and room), removing fields that were often irrelevant to employees' requests.
Streamlined the flow by leveraging employee context to prepopulate an Adaptive Card, allowing users to review, modify, submit, or cancel their request. Added a previously missing cancel option for greater user control.

Exploring Opportunities Beyond the Minimum Viable Product
Laying out ideas beyond the MVP
Research revealed a recurring need for greater visibility into ticket statuses, supported by employee interviews, external vendor research, and calls to the hospital's service desk . To address this opportunity beyond the initial MVP, I explored two potential solutions:
Conversational ticket status lookup
I designed a conversational workflow allowing employees to inquire about the status of an existing ticket directly through the AI, extending the experience beyond initial ticket submission.

Ticket status dashboard
Drawing inspiration from the external vendor’s concepts, I designed a centralized ticket-status dashboard adapted to OSUWMC’s needs and Microsoft Teams’ interface constraints, allowing employees to view and track their requests.


Reflection
Next Steps
Implementation is currently in progress with the project nearing pilot testing. Upcoming work includes:
Validating the conversational interactions with employees
Incorporating pilot feedback
Exploring additional IT support use cases

More Case Studies!







