Agentic (AI)-Powered Support, Expert to Expert
Goal
Improve the support experience for TurboTax Product Support Experts by redesigning how Tier 1 experts connect with Tier 2 SMEs. The goal was to reduce friction, automate context sharing, and integrate gen-AI-powered assistance (Agentic) ultimately improving efficiency, deflecting unnecessary escalations, and enhancing customer satisfaction.
My Impact
As the E2E lead designer, I guided the entire UX process, from discovery through launch. I conducted user research (Follow-Me-Homes), mapped escalation flows, defined edge cases, and designed scalable components within our design system. I also partnered closely with engineering and QA to test and validate every interaction, ensuring platform durability and adoption.
Outcomes
40% year-over-year reduction in Tier 1 escalations
24% escalation deflection during early cohort testing
20 minutes saved per deflected escalation
Portfolio type
Case study
Project timeline
Dec 2024 - Mar 2025
Sector
Fintech, Enterprise
Role
E2E lead on Product design, Service design, Design system, User research
Platform
Responsive Web App
Empathize
What is a tier 1 product support expert?
A Tier 1 Product Support Expert handles incoming contacts regarding billing, account, and product or service management issues. They may flex between different products based on customer demand. These experts are hired by Intuit to provide virtual assistance and advice to customers on a range of issues, including login problems and technical glitches. They play a crucial role in ensuring that customers have a positive and efficient tax preparation experience.
What is a tier 2 product support expert?
A Tier 2 Product Support Expert, also known as a Subject Matter Expert (SME), assists Product Experts by monitoring Tier 1’s performance and helping them overcome obstacles with their deep knowledge of product issues. They play a crucial role in supporting Tier 1 experts and maintaining the hierarchical structure of the business’s service delivery.
User Research
To better understand the current flow and experience, and to gain empathy for our experts, I conducted a handful of user research sessions known as Follow Me Homes. It became painfully clear that the experience was far from optimal. The experts have to manually input several pieces of information that the system should already know, leading to significant delays of up to two minutes. This presents a major opportunity to reduce Average Handling Time (AHT) and Customer Service Time (CST), which are primary metrics for evaluating expert performance.
A screenshot of a Follow Me Home being conducted with a TurboTax Product Expert
Omni-chat, a native Salesforce chat
The omni-chat was very archaic and carried zero context with it. It was not plugged into the tier 1 data to know basic knowledge like what site they are located at. Before connecting to a tier 1 it would ask several questions which would delay connecting to a tier 2. There were so many questions that it could take up to 2 minutes for tier 1 to answer all of them.
A demo of the current tier 1 to tier 2 chat escalation feature is Salesforce
Define
Problem statement
I am...
a TurboTax product support expert
I am trying to…
connect to a Tier 2 expert to get unblocked so I can quickly and efficiently help my customer resolve their issues
But…
the chat feature takes too long to connect me with someone
Because…
the system should already know all the context and shouldn't require me to answer questions it should already have
Which makes me feel...
like my efforts are wasted, especially when I end up just waiting for the Tier 2 expert to connect, leaving my customer waiting as well.
Business goal
Tier 1 product experts currently seek assistance from Tier 2 experts when they are stuck or unsure how to help a customer. They connect via omni-chat, a native Salesforce solution, where Tier 2 experts can provide support and handle up to five concurrent chats. To enhance this process, a new capability is required in lEP as part of the CS-LG-09 (Salesforce into IEP) project, ensuring that the Tier 2 chat support feature is fully available by the third peak in October 2024.
Requirements
The Tier 1 capability needs to be seamlessly integrated within Intuit Assist. Before automatically connecting to a Tier 2 expert, Intuit Assist should offer de-escalation with self-help answers from its curated knowledge base via the QnA plug-in. If a Tier 1 expert determines they need further assistance, the Human Handover (HHO) plug-in is activated, connecting them to the next available Tier 2 expert.
For Tier 2 experts, the existing Telephony widget will automatically provide all necessary context, ensuring a smooth transition. Tier 2 experts can then address the questions from Tier 1 and, if needed, join a screen share to view the customer's screen.
Business Requirements Document (BRD)
Ideate
User flows
There are two paths an expert can choose before connecting to a Tier 2:
Ask their question first.
Request a Tier 2 expert first.
Both options require an answer before connecting to a Tier 2. This approach supports our de-escalation strategy, where the Tier 1 expert attempts to resolve the issue first. Other use cases include scenarios where the contact is incorrect or the Tier 1 expert needs to edit the call summarization, along with a few other edit cases.
Example of basic user for for a tier 1 product expert to connect to a tier 2
Tier 1 flow
Ask question first
When a Tier 1 expert is stuck or unable to help the customer, they should go to Intuit Assist for assistance. They can ask their question and review the response. If the answer is insufficient, they can then request a Tier 2 chat. The system recognizes that they have just asked a question and initiates the routing process. Before automatically connecting them, we need to confirm that the current case is the one they need help with. Once confirmed, Intuit Assist will generate a call summary of the conversation so far and verify it with the Tier 1 expert. After confirmation, they will be connected to a Tier 2 expert.
Ask escalation first
Sometimes, a Tier 1 expert will immediately request a Tier 2 chat. However, the system knows they haven’t asked a question yet and will use our de-escalation strategy to encourage the Tier 1 expert to try self-help first. In this case, Intuit Assist will still ask what they can assist the expert with before connecting them to a Tier 2. If the initial response is not sufficient and the expert still needs help, the process will largely follow the steps outlined above.
Tier 2 flow
When a Tier 1 expert confirms they need assistance, the Human Handover (HHO) plug-in is invoked, connecting them to the next available Tier 2 expert. The engagement context is automatically passed on to the Tier 2 expert, ensuring they are fully informed. This allows the Tier 2 expert to treat the Tier 1 expert as their customer, addressing their questions directly. If necessary, the Tier 2 expert can join the same screen-sharing session with the Tier 1 expert and the customer.
Component documentation
Since this is a brand-new experience built for Intuit Expert Portal (IEP), meticulous documentation on its construction and functionality is crucial for a successful build, considering all its different states. Engineering needs to understand each inversion pattern to ensure the component functions correctly. Additionally, we must create this component at the platform level, making it durable and versatile enough for any other use case or archetype within IEP to utilize. The component should also adhere to the established UI patterns in our design system, ensuring it integrates seamlessly into the existing workflow.
Specs documentation to hand off to dev to make pixel perfect
Test
Test with smaller cohort
It was clear that the new design was a better experience, so we launched a small test to 20% of Tier 1 product experts to run water through the pipes. The test results were excellent, with 24% of escalation attempts deflected, reducing the percentage of contacts with escalations from 10.2% to 6.1% (60 ITC). The overall AHT is 99 ITC, and we observed almost 20 minutes of savings when escalations were deflected. Due to this success, we scaled the new experience to 100% of product experts a few weeks later.
Implementation
We will ensure that all development specifications, including edge cases, error states, and interactions, are thoroughly prepared. We’ll use durable components from our design system to avoid custom code or one-off solutions.
While the designs are being coded, the development team will provide updates during our daily scrum meetings. They will share their builds with me to ensure pixel-perfect accuracy. After that, they will provide a pre-production link so I can test the build, trying to break it through various scenarios and edge cases. I will then document any discrepancy and share it with them.
Once we have internal sign-off, the build will move to User Acceptance Testing (UAT). I will help create scenarios for real experts to test the build in both typical and complex situations, ensuring a pass/fail at every step. Any failures will be documented and addressed. Once it passes UAT, the build will go to production.
Happy path to connect a tier 1 Product Expert to a tier 2 Subject Matter Expert
Results
With the integration of gen-AI-powered self-help, call summarization, and attribute-based routing, the Intuit Expert Portal (IEP) successfully launched our Human Intervention Agent. This Tier 1 to Tier 2 chat feature enables Tier 1 experts to seamlessly connect with Tier 2 human expertise. It also offers de-escalation through self-help answers, which can be resolved by Intuit Assist, resulting in an 8% Tier 1 escalation rate—a remarkable 40% reduction year-over-year. This has led to a decrease in both Average Handling Time (AHT) and an improvement in Total Net Promoter Score (tNPS), with an Expert Tooling PRS consistently above 80.
Future
We all know that design is an ongoing process. Once a design is released, we monitor its performance and gather analytics. We then use these insights and metrics to continuously refine and iterate, uncovering new use cases and improving flows and patterns to enhance the user experience.
As we roll out these capabilities, we plan to leverage them across the platform, ensuring that other archetypes can also benefit from the same advantages.