BigTime AI Assistant

Senior UX Director   |  Made at BigTime, 2024-25

The First Step Into AI at BigTime

In late 2023, I started to lead UX strategy and customer research for BigTime’s first AI-powered product: the BigTime AI Assistant. This initiative marked an important milestone for the company, not only was it our first AI integration, but it laid the foundation for a more intelligent, scalable, and customer-empowered platform.

The AI assistant began as a rough prototype from the team's Hackathon, which was created in a short sprint before I joined. My task was to help transform it into a production-ready, customer-centered support feature. This project was not met without challenges, including those around personalization, backend architecture, knowledge base content management, and trust-building between teams.

The Team:
UI/UX Designers: Emily Lee, Jan Starzyk
Engineering: Alan Kern, Josh Reed, Will Thomas, Angel Rodriguez
PMs: Kevin Kim, Steven Fernicola
Product Education & Research: Brian O’Donnell

The Problem

Before the AI Assistant, customers relied on weekday-only live chat or manually navigating an extensive Knowledge Base. Both approaches were time-consuming and lacked immediacy when customers just needed quick answers.

Customer Goals

• Get instant answers (24/7)

• Minimize time spent searching or waiting

• Discover product features through natural interaction

• Fast resolution to troubleshooting

Business Goals

• Lower support ticket volume

• Improve time-to-resolution

• Empower support specialists to focus on high-impact work
  

• Reduce time troubleshooting foundational questions

Designing a Smart Assistant

We created a conversational AI assistant trained on BigTime’s public Knowledge Base, API documentation, Data Dictionaries, and XML schemas. Customers could now get instant support around the clock, directly inside the product. 

To ensure accuracy, I partnered with our Customer Success, Solutions Consulting, Implementation, and Product team to build a robust question bank. This training set became essential for tuning responses, benchmarking performance, testing accuracy, and identifying areas where content needed to evolve.

Core Challenges

Building your first AI product can come with challenges, and the BigTime AI Assistant was no exception. Here are some of the challenges we faced as we dialed up the release of this feature:

Lack of Personalization

In our first iteration for experimentation, the AI assistant is not firm-specific. It delivered generalized responses, which we addressed by curating prompts and clarifying scope through UX copy and microinteractions. We got early customer feedback about this gap, and it informed our future roadmap.

QA Testing & Accuracy

Testing accuracy with prompt based inputs is very organic and wildly unpredictable. We worked to develop internal questions that we can use and evolve to maintain accuracy of responses with each release.

Presentation & Formatting

We needed responses that felt trustworthy, scannable, and well-structured. I worked with the front-end and UX team to create UX patterns for response layout and quick iteration.

A New Team Tackling AI for the First Time

This was our first feature developed for AI as a company. I helped bridge unfamiliar territory and vision by facilitating cross-functional workshops on AI behavior, conversation flows, and setting user expectation frameworks. 

Latency Times

Customers expect instant answers, so we focused on minimizing delays. This required tight and constant coordination between design and engineering to ensure the interface signaled progress without feeling sluggish across 3P dependent endpoints.

Ethics and Privacy

Our first iteration of the Assistant did not use firm specific data, but we developed the architectural foundation for future use cases. We defined ethical and practical use of the AI Assistant to ensure that data is kept secure, and updated our Privacy Policies.

Real Impact

The AI Assistant went GA in 2024, after 6 months of early access and weekly iterations. The assistant now deflects 95% of incoming support tickets and considerable savings for the company. This empowers our support staff to tackle deeper, more strategic issues. We measure weekly usage, return visits, and top conversation themes to guide future improvements and content investments.

95% 

Support Tickets
Deflected

Accellerating Internal Expertise & Iterative Improvements

The AI Assistant has been a huge benfit for our customers, as well as our internal employees.
Here are some ways that we are embracing this new feature across the company:

Used Daily for Onboarding & Training
The AI assistant is used across the organization on a daily basis, as a tool to help resolve Customer Support questions, or for general training for employees on feature functionality.

Ongoing QA
We monitor feedback from customers and our Customer Support teams to track and repair Knowledge Base articles that may be outdated or is missing key details.

Iterative Improvements
After the launch, improvements were made to improve vector database scraping automation, reducing latency times, and improving response accuracy.

Evolving the Help Center

Following the AI Assistant’s success, I designed and deployed Help Center udpates with Pendo. We brought all self-service support tools into one view: BigTime AI Assistant, the Knowledge Base, Releases & Roadmap, and BigTime Academy, Webinars, and Blog into a unified help center experience. Live support is still accessible at the bottom of the help center, but this interaction model encourages self-service resolution first.

The Help Center architecture is powered by a blended stack of native development and Pendo’s no-code Resource Center, providing the flexibility we need to evolve fast with quick iteration.