Productionising an AI Support Assistant That Learns From Your Team

Company

Major US Sports Broadcaster

Headquarters

New York, US

Industry

Broadcast media

Discover how we helped this customer build a self-learning AI assistant - reducing time-to-first-response by 98% and cutting time spent resolving each issue by 24.8%.

The Customer

A major U.S. sports broadcaster, delivering real-time stats, news, and live streams to 50 million visitors per month, operates in an unforgiving technical environment. Even minor platform issues can result in immediate user disruption, reputational damage, and lost revenue. High performance and rapid problem resolution are critical to business continuity.

The Challenge

The client’s internal dev-support team received hundreds of bug reports daily. Despite having strong engineering talent, response times were slow, knowledge was siloed, and triage was mostly manual.

They’d built a prototype Slackbot named NORA, powered by a large language model, to assist with support queries. Early results were promising - NORA could answer basic questions and reduce the burden on support engineers. But without production-level infrastructure, she couldn’t scale beyond a small beta group.

Our Solution

We helped the client take NORA from a working demo to a fully operational support assistant-integrated into the team’s daily workflow and capable of scaling with demand.

Our five-week engagement focused on three core areas:

Knowledge acceleration – allowing NORA to learn automatically from real engineering conversations
Infrastructure readiness – upgrading the system for stability, scale, and live usage
Operational fit – aligning security, feedback, and integration with existing engineering practices

The result: a self-learning AI co-pilot that improved resolution times, reduced internal escalations, and embedded knowledge sharing into daily operations.

The Results

By the end of the project, NORA had been deployed to the wider engineering team-moving beyond beta to full production use. Key outcomes included:

Response time dropped from an average of 64 minutes to just 37 seconds
Time-to-resolution improved by 24.8%, thanks to faster triage and direct answers to recurring technical questions
Support load was reduced, with NORA handling a growing share of internal queries automatically
Escalations to senior engineers decreased, freeing them to focus on higher-value work
New hires can now be on-boarded faster, with NORA serving as a first-stop resource for platform knowledge

NORA didn’t replace human support-but she made it faster and more consistent.

AI That Learns From the Team

We redesigned NORA’s training workflow to allow her to learn from Slack, Confluence, and GitHub, in near real time. When subject matter experts answered questions, NORA learned from those exchanges. Connecting questions with solutions and continuously expanding her capabilities.

Scaled for Production

We introduced a lightweight evaluation framework to assess the quality of NORA’s responses and the capacity of the system. This helped us identify technical limits, implement performance optimisations, and roll out to a broader user base without compromising reliability.

A Collaborative Project

This wasn’t just a technical challenge - it was a team effort.

We worked side-by-side with the client’s developers, product managers, and platform leads. Together, we ran deep-dive workshops and code reviews to improve:

System architecture and LLM integration
Model training workflows and data sources
Deployment, security, and scalability

The result was a shared backlog, a unified team, and real momentum by week one.

Why It Matters

For business leaders, the key takeaway is simple: well-implemented AI doesn’t just automate, it unlocks capacity.

By using AI to streamline support, this client improved operational resilience, reduced time-to-resolution, and extended the reach of its engineering knowledge-without additional headcount.

That’s not a chatbot. That’s a strategic shift in how technical teams operate.