AI Conversation Summary
AI tool to auto-summarise customer calls so Optus agents no longer have to take notes manually.
What is it?
We crafted a 3-year AI strategy with Google to completely reimagine the contact centre experience at Optus.
Conversation Summary was our first step: an AI tool that automatically generates a summary of every customer call, removing the need for experts to manually scribble notes frantically and could focus entirely on the person on the call.
My contribution
Led contextual inquiry research in Bangalore
Coming up with a 3-year strategic plan with the squad
Collaborated with data scientists, engineers and legal team on AI accuracy and data privacy
Company
Optus
Role
Product design lead
Time frame
Nov 2022 - Jun 2023
Contact centre reality
Frontline experts juggle a lot at once:
authenticate customers
listen and problem-solve
detect sentiment
upsell
and record detailed notes
On average, they spend 80 seconds writing notes per call, which distracts them from the conversation and costs the business $1M a year in inefficiencies.
We knew AI could help… but we didn't know how yet.
The opportunity?
Because the opportunity wasn’t obvious, we started with research.
Our team travelled to Bengaluru to sit beside voice, messaging and retail experts and observe how they actually work. Before the trip, I collected assumptions, planned contextual inquiries and interviews, and set up daily huddles to synthesise insights as they emerged. This research helped us see the full picture of their workflows, pain points, and their expectation of AI.
Synthesise & ideate
We collected A LOT of insights! I synthesised observations to find common patterns across Voice, Messaging and Retail experts on a daily basis.
From this, I used a Venn diagram to capture shared pain points across all three channels and ran an ideation workshop with the cross-functional team. We dot-voted on concepts and explored how AI could help at each moment of the agent journey.
Building the AI strategy
Using the ideas generated in workshops, I mapped out a future-state journey and storyboard to show what an AI-powered contact centre could look like end-to-end.
This helps to visualise the north star with the wider business, share a clear vision, and gain buy-in for the multi-year roadmap.
What to build first
We couldn't launch everything at once! So I guided the team through an Impact vs Effort prioritisation, which helped us land on Conversation Summary as our first initiative:
High impact: note-taking was one of the biggest time sinks for experts
Low effort: Google’s AI summarisation tech was already available to us through our partnership
We've seen a demo of the Google Agent Assist model, and with that I sketched some concepts of the CXOne platform to run concept testings with the frontline experts to see if they're as excited as we are with the concept!
AI Design System
After rounds of testing and getting feedback, we started considering the detailed UI. The existing design system didn’t support contact centre workflows, especially because experts frequently minimise or resize windows.I worked with the Senior Product Designer to create and refine the AI Design UI Kit, adding new patterns, behaviours and breakpoints tailored for expert workflows. This fed directly into the broader design system.
A successful launch
Conversation Summary was a success with note-taking time reduced from 80 seconds to 0 — experts no longer need to write notes; with $1M cost-saving.
But more importantly, experts told us they could finally focus on the customer, not the admin work.
The work goes on…
We continuously improve on Conversation Summary by sending out survey to frontline experts and running focus group every month for feedback around usability and AI summary performance. Product manager, engineers and I would compare the AI generated article to the frontline experts' manual notes, to identify the gap and share feedback with Google to improve the model.
My learnings 🩵
In-person observation is irreplaceable. Bangalore gave us depth, but I supplemented it with contextual inquiry in Sydney to validate behaviours locally.
Keep the big vision in mind, but ship small. Many great ideas went into the backlog, but we continued assessing them based on cost and user benefit.
