AMCS Group
Role:
Senior UX/UI Designer
Team size:
1 designer, 1 engineer
Timeline:
6 months
Core tools:
Figma
Overview
The problem
AMCS’s original chatbot design was a conventional support window, and it was parked as priorities shifted. As agentic AI matured, the platform needed an assistant that could answer data-heavy questions and, over time, take action.
The solution
AVA, an agentic assistant in the same corner of the screen. It can handles data-heavy queries via tables. I proposed an expandable window to keep them readable, and a link through to full list screens. Initial roll-out will focus on the customer service workcenter.
Key outcomes
Starting point
Where the original chatbot fell short
Key Observations
Design process
Designing the opening state
The old chatbot offered a menu of questions. AVA opens with a greeting and an empty field, so people can ask for what they need. The screen behind it already shows which customer they're working on, and alot of the data options available.
Design solutions
From a menu to an open question: The old chatbot could only answer a predefined set of questions. AVA opens with a greeting and an empty field, so people can ask for what they need.
Answers that fit the data: Customer service questions often return tables, and in a narrow window they cut off. I proposed letting AVA expand to roughly double its width.
A short answer with a route to the rest: AVA shows the most recent five of 46 work orders and links to the full list, filtered to the customer, instead of paging through everything in chat.
Where it launches: I designed AVA to open from the help menu in the header, which lists Help Center, Chatbot, Keyboard shortcuts and Leave feedback. The proof of concept used a standard floating button instead. Which works better is still under consideration.
Status and next steps
Reviewing the proof of concept
An engineer built a first proof of concept, and I reviewed it against the design. My feedback covered three areas:
Branding: the colours didn't match the platform, so I asked for them to be corrected.
Density: the padding was too generous for answers that contain large tables, so I asked for it to be reduced.
Launcher: the build used a floating button in the bottom-right corner, not the header help menu. I asked whether the help menu would work in practice, in particular what happens when the chat is minimised. That is still under consideration.
Where it goes next
AVA starts with customer service. Routing, dispatch and billing validation are planned. In parallel, I oversaw agentic concepts for the customer portal, including an onboarding agent that guides prospective customers step by step through setup.
What I'd test next
Whether people know what to ask when AVA opens with an empty field, and whether suggestions tied to what it can do would help.
How AVA handles unclear requests and mistakes.
When it hands over to a person.
How people use the expanded window in real work.
I'd test these before the agent starts taking actions on its own.
Reflection
Designing for tables showed that a chat window built for short replies doesn't suit operational data. The bigger lesson was deciding what belongs in the conversation and what belongs on the page it links to.