All work

Case study 02 · AI · Enterprise contact centre · SaaS

Helm AI supervisor console

Product design for the supervisor module of an omnichannel contact centre console. One grid to watch every agent on every channel, and a way to step into a live conversation without leaving it.

Role
Product strategist and owner
Built for
Contact centre supervisors
Sector
Enterprise customer service
Platform
SaaS, cloud-based system
Year
2022 to 2023
01

Project Overview

Helm is the agent and supervisor workspace of a contact centre platform. This case study covers the supervisor side. A supervisor looks after a team of agents who handle phone calls, WhatsApp, Messenger, email, SMS and Instagram at the same time. They need to know who is stuck, who is overloaded and which conversation needs a second voice, then act on it quickly.

The design brings that job into one screen. A live grid shows each agent against each channel with a running timer. From any cell the supervisor can open the conversation, message the agent privately, speak to the customer, or join a call as a third party.

02

Platform

  • Desktop
  • Web app
  • Voice and digital channels

The supervisor console is built for wide screens that stay open all day beside other tools. Density and quick scanning mattered more than visual flourish.

03

Challenge

A supervisor watches many agents and many channels at once. Each agent can hold several digital chats while on a call. The problem was to make that load readable at a glance, and then make intervention fast and safe.

Show the state of every agent on every channel without opening anything.

Make delays and breaches visible before the customer complains.

Let the supervisor step in without losing sight of the rest of the team.

Keep the customer experience calm. Coaching an agent and speaking to a customer are different acts and must feel different.

04

Strategy & Approach

I treated the grid as the home of the product and built everything else around it. Three decisions shaped the work.

  • One matrix, not many lists

Agents as rows, channels as columns. The same layout answers who is busy, on which channel, and for how long.

  • Reveal on demand

Hover shows agent ID and the customers behind a timer. Expand All stacks concurrent interactions. Detail stays hidden until asked for.

  • Act in place

Conversations open in a side panel and in small windows over the grid. The supervisor never loses the team view.

05

My Contribution

I owned the supervisor experience from flow definition to the clickable prototype: the structure of the grid, the interaction model for monitoring and intervention, the screen designs, and a prototype of one end to end flow used to walk stakeholders through the product.

  • One merged flow for monitoring, AI assist, chat intervention and voice join
  • Screen design and component states
  • Interactive Figma prototype
06

Scope

The prototype covers one end to end supervisor flow on the Team Performance Stats screen, with AI assist built in.

  • Spot on the grid
  • Understand with AI
  • Coach or speak
  • Join the call
07

User Research/Journey

  • User Interviews

Over the first discovery phase I ran 20 one on one interviews: 9 floor supervisors, 6 team leads, 3 operations managers and 2 quality analysts, across sales and support teams. Each session was about 45 minutes and included a walkthrough of how they monitor a shift today.

Problems are found late, usually when an agent asks for help or a customer complains.

Context is scattered. The timer is in one tool, the transcript in another, the call in a third.

Supervisors want to guide quietly first and only then speak to the customer.

08

User Personas

From the 20 interviews I shaped three personas. The first two drove most of the decisions on the grid and the intervention flows.

  • Hassan Raza, floor supervisor

Watches 10 to 25 agents live across calls and chats. Wants early warning, a fast way into a conversation, and a way to coach without the customer noticing. “By the time an agent asks for help, the customer is already angry.”

  • Ayesha Siddiqui, operations manager

Looks at queue and team totals. Wants to know if the floor is coping and where to move people. “I do not need every chat. I need to know where to send two more people.”

  • Bilal Ahmed, agent

Handles a call and two chats at once. Wants guidance that arrives privately and refers to the exact message he is stuck on. “Tell me quietly. Do not jump into my customer’s chat before I can fix it.”

09

Research Workflow

  • Interviews and shadowing
  • Task and flow mapping
  • Wireframes
  • Prototype and walkthrough

Interviews and shadowing showed how supervisors monitor and intervene today. I then broke the work into four tasks: watch, decide, coach, join. Wireframes tested the agent by channel matrix against other layouts, and the Figma prototype of one end to end flow was walked through with stakeholders and refined.

10

User Roles & Priorities

Floor supervisor: spot delays and step in fast. This shows up as red timers, hover detail, and the chat and call panels.

Operations manager: read the state of the whole team. This shows up in the Team ABC panel with active agents, calls, queue and callbacks.

Agent: get help without losing the customer. This shows up as a private agent chat with a quoted message.

11

Defining The Architecture

The architecture is shallow on purpose. The grid is the root. Every deeper view opens beside it or over it, so one click back is always the grid.

12

Supervisor Journey Map

The journey follows one intervention from the first sign of trouble to resolution: notice, understand, read, coach and step in. Each stage has a need and a design answer.

13

Core flow: Spot, coach and join, with AI assist

The three flows I first designed separately, reading the grid, stepping into a chat and joining a call, are really one job. A supervisor spots a problem, decides how to help, and acts. I merged them into one flow and added AI assist at the points where it saves the most time.

  • Spot
  • Understand
  • Coach
  • Join
  • 1. Spot it on the grid

The grid is the product. Each cell is a timer for the agent’s current interaction on that channel. Timers that pass the threshold turn red, and hovering an agent shows who they are without opening a panel. Linked interactions group a customer’s WhatsApp, email and calls, so the supervisor sees the whole story.

  • 2. Understand it with AI

Before opening anything, the supervisor sees an AI signal on each row, such as a sentiment drop or an escalation risk. Opening the conversation shows a short AI summary and a suggested reply, so the supervisor reads two lines instead of the whole thread. The signal only advises. The supervisor decides, and the agent is never replaced.

  • 3. Coach first, then speak

The supervisor can whisper to the agent in a small window over the grid, with a quoted customer line attached so the guidance has context. Only when needed do they write into the customer thread, where the message carries a Supervisor label and its own colour so everyone sees who is speaking.

  • 4. Join the live call

A red timer in the Voice column leads to the call. The call panel shows the customer, the agent and the supervisor side by side with live waveforms. Pause, mute and end sit on the left, and the transcript stays open on the right so the supervisor keeps reading while they listen.

▶

Interactive prototype

A working feel of the supervisor console. Press play for a guided walkthrough, or click any agent to take over. Try Listen, Whisper and Barge in, and watch the AI assist panel. Timers are live and turn red when a wait gets too long.

Or click any agent row
Helm · Supervisor · Team ABCLive
6Active agents
2Live calls
7In queue
3Callbacks
AgentChannelCustomerWaitingAI signal

14

Usability Testing Based On Heuristic Evaluation

I reviewed the flows against standard usability heuristics. This is where each one shows up in the design.

  • Visibility of system status
  • Recognition over recall
  • User control and freedom
  • Consistency
  • Error prevention
  • Flexibility and efficiency
  • Minimalist design

Visibility of system status comes from live timers, red alerts, the queue count and the availability timer in the left rail. Control and freedom comes from every panel closing back to the grid. Consistency comes from using the same chat component in the side panel and the agent window. Error prevention comes from labelling supervisor messages with their own colour.

  • A/B Five Minute Testing

I tested the prototype with 8 of the supervisors and team leads from the interviews, in two rounds. Round one was a five second test of the grid. Round two was a click test of the merged flow with task prompts such as “find the agent who has waited longest on WhatsApp” and “help this agent without the customer seeing it”.

  • Conclusion & Learnings

Most people found the delayed agent within the first few seconds, so the red timer rule stayed. Early versions let the supervisor type into the customer thread by default. Testers wanted a private route first, so the agent window with a quoted message was added. The Supervisor label and colour came from testers asking how the customer would know who was writing, and the Linked Interactions dropdown came from testers asking where the related emails and calls were.

15

Some key takeaways from this project are

A dense screen can be calm if the structure is strict. One matrix with one rule for colour carried the whole product.

Keep the home screen on screen. Opening things beside the grid, not instead of it, made the flows feel fast.

Different acts need different looks. Coaching an agent and talking to a customer share a thread, so the label and colour do the work.

Build the prototype around real tasks. One end to end flow made reviews with stakeholders short and focused.

Next time I would test the red timer thresholds with real supervisors earlier, because each team defines a delay differently.

Thank you for your time!

Work together

Have a complex problem of your own?

Start a conversationSee all work