Back to projects

The agent who always
knows the answer.

Agent Copilot is a real-time AI panel in Verint's contact center desktop. It follows the conversation as it happens and surfaces what the agent needs before they have to search.

Background

Verint had strong tools for routing, recording, and workforce management, but nothing for the agent in the middle of a live conversation. Agent Experience was the initiative to change that, and the Copilot panel was the bet at its center.

It started as a hackathon. A panel that reads the conversation in real time and responds with context, knowledge, and guidance was new territory for Verint, and nothing in the existing product pointed to what it should look like. Going from prototype to production meant designing something reliable enough to run during any live customer conversation.

Hackathon Shipped
Shipped design of the agent desktop: the chat on the left, the customer profile and interaction history in the middle, and the Copilot panel on the right answering the agent's questions in cards, each with sources, a helpful rating and a box to ask more.
Hackathon design of the agent desktop: the chat on the left, customer details in the middle, and a Verint Copilot panel on the right with a robot mascot, a green chat summary card and a list of facts about the customer.
From the hackathon to the shipped product. Drag across to see either in full, or open it full screen.

The problem

Agents are expected to know everything, instantly. While a customer is talking, they keep track of context, search the knowledge base, follow compliance rules, manage the conversation, and write a wrap-up at the end.

All of it is manual. All of it adds up. Nobody was helping the agent.

The Copilot

A dedicated panel on the right side of the agent desktop. It tracks the conversation and responds, with no search, no switching screens, no extra effort.

Three moments carry the feature. At the handover, context is already waiting: a natural-language summary of the bot conversation sits in the acceptance screen, before the agent has even taken the chat. During the conversation, answers surface before anyone asks for them. At the end, the wrap-up writes itself. Two to three minutes of manual write-up becomes a 20-second review.

Below is one real conversation, step by step, with the reasoning behind each decision.

1 / 1

One real conversation, step by step. Drag the slider or pick a chapter.

Result

The Copilot was ready to ship on its own, ahead of the full Agent Experience launch. It became the first AI-first feature in Verint's contact center desktop and the first real-time assistant of its kind shown at Verint Engage. The figures below are from a customer deployment of in-the-moment knowledge access.

20s
Shorter calls
saved per call, reported by a bank using in-the-moment knowledge access
£6M
Annual savings
what that 20-second cut is worth across a UK contact center of about 1,200 agents
1st
First of its kind
real-time AI suggestions inside a live interaction, a new category for Verint

What I'd do differently

Scope

Set a hard scope boundary at the start.

We had many stakeholders, each with a slightly different idea of what Copilot should be. Without a scope and timeline locked in early, those competing visions kept resurfacing mid-sprint and pulled the design in different directions. A cutoff agreed upfront would have given the team something to push back against and made prioritization faster.

Alignment

Document every decision in writing, every time.

We held a lot of alignment meetings, but alignment that lives only in people's heads doesn't survive the next meeting. Decisions that weren't written down got reopened, the same questions came back with different people, and the design paid for it in extra iterations. A simple decision log (what we decided, why, and who was in the room) would have saved a lot of time.