Support metrics glossary
Agent assist. The AI drafts, the agent sends.
Agent assist is software that drafts replies, finds answers and summarizes threads for a human agent who clicks send. It is a category noun, not a product name, and it lives at the other end of the queue from the customer-facing bot.
Last updated 14 August 2026
What agent assist means
Agent assist is AI that works on the agent's side of the desk. It reads the incoming ticket, drafts a reply from your documentation, pulls up the order in question, and summarizes the thread. The agent reads the draft, edits what needs editing, and clicks send. The customer never sees the machine. They see a fast, correct answer from a person.
The contrast that matters is the customer-facing bot. A bot talks to the customer and owns the conversation until it gives up. Agent assist talks to the agent and owns nothing. One removes the human. The other removes the busywork around the human.
What it does during a shift
Strip the branding and agent assist does four jobs. It drafts a first full reply from your help center and similar resolved tickets. It searches: the article, the order state and the account record, already open. It summarizes: a forty-message thread becomes six lines at handoff. And it does the paperwork: tags, fields, disposition codes.
None of the four sends anything. The agent is the author, the machine is the typist.
The part the demos skip
Here is the flat version nobody puts on a landing page. Agent assist is only as good as the knowledge base it reads, and most knowledge bases are stale. Keeping one current is real, unglamorous work, and we say that as a company that does it for a living. A draft built on a two-year-old article is a confident wrong answer, typed faster. Demos never show this. Demos run on ten perfect articles.
The measurement is just as soft. Teams buy it to cut handle time, then report suggestion acceptance, which counts clicks on the draft, not faster closes. Measure the queue, not the clicks.
Not a bot, and the difference is operational
Both get sold as support AI, and they are not interchangeable.
| Dimension | Agent assist | Customer-facing bot |
|---|---|---|
| Talks to | The agent, inside the helpdesk | The customer, in chat or email |
| Who clicks send | The human, every single time | The bot, until it escalates |
| When it is wrong | A bad draft the agent deletes | A wrong answer the customer already read |
| Metric it moves | Average handle time, cost per ticket | Deflection rate, containment rate |
The math, with the arithmetic shown
The metric agent assist exists to move is average handle time, and cost per ticket rides along with it. The arithmetic is boring. That is the point. Say a ticket takes 10 minutes: 6 talking or typing, 4 searching, reading, filling fields. A good draft and summary take that second block from 4 minutes to 1. New handle time: 7 minutes.
Now scale it. Five hundred tickets a month at 10 minutes is 5,000 minutes. At 7 it is 3,500. The team gets about 25 hours back without hiring anyone. That is the whole business case. Any bigger number shown without the arithmetic is a guess.
Which one you buy
If your queue is repetitive questions with clean answers, a bot deflects them before an agent reads them, and containment rate tells you how honestly that bot is measured. If tickets need judgment or a refund decision, deflection is the wrong goal and agent assist is the right one. Most queues need both. The cleanest bot example is WISMO, the question your order system answers on its own.
Questions, answered first
What is agent assist in customer support?
AI that helps a human agent during a conversation. It drafts the reply, finds the answer, summarizes the thread, fills the fields. The agent reviews and sends. The customer talks to a person throughout.
Is it the same as a chatbot?
No. A chatbot talks to the customer and sends its own answers. This talks to the agent and sends nothing. The failure modes differ too: a bot fails in public, a draft fails in private.
Does it replace support agents?
No, and tools sold that way underperform. It removes the searching, typing and summarizing around the conversation, not the conversation itself. Judgment, angry customers and money decisions still belong to the person clicking send.
Which metrics does it improve?
Average handle time first: drafting and summarizing cut the minutes around each ticket. Cost per ticket follows, since the team closes more per hour. First response time improves on email queues, where the draft is waiting when the ticket opens.
What does it need to work well?
Current documentation. The draft is built from your knowledge base and past tickets, so stale articles produce stale drafts at scale. Second, a fast edit step: if editing is slower than retyping, agents retype and the gain evaporates.
One page over
See the customer-facing end of the queue.
The largest bot-deflectable ticket category in ecommerce, priced with the arithmetic shown.