Guide · AI and knowledge base
How to roll out agent assist in a support team
In short
Agent assist suggests articles and drafts a reply while an agent works on a chat or ticket; the agent edits it and sends it. It helps most where the answers already exist in writing and agents spend time finding them. Roll it out on one team first, keep every reply in the agent’s hands, and judge it on your own before-and-after numbers, not a borrowed ROI figure.
Reviewed 9 November 2026. Terms used: Agent assist, RAG (retrieval-augmented generation), Ticketing system.
Seven steps to roll out agent assist
Most of the work is in the articles and the habits, not the switch.
Fix the knowledge before the AI
A draft is only as good as the articles behind it. Check that your most-asked answers are written down, current and in each language you support, and retire articles that contradict each other.
Decide what it may read
Customer-facing articles are safe to quote. Internal notes, such as refund exceptions or supplier contacts, may be fine to draft from but not to paste. Limit internal articles to the teams that should see them.
Keep the agent in charge of every reply
Drafts are suggestions. Agents read, edit and send; nothing goes to a customer by itself. Say that to the team out loud, so nobody sends a draft without reading it.
Start with one team and its common questions
Pick a team with many repeated questions and good articles, and run it for a month before you widen it.
Teach agents to check the source
Each draft should say which articles it used. Agents open the source when something looks off, and report a wrong article instead of only fixing that one reply.
Turn empty searches into articles
When an agent searches and finds nothing, an article is missing. Collect those searches and write the most common ones each week.
Judge it on your own numbers
Compare the month before and the month after for the team that used it: first reply time, reopened tickets and ratings. Your volume and your articles decide the result, so a figure from somewhere else won’t tell you much.
Agent assist vs a customer-facing AI bot
Who reads the answer first
- Agent assist
- The agent
- A customer-facing AI bot
- The customer
What it may draw on
- Agent assist
- Internal and public articles the agent may see
- A customer-facing AI bot
- Published customer articles only
Who sends it
- Agent assist
- The agent, after editing
- A customer-facing AI bot
- The bot, straight away
When it’s unsure
- Agent assist
- The agent writes the reply
- A customer-facing AI bot
- It hands over to a person
How it works in Axonave
Suggested articles and a drafted reply, sent only by the agent.
In Axonave, Agent Assist opens beside the chat or ticket composer. It suggests five articles from the customer’s last two messages and drafts a reply from the internal and external knowledge that agent may see: short plain text in a chat, paragraphs in a ticket, citing its articles and warning when it uses internal ones. A ticket draft reads the subject and the public messages, never internal notes. The draft goes into the reply box for the agent to edit and send; nothing is sent from the panel. A search that finds nothing is kept as an unanswered question, grouped with the bot’s, for a drafted article.
- Insert an article’s text, or a link to an external one
- Internal knowledge bases can be limited to roles or teams
- On Pro, or on Basic with an AI pack
Agent Assist beside Wong Kar Yan’s WhatsApp chat in our demo workspace: suggested help-centre articles and a drafted reply in Bahasa Malaysia, for the agent to check.
Common questions
Does Agent Assist ever send a reply by itself?
No. The draft lands in the reply box, and the agent edits it and sends it.
What does it cost?
It uses the AI tokens in your plan. Pro includes 2,000,000 a month, and an AI pack of 1,000,000 is $20 a month, on Basic too. There’s no charge per draft.
Who can use it?
Admins, managers, supervisors, agents and bot managers, on chats and tickets they may reply to. Auditors and viewers can’t.
Do you publish an ROI figure for it?
No. We haven’t measured one we’d stand behind, and yours depends on your volume and your articles. Measure first reply time, reopens and ratings before and after.
See it working with your own customers.
We reply within one business day by email or WhatsApp.