How to test an AI support agent before it goes live
With Aide you test on your own conversations before any customer sees a reply: preview the draft a scenario would write on real conversations from the last 28 days, chat with the agent yourself in its preview, and roll each helpdesk inbox out from Watch only to test senders, a share of conversations and then everyone. Nothing is sent to customers or written to your helpdesk while you preview or watch.
Updated · Every plan, Free includedThe ways to test Aide before going live
| Match count on real conversations | While you build a scenario, Aide shows how many of your conversations from the last 28 days its conditions match, exact for small sets and estimated above that, with up to five real examples. |
|---|---|
| Preview draft | Owners and admins run Aide's reply on any of those examples with the saved scenario, even while it is switched off. You see the reply, or why Aide would hand off, skip or hold back, and which scenarios applied. Nothing is sent or written. |
| Agent preview | Chat with the agent from its page, as a customer would. The preview says what it is answering from, for example "Answering from 120 pages of acme.com". |
| Drafts on recent tickets | Connect Zendesk, Gorgias or Front and Aide drafts replies to up to 10 recent open tickets and 3 recently solved ones inside Aide, with your team's actual reply beside each solved one. |
| Watch only | A helpdesk inbox or mailbox in Watch only decides and drafts on live traffic inside Aide and writes nothing to the helpdesk. The inbox lists every conversation a scenario ran on, Watch only included. |
| Staged rollout | Then test senders (addresses you choose), a share of conversations, and everyone. Per inbox, each reply is sent, drafted or neither, with the reason shown. |
| Topic levels | New topics start at Drafts. Aide sends replies and acts on its own only on topics you move to AI Agent. |
| Cost | Previews and spam are never counted as AI replies. Replies to test senders count like any reply, and drafts Aide keeps inside Aide in Watch only count once per customer message, like any draft. |
A test-before-launch routine
- Step 1
Build the scenario
Set the conditions (topic, order, shipment, fields) and the instructions or actions. The editor says whether Aide drafts, sends or writes nothing on that topic.
- Step 2
Check what it matches
See how many conversations from the last 28 days match and open the real examples, before the scenario is switched on.
- Step 3
Preview the draft
Run Aide's reply on those examples. Fix the instructions or knowledge, save and preview again until the answer is right.
- Step 4
Roll out in stages
Keep the topic at Drafts, then AI Agent. Move each inbox from Watch only to test senders, a share of conversations and everyone.
Watch it work before customers do
Every example below is a preview or a staged rollout, not a reply to a customer. This is what your team sees before Aide answers anyone.
Tested before launch. Checked after.
Testing decides what goes live. These keep it accountable once it does.Test on the conversations you already have
Connect your helpdesk or add Aide Chat, and testing runs on your own conversations. No migration and no engineering.
Drafts on recent tickets as soon as you connect.
Watch only, then test senders, per inbox.
Per-inbox rollout from Watch only to everyone.
Order and shipment conditions in every preview.
Contact fields as scenario conditions.
Search knowledge and read conversations from a Claude chat.
Frequently asked questions
In Aide you test on your own conversations: see how many conversations from the last 28 days a scenario matches, preview the draft Aide would write on them, chat with the agent in its preview, and keep a helpdesk inbox in Watch only before widening it to test senders, a share of conversations and everyone. New topics start at Drafts, so a person sends every reply until you move the topic to AI Agent.
Yes. Scenario match counts and Preview draft run on your real conversations from the last 28 days. When you connect Zendesk, Gorgias or Front, Aide also drafts replies to up to 10 recent open tickets and 3 recently solved ones inside Aide, with your team's actual reply beside each solved one.
No. A preview shows the reply Aide would write, or why it would hand off, skip or hold back, and nothing is sent to the customer or written to the conversation or your helpdesk. An inbox in Watch only drafts inside Aide and writes nothing to the helpdesk.
Yes. Each helpdesk inbox moves through Watch only, test senders (only addresses you list get Aide's replies), a share of conversations, and everyone. You can also start with one topic at AI Agent and keep the rest at Drafts or None.
Owners and admins. A preview uses the saved scenario, even while it is switched off; with unsaved changes it asks you to save first. A workspace can run 30 previews an hour.
No. Previews and spam are never counted. Replies Aide sends to test senders count like any other reply, and drafts Aide keeps inside Aide while an inbox is in Watch only count once per customer message, like any other draft.
Every reply is checked before it is sent, refunds wait for your team by default, Aide says it is an AI when asked, and each conversation shows the topic, sources and actions behind every reply. The audit log keeps every settings change for two years.
Start free with 150 AI replies a month and no card, with every feature included. Paid plans start at $39 a month with a 7-day free trial, and you pay for AI replies, not seats. See aide.app/pricing.
Governance and trust, defined
The vocabulary of testing and trusting AI agents, defined.
Test Aide on your own conversations, free
Every feature is on the Free plan, previews included. Connect your helpdesk and see Aide's drafts on your recent tickets in minutes.