Learner mode: how to let the AI earn its autonomy topic by topic
How to turn on Volga's learner mode: the AI suggests, your team decides, and each topic answers on its own only after proving it gets things right.
In a nutshell
Learner mode changes when the AI answers: instead of replying on its own, it suggests each reply in the Inbox and your team uses, edits or discards it. Volga measures how often it gets each topic right, and you choose which ones can be answered automatically.
What changes when you turn it on
Turning on an AI assistant is scary for a good reason: you don't know how it will answer until it already has. Learner mode lets you try it with your real customers and questions, without the AI sending anything on its own.
While it's on, your customers' messages generate reply suggestions in the Inbox, and a person decides what to do with each one. At the same time, Volga measures how many suggestions go out unchanged in each topic. With those numbers you decide, topic by topic, what can be answered automatically.
How to turn it on
An admin turns it on and there's nothing to install: it uses the same knowledge base your assistant already has.
From that moment, topics you haven't approved stop being answered automatically and become suggestions. It's worth telling your team first, because replies now depend on them.
- Sign in as an admin: agents can't turn it on or off.
- Open Intelligence → Learner mode.
- Tap “Turn on learner mode” and confirm. The Audit log records who did it and when.
- If your business hours are set, choose what happens outside them: “The AI answers what it knows” or “Wait for an agent”.
Use AI with your own business information.
Add products, policies and FAQs, and define what it can answer and when it hands off to your team.
What your team does in the Inbox
Conversations with a suggestion show the “Proposal ready” badge. When they open one, agents see the AI's proposal and have three options.
The customer won't get anything until someone sends it. If the customer writes again, the suggestion is redone with the new message. If the knowledge base doesn't have the answer, the suggestion says so: “This isn't in the knowledge base. Please reply yourself.”
The usual WhatsApp rules still apply: if more than 24 hours have passed since the customer's last message, you have to reply with an approved template.
- Use: sends the reply as is.
- Edit: fixes it before sending.
- Discard: removes it so the agent can reply by hand. They can give a reason: wrong information, not the right tone or something else.
When a topic can be answered automatically
Volga sorts each suggestion into a topic and keeps a separate count.
A topic is ready when, over the last 2 weeks, the team used at least 90% of its suggestions unchanged, with a minimum of 30. An edited or discarded suggestion doesn't count as right. Anything that doesn't fit a topic stays under “Other topic” and always remains a suggestion.
When a topic gets there, a card appears on Home. With “Yes, let it answer”, that topic starts being answered without waiting for anyone; with “Not yet”, the card comes back after 7 days. Nothing is approved automatically: an admin always decides.
- Hours and location.
- Prices.
- Stock.
- Shipping.
- Payment methods.
- Products.
- Warranty and returns.
How to undo it
Replies the AI sends on its own are marked in the conversation with “Replied on its own” and the topic. If a topic stops convincing you, “Back to suggestions” stops it. After that, it isn't offered again for 30 days.
If you turn off learner mode, the AI goes back to answering everything it knows on its own, as it did before, and approved topics are kept for when you turn it back on.
If an agent pauses the AI in a conversation, it keeps suggesting there but doesn't answer on its own, even if the topic is approved.
What changes by plan and channel
On plans with AI it suggests on WhatsApp, in Instagram direct messages and in the web chat. Instagram comments are still answered by the AI on its own.
On the Humana plan, which doesn't include the AI assistant, it suggests on WhatsApp and in the web chat, up to 300 suggestions a month, and nothing is answered automatically. When it reaches the cap, the Inbox keeps working as usual until the next month. For a topic to be answered automatically you need a plan with AI.
Frequently asked questions
Questions about this guide
Does the AI learn from corrections on its own?
It doesn't change how it answers on its own. Learner mode measures which suggestions go out unchanged and shows you which ones were edited or discarded. That tells you what's missing or wrong in your knowledge base, and replies improve when you fix it.
Does the customer notice the difference?
They get a normal message. The difference is timing: while a topic is in suggestion mode, the reply goes out when someone on the team sends it, so you need people attending.
Who can turn it on and approve topics?
Only an admin. Agents see the suggestions in the Inbox and decide what to send, but they can't turn the mode on or approve topics.