Training before real calls

Before your AI employee answers real customers, it trains. It takes practice calls from pretend customers, the same way it will take real ones: booking, prices, emergencies, upset callers, asking for a person, and a few callers trying to trick it. You see how each call went.

Find it on your employee's Training tab, or under Training in the menu.

Before training can start

Finish these first, because each one changes how your employee answers:

• Add your services and service area in Business profile.
• Set a hand-off number.
• Add business knowledge on the Knowledge page.
• Confirm your opening hours, if it books appointments.

Then press Start training.

What passing means

A round counts when all of these are true:

• At least 25 training calls finished.
• At least 90% of them went well.
• No must-fix calls went wrong.
• It trained on a real AI model.
• Nothing changed since: not the employee, your business details or your knowledge.

If a round doesn't pass, you'll see "needs your help", with what to fix. Training doesn't change your employee by itself; what you change does. Fix what's listed, then train again.

Accepting the result

If you're happy with how the calls went, you can choose Accept and continue instead of fixing. You confirm you understand it may make the same mistakes with real customers. It only counts for that exact version; change anything and it trains again.

Training length

On a paid plan, you can pick a longer round to catch rarer mistakes: Quick training, Standard, Thorough, or Full sweep. The trial uses Quick training.

Trying it yourself

You can talk to your employee in the browser any time, by typing or speaking. Or have it call your phone: US or Canada, up to 5 test calls a day, 5 minutes each, free. These don't count as training.

Bookings, leads and messages from practice calls are marked as practice.