Your AI Agent

Movra's AI agent replies to your studio's WhatsApp messages 24/7. Here's what it can do, what it won't do, and how you stay in control.

What it is

Think of the agent as a receptionist who never sleeps. When a customer messages your studio on WhatsApp, the agent reads what they said, figures out what they need, looks things up in your booking system if it has to, and writes a reply in your studio's voice — usually in under 30 seconds. You see every conversation in your Inbox.

What it can do

Answer the basics, instantly

Class times, prices, what's included in your trial, where you're located, what's on this week. The kind of question a brand-new lead asks before deciding to come in.

Example: "What time is the Saturday morning Bachata class?" — the agent checks your live schedule and replies with the time, the instructor, and the address.

Sell trials, packs, and memberships

On the Front Desk plan, the agent can take payments. It sends a secure Stripe checkout link, the customer pays in the chat, and the credits land in their account automatically.

Example: "How much is the 10-class pass?" — the agent quotes the price, asks if they'd like to buy it, and sends a payment link if they say yes.

Book new classes

If a customer already has credits or an active pack, the agent can book them into a class directly. It checks the schedule, confirms with the customer, then deducts the credit and adds them to the roster.

Example: "Use my pass to book me into tonight's Salsa Intermediate."

Cancel and reschedule

The agent handles cancellations and reschedules the same way your booking system does — including the late-cancellation window. If the customer asks to cancel after the cut-off, the agent tells them the credit will be forfeited and asks if they want to proceed anyway. It never invents a policy.

Example: "I can't make it to tonight's class." — even without the word "cancel", the agent recognises the intent and walks the customer through the right action.

Booking a trial for two? Each person needs their own email

When a customer wants to bring a friend to a trial class, the agent asks for each person's own email address — it won't book two people under one shared email. A trial is one per person, and your booking system checks trial eligibility by email, so each guest needs a separate email to be set up as their own client and get their own trial. If the customer can't give a separate email for their friend, the agent hands the chat to your team rather than booking both on one email.

Example: "Can I book my friend too? Just put us both under my email." — the agent explains each person needs their own email, asks for the friend's, and sets both up correctly for the trial.

Look up a customer's account

Packages, credits remaining, upcoming bookings. The agent checks the customer's identity first (a quick 6-digit code to their email) before showing any account info — so nobody can spoof their way into someone else's data.

Capture leads from your ads

When someone fills out a lead form on Facebook, Instagram, or your website, the agent messages them on WhatsApp first — usually within 30 seconds. Speed-to-lead is the single biggest predictor of whether a lead converts to a trial.

What it won't do

Just as important as what the agent does is what it deliberately stays out of:

Refunds and disputes. Anything involving money going back to a customer escalates to you — every time.
Inventing policies it doesn't know. If the answer isn't in your knowledge or the booking system, the agent says it'll check with the team rather than guessing.
Off-topic chat. If a customer asks something outside the studio's scope (advice on training, gossip, general questions), the agent politely brings the conversation back to what it can help with.
Account changes a human should approve. Transferring a membership, freezing a pack, or changing pricing for one customer always goes through you.

How it thinks

When a customer sends a message, the agent breaks it down: what is the customer actually asking for? Sometimes that's one thing ("what time?"). Sometimes it's several ("cancel my Wednesday class and book me into Friday instead"). Either way it tackles each piece in turn, then writes one reply that covers everything.

If a customer's tone is vague but the intent is clear, the agent still gets it. "I can't make it tonight" is a cancel request even without the word "cancel". "Use my credit to come Friday" is a booking even if they don't say "book".

When it asks for your help (escalation)

The agent hands a conversation to you whenever:

The customer explicitly asks to speak to a human.
The customer requests a refund.
The booking system returns something unusual and the customer insists something is wrong (the agent tries twice, then escalates rather than going around in circles).
The request is outside what the agent knows how to do.

When this happens, three things happen at once: (1) the AI pauses on that conversation, (2) a WhatsApp template message lands on the handoff phone you've configured in Settings → Escalation, and (3) the conversation gets tagged in your Inbox so your team knows it needs a human. You can take it over right there.

One thing to know: after a handoff the AI stays paused on that chat — it does not jump back in on the customer's next message unless you've told it to. The next section explains exactly how it comes back.

After a handoff: how the agent comes back

A paused conversation shows a small status badge in your Inbox. It tells you exactly how — and whether — the agent will return:

Auto-resumes at [time] — the agent comes back on its own at that time. Until then, the chat is yours.
Paused · manual — manual unpause only. The agent will not come back by itself; it waits until you resume it, no matter how long that takes or how many times the customer writes in.
Blocked — the AI is switched off for this one contact and never auto-resumes. Messages still arrive in your Inbox, but the agent stays silent until you turn it back on.

Which one you get is controlled by your Auto-resume setting in Settings → AI → Escalation. Set it to "Never (manual unpause only)" and an escalated chat stays paused until a person resumes it — the safest choice, since a human always closes the loop, but it does mean a handoff can keep the agent quiet until someone acts, even on a brand-new question weeks later. Prefer the agent to pick back up on its own? Choose a window instead: 15 minutes, 30 minutes, 1 hour, 2 hours, 4 hours, 24 hours, or a custom number of minutes. (Purely technical hiccups — a booking-system or connection blip — resume on their own after a few minutes regardless of this setting; only genuine "needs a human" handoffs follow the window you pick here.) It's worth checking this setting so a handoff doesn't leave the agent silent longer than you intend.

You can also override the window for a single conversation from the pause menu in the Inbox — auto-resume in 15 minutes, 1 hour, or 4 hours, fall back to your studio default, or block the AI for that contact entirely — without changing your studio-wide setting. One thing to know: if the chat is already paused, changing the duration here applies to the next time it pauses — the current pause keeps the timer it already has. To bring the agent back on the current chat right away, use Resume AI (below).

And whenever you want the agent back right now, click Resume AI on the paused chat. It reads everything that happened while it was off, then carries on from there.

Waking the agent manually

Sometimes a conversation is active but the agent hasn't replied — a customer wrote something that didn't need an answer, or you moved things along yourself and now want the agent to take the next step. The Wake button (🔔, at the top of every conversation) tells the agent to take another look and act right now.

When you wake it, you can add a short private note — up to 280 characters, something like "offer her the Friday 7pm class" — that steers this one reply. The customer never sees it. If the agent looks and decides nothing needs sending, it tells you that rather than messaging the customer just to fill the silence.

Two things to know: there's a short 5-minute cooldown between wakes on the same chat, and Wake only works when the AI is active. If the conversation is paused (or your whole studio is paused), resume it first — waking a paused chat won't do anything.

You're always in control

Everything the agent does is configurable. The five most useful controls:

Pause the AI on any conversation — one click in the Inbox. You take over the chat manually. When you're done, unpause and the agent picks up where it left off (it reads what happened while it was off, then continues). You choose whether it comes back on its own or waits for you — see After a handoff above — and you can nudge it to reply on demand with the Wake button.
Voice and tone — Settings → AI → Prompts. Tell the agent how to speak: warm, professional, friendly with emojis, no emojis, formal, casual. The change applies to every reply from then on.
Custom answers — Settings → AI → Knowledge. The little things that are specific to your studio: parking, dress code, your "no late cancels" policy, the address of your second location. The agent pulls from this when a customer asks.
Working hours — Settings → Studio. The agent uses this to phrase replies ("we open at 9") and decide when something is genuinely urgent versus when it can wait.
The Playground — Settings → AI → Playground. Type any customer message and see exactly what the agent would reply, without sending anything. Useful for testing a new prompt or seeing how the agent handles a tricky scenario before it happens for real.

Tips for getting the best out of it

Write your knowledge in plain customer language. "We're a 5-minute walk from Bugis MRT, exit B" works better than "Location: 123 Ophir Rd".
Test in the Playground before going live. Throw your three most-common customer questions at it. If the replies sound like your studio, you're ready.
Don't over-instruct. The agent works best when it has clear policies, not a long script. "Late cancels forfeit the class credit" is enough — you don't need to specify what it should say in every edge case.
Set the handoff phone before you go live. Settings → Escalation. The first time the agent escalates, you want the notification to actually land somewhere.