Questions, answered

An afternoon. You pick a role, point it at your website so it can read your services and prices, check what it learned, then paste one line into your site for the chat and forward your support inbox to the address we give you. The reading is the part that takes longest, and it happens while you do something else.

Your website, to start. It reads your pages and builds its own notes from them, which you can then correct — that is usually enough to answer most of what people ask. Anything not on your site, like your callout fee or which suburbs you cover, you paste in as notes. You are not writing scripts or building conversation flows; you are telling it facts.

No, and this is the part most people expect to be harder than it is. You keep the address your customers already write to and forward it to ours. Replies go out with your address on them, so as far as your customer is concerned they wrote to you and you wrote back. Undo the forward and everything is exactly as it was.

There is a live one on our demo page, answering as a real AI employee rather than playing a recording. Ask it something awkward. It is the same software you would be buying, with our business loaded into it instead of yours.

A virtual assistant is a person, shared with other businesses, working their hours, at roughly $25 to $40 an hour. An AI employee works one business — yours — replies at 3am, and costs a flat monthly rate. Keep your people for the work that needs judgment. This is for the work that needs somebody to answer.

It hands over. You set the handoff target per employee: a person, a team, an inbox. The customer is told a human is picking it up, and whoever does already has the whole transcript. It is built to hand over early rather than bluff.

No. Asked directly whether it is a bot, it says so — and that question also counts as asking for a human, so it hands the conversation to your team at the same time. Claiming to be a person is the kind of thing that ends up in a screenshot, and it is the one trick that would cost you more than it ever saved.

It books. It offers times that are genuinely free in the calendar you already keep, holds travel time separately from the appointment itself, and writes the booking in. Two people cannot take the same slot, and whoever owns that service gets told.

Every conversation, in full, with a summary and what it decided. You can search across all of them. Nothing is hidden from you — including the messages it chose not to answer, and why.

Yes, and without charging you for it. A support inbox gets agency pitches, supplier invoices, recruiters and phishing. Those are recognised, set aside without a reply, and kept — so you can look at what it decided rather than taking our word that it decided well.

That is the right question, so here is the honest answer. Every AI employee runs with automatic failover to a second model provider, so one provider having a bad hour does not take your chat and email down. Conversations that failed over are flagged in your dashboard, with timestamps, so you can check rather than trust us. What we will not claim is that it never fails. Nothing never fails. What it does is fail in ways you can see.

It is worst in its first two weeks, before it has seen your edge cases. It will occasionally take a message when it could have booked the job, and occasionally answer a pricing question more confidently than it should. Review the transcripts daily at the start — they are all in your dashboard — and correct what it got wrong. Most of that settles inside a fortnight.

Your pages, your documents and the notes you write. It answers from those and nothing else. Asked something they do not cover, it says it does not know and takes the person's details rather than guessing — because a confident wrong answer about your prices costs you more than no answer at all.

You can move any employee to a stronger model from its own page, which is better with long conversations, awkward customers, and working out what somebody meant rather than what they typed. It costs twice the credits per reply, and the page tells you that before you switch. Most businesses run the standard one and put the premium model only on the role that needs it.

One unit of work done for a customer. A typical written reply is about one credit, and a minute on the phone is one credit, so a balance means the same thing whichever way your customers reach you. That is the whole reason the number exists — you should be able to look at it and know roughly how many more conversations you can have without translating between minutes and messages first.

Length. A short confirmation costs a fraction of what a detailed answer costs, because it is a fraction of the work. Billing by length rather than a flat rate per message means a one-line reply is not charged as though it were three paragraphs, and it stops the bill drifting away from what actually happened. Your dashboard shows the balance falling in real time if you want to watch a few conversations and get a feel for your own traffic.

Anything that is not an answer to a real customer. Cold outreach, supplier invoices, newsletters, auto-replies and spam are recognised and set aside without a reply and without a charge — they are still filed so you can see what was decided. Conversation summaries, the alerts we send you, and everything you do in the dashboard are free too. You are charged when your AI employee speaks to somebody, and not otherwise.

Your plan's allowance is added on the day your subscription renews, and anything you did not use stays where it is. If you finish May with 10 credits left and your plan includes 250, you start June with 260 — the allowance is added to your balance rather than replacing it. A quiet month is not a wasted one.

No. Credits do not expire, and there is no monthly use-it-or-lose-it deadline. Whether they came with your plan or you bought them in a pack, they sit in your balance until the work uses them. You paid for them, so they are yours until they are spent — a seasonal business that is quiet in January and busy in June is not penalised for the shape of its year.

Your AI employees stop answering. That is deliberate, and it is the alternative to a surprise invoice — we would rather stop and tell you than keep working and bill you for it. Nothing is lost while you are stopped: conversations, contacts and history stay exactly as they were, and buying credits starts things again immediately.

Yes. Set a low-balance level on your billing page and you are emailed when the balance falls past it, with further warnings as it keeps dropping. Pick a number that gives you time to act rather than time to panic — if a normal day costs you thirty credits, being told at ten is being told too late.

No. It is a seat per AI employee per month, the way payroll works. Add one when you need the cover and drop it the month you do not. No setup fee, no minimum term, and no conversation with anybody to cancel.

No. We do not use your conversation content to train our own models, and we contract with the model providers on terms that do not permit them to train on it either. Your customers' questions are not raw material.

Categories of provider — hosting, database, email delivery, model inference, and so on — each under a contract limiting them to that purpose. They are listed on our privacy page. If you need the specific companies for a vendor review or a data processing agreement, ask and we will send the current list.

Both. Ask and we will export it or delete it, within 30 days and usually sooner. Closing your account starts a 30-day window before data is removed, so changing your mind is recoverable. We will not treat you worse for asking.

Yours. For the people your AI employees talk to, you are the data controller and we are your processor, acting on your instructions — your own privacy notice governs what you tell them. For your own account details we are the controller. Both are set out on our privacy page.

Open the conversation in your dashboard and read the transcript — it will usually show exactly where it went off. Most errors trace back to a gap in the business notes on the employee, not to the model. Add the missing fact, and it stops. If it keeps happening after that, send us the conversation link.

Check three things in order: the employee's status is Active rather than Draft or Paused, its email channel is switched on, and your inbox is actually forwarding to the address we gave you. The employee page shows the first two. If all three look right, send us the address you wrote from and roughly when.

The usual cause is credits running out — that stops work deliberately rather than running up an unbounded bill. Your billing page shows the balance. The other cause is a failed payment, which you would have had an email about.

Add them to your do-not-contact list from the contact record, which suppresses them across every outbound employee immediately. Do this even if the request came through another channel.

Set escalation targets on the employee. A customer who asks for a human, gets upset, or raises something outside the employee's brief is handed over there, and whoever picks it up gets the summary first. You can also open any conversation and take it over yourself.