Build with AI

The Train your AI Receptionist section is the first section of your AI Receptionist’s editor and the fastest way to shape it — on day one and any day after. Tell your Receptionist what it should handle on a call, in plain English, and it sets itself up as you talk — the greeting, the journeys, the details to collect. You can fine-tune everything it builds in the other sections whenever you like. To get there, open AI Receptionists in the sidebar and pick your Receptionist — the editor opens on the Train your AI Receptionist section. Choosing Build with AI when you create a new Receptionist lands you here too. The editor lists its sections down the left — Train your AI Receptionist, Personality & voice, Journeys, Outcomes, Knowledge, Business numbers, Transfers and After the call — with your Receptionist’s name and status above them and the Checklist at the foot. On a narrower screen the list becomes a picker at the top that shows the section you’re in; click it to choose another. Along the top of the editor sit Versions, Test and a ⋮ menu, plus Discard draft and Publish changes whenever you have unpublished changes.

​

The two halves of the screen

The chat on the left is where the training happens. A tip above it uses your Receptionist’s own name — “Tell your Receptionist’s name what to do, just like briefing a new team member.” — and that’s exactly the idea. You can close the tip with its ×, and it stays closed. Type what you want in your own words — your Receptionist already knows your business from your workspace’s business profile on Business details, which every AI Receptionist reads at the start of each call, so you don’t need to explain who you are. It also reads what you’ve typed in the What your Receptionist always knows box in the Knowledge section, so it can answer from it — if your prices are there, ask “do you know our prices?” and it quotes them back. It can’t change that box, though: you edit it yourself in the Knowledge section. It answers as itself, in the first person: a brand-new Receptionist asks what you’d like it to take care of when someone calls, while one with journeys already set up opens by telling you what it can do today and asking what you’d like to change or add. Naming a job — “book an MOT”, say — starts a short interview rather than a build: your Receptionist asks one thing at a time (what to ask the caller, what to collect, where it goes, what to do if it can’t be done, and what to do when someone asks for a person), then shows you the plan with a separate list of anything it would be assuming, and builds only when you say yes. If you’d rather skip the questions, say “just build it” or “sensible defaults are fine” and it will, listing its assumptions afterwards. The Checklist on the right is Moneypenny AI’s short list of what’s worth doing next for this particular Receptionist — its header shows how many items are left, or All done once there’s nothing outstanding. Each row carries a button that does the thing — asking your Receptionist, opening the right section, starting a test — and the list keeps itself up to date as you build; Your Checklist below explains how it works. You can hide the Checklist with its × to give the chat the full width, and bring it back with the Show the Checklist button above the chat, or from the Checklist entry at the foot of the section list, which also shows how many items are left. On a narrow window or a tablet there’s no room for the Checklist beside the chat, so it becomes a second tab on the chat frame — Chat and Checklist, with the same count on the Checklist tab — and everything on it works the same way. To try what you’ve built, click Test at the top of the editor, or the Test tab on the right edge of the screen — the test drawer floats over the right-hand side of the page with a chat conversation and a live voice call, and it works from every section of the editor. In this section it covers the Checklist and a little of the chat’s right-hand edge, so the conversation stays in view while you test; close the drawer and the Checklist is back.

​

Build, watch, test

​

When nothing changed, or something went wrong

Your Receptionist speaks in its own words only when its change was saved, or when it’s simply answering a question. Anything else comes from Moneypenny AI itself, as a notice that looks different from a chat balloon — so you always know whether your Receptionist really changed:

• “No changes were made to your Receptionist.” — your request left your Receptionist exactly as it was. Nothing was saved and there’s no What changed record.
• “I couldn’t make a valid change for that — could you rephrase or simplify what you’d like?” — your Receptionist couldn’t turn the request into a change that works. Nothing was saved; try it in different words, or in smaller pieces.
• “Your Receptionist changed while I was working. Nothing was updated; please ask again.” — your Receptionist was edited from somewhere else during the turn (another browser tab, say). Rather than build on an out-of-date copy, it stopped. Ask again and it starts from the current version.
• “The reply was cut short. This is the latest saved version of your Receptionist; the change may still be finishing.” — the connection dropped before the reply arrived. The editor shows the latest saved version of your draft; look at it before asking again, because the change may have been saved even though the reply was lost.

If a business profile change was agreed in the same turn, the notice says whether it was saved — a profile change is saved on its own, so it can go through even when the change to your Receptionist didn’t. You can keep editing by hand in the other sections while your Receptionist is working, but if you do, its change isn’t applied over yours. Your own edits stay on screen, and the “This Receptionist changed elsewhere” notice appears at the top of the editor; click Show latest to see your Receptionist as it saved itself — the hand edits you made in the meantime are replaced, so note them down first if you want to redo them. Editing from more than one place explains how that notice works.

​

Changing your business profile from the chat

Your business profile — your business name, what the business does, what you offer, your opening hours, your sector and your contact details — belongs to the workspace rather than to any one AI Receptionist, and every Receptionist reads it at the start of each call. So when you tell your Receptionist something about the business (“we’re open until 7 on Thursdays now”, “we’ve started offering emergency call-outs”), it doesn’t write that into a journey step: it offers to update the profile instead. Opening hours in particular always live in the profile, so a journey step can simply say “only offer times within our opening hours”. Because a profile change is live for every Receptionist in the workspace the moment it’s saved, your Receptionist first tells you exactly what it will change and waits for you to agree. Say yes and the profile is updated, with the change listed under What changed — “Updated the opening hours”, “Renamed the business to Rivermead Physio”, “Updated what the business offers”, “Updated the business details” for contact details. Rename your Receptionist or the business and it rewrites its greeting with the new name in the same step, so callers never hear an old name. Only workspace admins can change the profile this way. If you have the User role, your Receptionist tells you that a workspace admin can change it. Admins can also edit it directly on Business details, under Your business.

​

Transferring callers to a person or team

Tell your AI Receptionist who a kind of call should go to — “Any calls about lettings should go to Wendy Lloyd-Goodwin”, say, or “New business enquiries should be put through to the Sales team” — and it adds a step with the Transfer to a person or team you pick action. If that person or team is already in your Receptionist’s Transfers section, the step comes back pointed at them, and your Receptionist confirms who those calls now go to. This is a change to your draft like any other, so it goes live when you publish, and nothing in the Transfers section itself changes. Use the names as they appear in the Transfers section. Your Receptionist always knows your teams; it recognises a person when you say their name, and a first name or a surname on its own is enough when only one person on the list has it (“send weddings to Daniel”). A nickname isn’t — “Dan” won’t find Daniel — and nor is a description: your Receptionist won’t guess that “the front desk” means your Reception team, and checks with you instead. Your Receptionist only ever chooses from the people and teams already in that section. It never adds anybody, never asks you for a phone number, and tells you the rest in words — so three replies are worth knowing:

• Somebody who isn’t on the list yet. The step is still added, with nobody chosen to ring, and your Receptionist tells you they need adding in the Transfers section first. Add them in the Transfers section (in the same section list as Train your AI Receptionist), then either pick them in the step’s Who to ring list in the Journeys section or come back to the chat and say so: “she’s on the list now — point the step at her.” Until somebody is chosen, Publish changes is greyed out, with the reason beside it: “The hand-off on ‘Complaints’ needs somebody to ring.”
• Two people could be meant. Say “Billing questions should go to Smith” with a Jack Smith and a Robert Smith on the list and your Receptionist asks which one rather than guessing. Answer with the full name and the step is pointed at them. With more than three possible matches it asks for a surname instead of reading names out.
• Somebody who can’t be rung right now — switched off as away, or with a number that can’t be used. They are on the list, so your Receptionist doesn’t send you to add them; it says they can’t be rung at the moment and asks who else. As above, the step is added with nobody to ring until you choose.

Which action the chat picks depends on who does the choosing. When you name the destination, as in every example above, it’s Transfer to a person or team you pick. Only when you want callers put through to the person they ask for — “If a caller asks for somebody by name, put them through to that person” — does it use Transfer to the person the caller asks for, which listens for a name on the call and searches everyone in People (people only — a caller can’t ask for a team). The two ways a caller reaches somebody explains the difference.

​

Booking, recording and sending through your connected apps

Tell your AI Receptionist to book, record or send something through one of your connected apps — “book the appointment into our calendar”, say — and it adds the action to the right step. For each detail the app needs, it works out who supplies it: a fixed setting of your business (which calendar, how long a booking lasts, whether invitations are emailed), a detail the caller gives on the call (their preferred time, their email address), or something your Receptionist decides in the moment (the wording of a note). It asks you in plain words for anything only you can answer, such as which calendar to use, and where it has assumed a setting instead it says so in its summary — “I’ve assumed 30-minute bookings and that invitations are emailed — say if not” — so you can correct it by replying. Dates and times need nothing from you: whatever the caller says goes through in your Receptionist’s time zone as an exact date and time — and the same goes for numbers, email addresses and phone numbers, which your Receptionist records in the shape the app expects — so there’s no need to describe formats or time zones in a step’s instructions. Where an app needs a window rather than a single moment — the span an availability check looks at — your Receptionist works the far end out from the caller’s time and the length of the booking, so you never do that arithmetic yourself. Where one action depends on a check — an availability check before a booking — your Receptionist links the two so the booking runs only after the check has come back for the details as they finally stand.

​

Your Checklist

The Checklist beside the chat is Moneypenny AI’s suggestion of what to do next, written for the AI Receptionist you’re building — a Receptionist that takes bookings is prompted to connect its calendar, one that takes messages to switch on the after-call summary so messages reach someone. Each row has a short title and a one-line hint underneath saying why it matters. It’s a prompt, not a gate: nothing on the list stops you testing or publishing. Each open row has one button that does the thing. For anything the chat can build — your Receptionist’s name, greeting and personality, its journeys, and your business profile (its name, opening hours and the rest) — the button is Train followed by your Receptionist’s name — Train Barney, say — and a ready-written request is sent to the chat for you, just as if you’d typed it. It opens the conversation rather than making the change: your Receptionist asks you for anything it needs (your opening hours, say) or shows you the change to confirm, and nothing changes until you’ve answered or said yes. For anything that lives elsewhere, it’s Open … — Open Business numbers, Open After the call, Open Knowledge, Open Connected apps, Open voice settings and so on — while Start a test opens the test drawer and Publish now publishes for you. When a row opens another section of the editor, a Back to your Checklist link appears at the top of that section, and the browser’s Back button returns you to the list too. A row leaves the list in one of three ways:

• Mark done. Tick it off yourself. There’s no undo: a ticked row stays ticked — across reloads, and for anyone else who edits this Receptionist.
• Moneypenny AI spots it’s done. When it can see the thing has happened — a business number assigned, knowledge attached, a journey built, opening hours saved to your business profile — the row ticks itself with a green “Done — we spotted this on your Receptionist” note. These can’t be unticked.
• Not needed. Takes the row off the list. A message — “Removed from your list — we won’t suggest it again.” — gives you a few seconds to press Undo; after that the removal is permanent and the item isn’t suggested again.

Done rows fold away under a Done (n) toggle so the open ones stay in view, and each carries a single Remove — now and then a row ticks itself on a reading that’s wrong for your Receptionist, and this takes it off the list, with the same few seconds to Undo as Not needed. When nothing is left the list says “All set — nothing left to do for now.” It refreshes when you open this section and after every chat turn — the first time, it briefly shows “Working out what’s next…” while the suggestions are written. Two rows look after themselves and have only their button — no Mark done, Not needed or Remove: “Try a test conversation” (Start a test) clears once you’ve tested the current draft, and “Publish your changes” (Publish now) appears whenever there’s something unpublished and goes when you publish. A third row is looked after the same way, and it’s the only one on the list you can turn down. Have a person pick up the calls your agent can’t finish — “One of our receptionists answers instead of your caller leaving a message. Ask us to set it up.” — offers Open Moneypenny receptionists, which lands straight on that sub-tab of Transfers. Once your account has a receptionist queue this Receptionist isn’t using, the row reads Choose who picks up the calls this agent can’t finish instead, and the same button takes you to the switch. It has no Mark done, because the list works out for itself whether this Receptionist hands callers over: the row drops off as soon as you switch the hand-off on, and comes back if you switch it off again. It does have Not needed, unlike the other two — press it and it’s gone for good, with the same few seconds to Undo as any other row. That one goes further than the rest: it’s a decision about the whole account rather than this Receptionist, so you won’t be asked again on any of your AI Receptionists, including ones you build later. Only an admin sees the row, since only an admin can ask us for a queue or work the switch. You won’t be offered it while a request for a queue is already with us, since there’s nothing to do but wait, and it doesn’t appear on an account that can’t be sold the service — one without a paid plan, or one that isn’t taking calls. If your Telephone Answering is part of a Moneypenny agreement, only the “choose who picks up” version appears, once we’ve set a queue up for you.