Skip to content

Overview ​

Overdesk can send replies to your customers itself, on Help Scout, Zendesk, Intercom and FreeScout. A workspace that turns sending on has Overdesk send the replies its guardrails approve, straight into the conversation in your help desk, and every other reply stays a draft or a note for your team. A workspace that does not turn it on keeps drafting exactly as before: every draft waits for a person.

Turning it on ​

Sending takes two keys, and it is off until both have turned:

  1. The Overdesk team enables sending for your workspace. Email hello@overdesk.io to ask.
  2. An admin of your workspace opens Guardrails, chooses Send replies automatically under Send mode, and confirms.

Until the first key has turned, Send replies automatically cannot be chosen, and nothing your workspace does can turn sending on by itself. Only an admin signed in through the dashboard can change the send mode.

To stop, choose Stop sending. The workspace goes back to Draft assist, and from then on every reply waits for a person again. A reply that was already on its way to your help desk at that moment still arrives. The Overdesk team can also turn sending off for your workspace at any time.

Overdesk also switches a workspace back to Draft assist by itself when its sending looks wrong: 3 replies in one conversation within an hour, 3 replies in an hour that your help desk refused or couldn't confirm, or a sign that the reviewer took one of Overdesk's own replies for your team's. It tells you under Alerts, and in Slack if your alerts go there. An admin turns sending back on under Guardrails once those conversations have been checked.

What is sent ​

Overdesk sends a reply only when all of these hold:

  • Sending is on for your workspace.
  • The conversation arrived on a channel Overdesk sends on. See Where Overdesk sends. A reply on any other channel waits for your team.
  • The reply was drafted in one of Overdesk's own runs: when a new customer message arrives, or on the 15-minute schedule, which Run auto-draft starts early. A follow-up nudge is written in the daily follow-up run instead, and is sent only when Also send follow-ups is on. See Follow-up nudges.
  • Your guardrail rules approve it. A rule under Answer automatically when has to match, and no rule checked before it can have held the reply. See Rules.
  • Nothing in the next section holds it.

A draft somebody asks for on one conversation is never sent: Re-draft on a draft's page, /overdesk draft in Slack, and drafts requested through the API all stay drafts.

Every reply that is not sent arrives in your help desk exactly as it would with sending off, as an internal note or a draft reply, and waits on Agent drafts for your team. A reply Overdesk held back also gets a short internal note of its own, headed "Overdesk held this reply", saying why it was not sent, on every help desk. When the conversation is handed to someone for that same reason, the hand-over line says it instead: at the top of the draft note, or in the "Overdesk escalation" note with draft replies. See Where drafts go.

Where Overdesk sends ​

Overdesk sends on email on every help desk, and on the other channels named in the first row of this table:

Help ScoutZendeskIntercomFreeScout
Overdesk sends onEmailEmail and API ticketsMessenger and email conversationsEmail
Waits for your team, for exampleChat and phoneWeb form, chat, messaging, SMS, voice and socialWhatsApp, SMS, phone and socialPhone, chat and custom
Status after a reply, your choicePending, Active, Closed, Leave as isOpen, Pending, Leave as isClose, Leave as isActive, Pending, Closed, Leave as is
Starts asPendingPendingClosePending
A reply that hands over leaves itActiveOpenOpenActive

A reply on any channel that is not on the first row waits for your team as a draft or a note, and the decision log gives the reason "Overdesk doesn't send on this conversation's channel yet". That includes a conversation whose help desk didn't say which channel it came in on, except on Help Scout, where it counts as email. See Decisions.

Your help desk adds to the email your customer receives, and Overdesk cannot remove it:

  • Zendesk adds its own footer, "This email is a service from" your company's name and "Delivered by Zendesk", the quoted original message and feedback buttons. Anyone copied on the ticket gets the reply too, so Overdesk checks each of them before it sends (see the list below).
  • Intercom adds the signature of the teammate Overdesk replies as, and "We run on Intercom". Replying also assigns the conversation to that teammate.

What is never sent ​

Whatever your rules say, Overdesk does not send:

  • a reply in a conversation it handed to a teammate, until that issue is resolved (see Escalation);
  • a reply in a conversation a teammate forwarded;
  • a reply to a conversation that changed while Overdesk was writing it: a new message, a reply or note from your team, a new status such as closed or spam, a different customer, someone newly copied on it, or a move to another inbox;
  • a reply in an inbox you took out of Allowed inboxes while Overdesk was writing it;
  • a reply in a conversation where Overdesk couldn't confirm its last reply went out, until someone on your team has added a note there or replied in the help desk itself (a reply in the first 30 minutes doesn't count, because it could be the one Overdesk tried to send, and neither does a note or automatic reply from a workflow or another app) and the issue is dealt with;
  • a reply or a follow-up nudge to a customer you have erased (see Erasure and deletion requests);
  • a reply to a customer under Excluded senders (see Privacy and data);
  • a reply that would also reach someone copied on the conversation, such as a CC on a Zendesk ticket, a CC or BCC on a Help Scout conversation, or another person in an Intercom conversation, who is under Excluded senders, on the customer list of a rule that holds replies, or was erased at any time. The note your team gets says only that someone the reply would reach is on one of those lists, never who or which list. If Overdesk can't read who is copied, it waits for your team too;
  • a reply that contains what looks like a private key, a licence key, an API key, a password or code written beside its label (a coupon code, an access token), a password written beside its username, or a WordPress application password written without a label, in a shape Overdesk recognises (see How customer data is handled), even inside a link or one the customer sent themselves;
  • a reply that contains what looks like a cloud storage secret key, such as an AWS secret access key, even with no label beside it, unless it is part of a link;
  • a reply that repeats a key the customer sent, a password or code written beside its label or its username, or a WordPress application password written without a label, in a shape Overdesk recognises: it keeps a placeholder where the value was, and waits for your team;
  • a reply that asks the customer for access to their site, such as a temporary login link, a temporary admin user or a collaborator request, or says the writer will log in to it. Overdesk never logs in to a customer's site; the reply goes to the person you marked as logging in to customer sites (see Escalation);
  • a reply to a customer who posted a link that logs in to their site since the last reply from your team that reached them, such as a temporary login link or a WordPress reset link. Overdesk hides the link from its records, raises an alert titled "Login link posted" and, when it can, hands the conversation to the person you marked as logging in to customer sites (see Escalation);
  • a reply that claims a person wrote it, or talks about where its answer came from;
  • a reply Overdesk tried a test site on, even when your rules approve it;
  • a reply that was allowed to ask the customer for their Site Health info, even when your rules approve it;
  • a reply that carries characters a reader cannot see, which can hide text or change the order of what your customer reads;
  • a reply to a message that isn't a request for support: marketing, spam, a scam, an automatic notice from a website or service, a bounce, an out-of-office reply or another help desk's acknowledgement;
  • a reply to a customer who asked for a person;
  • a reply that promises something your team has to do, such as a refund, or fixing something by a date. In a reply written in English, a second check, Overdesk's own and not the reviewer's, also looks for these promises. A reply whose only promise is passing the question on, or that the customer will hear back, can be sent, and Overdesk then hands the conversation to whoever keeps that promise (see Escalation);
  • a reply that repeats something only your internal notes say, or contradicts what the customer was already told, by your team or in an earlier reply Overdesk sent;
  • a reply that asks the customer again for something they already gave, without saying why;
  • a reply with a link or an email address that didn't come from your team's own material, such as one from a page or screenshot the customer shared;
  • a reply while Overdesk can't confirm its safety checks ran in the last 15 minutes: it waits for your team as a draft, and sending resumes by itself;
  • a follow-up nudge in a conversation where Overdesk did not itself answer the customer's last message, where a teammate wrote in the conversation (a note or a reply) after Overdesk's reply, or where the last reply told the customer, in wording Overdesk recognises in English, that your team would get back to them, and any nudge when the screen did not read the customer's last message as a request for support (see Follow-up nudges);
  • a reply the limits below stop.

See Holds on what a reply says for each of these.

Some of your starter rules hold replies too, and they start switched on: Human agent requested hands a customer who asks for a person to your team, Automatic replies and bounces holds out-of-office messages, no-reply senders and bounces, and the rules under Always hand to a person keep billing, refunds, legal and account security requests with your team. Turning one of those off lets its conversations be answered, apart from what the list above holds whatever the rules say: a customer who asked for a person, and a message a machine sent, still wait for your team.

Limits ​

  • Replies per hour: at most 10 replies an hour for the whole workspace by default. An admin can change it under Rate limit on Guardrails → Rules. When the hour's replies are used up, the rest wait for a person. A follow-up nudge Overdesk sends counts as a reply here, and a workspace sends at most half this limit, up to 5, in nudges a day, in your workspace's time zone, rounded down, so a limit below 2 sends none.
  • Loop guards: Overdesk does not reply when the customer's message arrived within 3 minutes of its last reply in that conversation, or repeats one of the customer's earlier messages word for word. Both are signs of another automated system answering Overdesk, so those replies wait for a person.
  • Replies per conversation: after 5 replies in one conversation in 24 hours, the next reply waits for a person. An admin can change that number under Replies per conversation on Guardrails → Rules. The same number limits the notes Overdesk answers in one conversation when your team asks it in a note, counted separately. See Ask Overdesk in a note.

Otherwise a conversation has no cap: each new message from a customer can get its own reply for as long as the conversation runs, as long as every check above passes.

After a reply ​

After it sends, Overdesk sets the conversation to the status you choose for each help desk, on Integrations under After Overdesk sends a reply, or in Settings → Sending. The choices for each help desk are in the table under Where Overdesk sends. Pending, the start for every help desk but Intercom, leaves the conversation waiting for the customer's answer, and Leave as is keeps whatever status the conversation had.

  • On Zendesk, Solved and On-hold are not offered. Solved would assign the ticket to the agent Overdesk replies as, and a Zendesk account can switch On-hold off. A workspace that had one of them chosen sends with Pending.
  • On Intercom, Overdesk closes the conversation as a second step after the reply. If the customer writes again, Intercom reopens it. In our testing, closing a conversation this way did not send the customer a rating request.
  • A reply Overdesk hands over after sending, because it passed the question on or said the customer will hear back, leaves the conversation in your team's queue whatever you chose, because your team owes the customer the next step. That is Active on Help Scout and FreeScout, and Open on Zendesk and Intercom (see Escalation).
  • On Intercom, the reply itself assigns the conversation to the teammate your access token belongs to, and Overdesk then assigns it on to the person who keeps the promise. When your team has dealt with a conversation Overdesk left with it, reply or add a note to the conversation to hand it back.

A line at the end of each sent reply saying AI helped write it is available under Add a line saying AI helped write the reply in Settings. It is off until you turn it on. Some regions require telling customers when AI helped write a message.

A reply your team sends from the help desk is unchanged by any of this.

Seeing what was sent ​

  • On Agent drafts, a reply Overdesk sent carries the Answered automatically badge, and its page says Overdesk sent it after your guardrails approved it. A nudge has no page there: its decision shows on Guardrails → Decisions only.
  • On Guardrails → Decisions, every reply's decision is listed with the rule that made it. See Decisions.
  • When a send ends without a clear answer from your help desk, for example because the request timed out, Overdesk reads the conversation. If its reply is there, word for word, it counts as sent, as if the help desk had answered. If it isn't, or Overdesk can't tell, your help desk didn't confirm the reply: Overdesk leaves a note in the conversation asking your team to check before replying, and the reply's page on Agent drafts says the customer may already have it. Overdesk sends nothing more there until someone on your team adds a note or replies (on Intercom, a note), and each new customer message gets a draft for your team instead.
  • A reply Overdesk sent is never read back as your team's own words: it does not become a past answer in your knowledge base, it is not graded as a reply your team wrote, the reviewer that checks a later reply in the same conversation reads it as Overdesk's, not your team's, and analyses, Weekly intelligence and Response quality leave it out, so what they find is about your team's work. This holds after the draft ages out, because Overdesk keeps a record of the reply without its text (see Data retention). Response times still count it, because your customer got a reply.