Skip to content

Decisions ​

The decision log records what your rules decided for every reply draft Overdesk writes, and for every follow-up nudge when your workspace sends them, and why. A nudge has three holds of its own: the customer's last message was not one Overdesk answers on its own, a teammate wrote in the conversation after Overdesk's reply, and the last reply told the customer your team would get back to them. They are listed below. Use it to see which rules fire on your real queue, and how often, before you hand any of them to a person or turn sending on.

Where it lives ​

Open Guardrails → Decisions. The Decision log lists the newest decisions first, and Load more at the bottom reaches older ones. Everyone in the workspace can read it.

Reading a row ​

ColumnWhat it shows
ConversationThe conversation number, linked to the conversation in your help desk
DecisionDraft only when a rule held the draft or none approved it, Escalate when a rule under Escalate when matched, Automatic answer when a rule under Answer automatically when did
RuleThe rule that matched, by the name you gave it
ReasonThe rule's own sentence, such as "Billing requests always go to a person." When the reviewer left a reading of that particular draft, it follows on a second line
WhenHow long ago the decision was made

While sending is off, each decision carries a small grey label beside it. It marks the row as a record of what the rules would have done, and the draft it describes waits for a person like every other, including an Automatic answer: in Draft assist, that row records a draft the reviewer approved, and nothing was sent.

Once sending is on, a row without that label is a live decision. An Automatic answer there is a reply Overdesk sent, or tried to: if the help desk refused it, the draft was placed in your help desk for your team instead, and the draft's page on Agent drafts says so. A reply that one of the checks on Sending held shows Draft only, with that check as its reason.

When no rule picked a conversation out, the Rule column shows Overdesk's default entry. That is the ordinary result for a draft none of your rules matched.

These entries are Overdesk's own holds, on what a draft says, on what the customer's message is, on a draft Overdesk tried a test site on, on a reply that was allowed to ask the customer for their Site Health info, or on a reply that was cut off before it finished, and nothing in your workspace switches them off (how they work):

  • The reply says a person wrote it: the draft says a person wrote it, or denies that AI did.
  • The reply says it was tested, but no finished test backs it up: the draft claims a test with no finished test on the test site behind it. A test site that started and checked nothing is not a finished test.
  • Overdesk tried a test site on this draft, so a person checks it first: your rules would have answered the draft automatically, and Overdesk had tried a test site on it. Replies that rely on a test site are not sent automatically yet.
  • Overdesk let this reply ask for the customer's Site Health info, so a person checks it first: the reply was allowed to ask the customer for their Site Health info, whether or not it made the ask. Replies like this are not sent automatically yet.
  • The reply was cut off before it finished, so a person checks it first: the writing stopped before the reply was complete, so part of it may be missing. It is never sent automatically, as a reply or as a follow-up nudge, and waits for a person.
  • The reply talks about its own sources or caution: the draft talks about what it was given or found, or that it won't guess, instead of the customer's question.
  • Not a request for support: the customer's latest message is marketing, spam, an automatic notice, a bounce or an out-of-office reply. A nudge is held here too unless the screen read the customer's last message as a request for support, so with Screen out non-support mail off every nudge is held as this.
  • An automatic message, like an out-of-office reply: the latest message was sent by a machine, found from its sender and how it opens.
  • The customer asked for a person: held even when Human agent requested is off.
  • The customer's last message was not one Overdesk answers on its own: a follow-up nudge was held because Overdesk did not itself answer the customer's last message, or could not confirm that it did.
  • A teammate wrote in the conversation after Overdesk's reply: a follow-up nudge was held because someone on your team wrote there, a note or a reply, after Overdesk's reply, so it is waiting for a person. A nudge is only ever a follow-up to Overdesk's own reply. It is also the reason shown when Overdesk could not read which replies it sent itself.
  • The last reply told the customer your team would get back to them: a follow-up nudge was held because the last reply told the customer, in wording Overdesk recognises in English, that your team would get back to them, so it does not nag a customer who was asked to wait.
  • The reply promises something the team has to do: a date, a refund, passing the question to your team, getting back to the customer later, or something done that the conversation doesn't show. In a reply written in English, a second check, Overdesk's own and not the reviewer's, also looks for these promises (how). A reply whose only promise is passing the question on, or that the customer will hear back, is not held for this when it can be handed to the target it names or to the person who keeps the promises replies make. See Escalation. A reply asking the customer for access to their site is held for this too, and goes to the person you marked as logging in to customer sites (how).
  • The conversation changed while we were deciding: a reply or nudge was held because the conversation changed between Overdesk writing it and sending it (a new message, a reply or note from your team, a new status, a different customer, someone newly copied, a change to a tag your rules read, a channel Overdesk does not send on, or a move to another inbox), or because Overdesk could not read the conversation again to check. It is also shown when the conversation's inbox is not in Allowed inboxes, which is true for replies and nudges alike.
  • The reply repeats something from a teammate's note: the draft tells the customer something only your internal notes say.
  • The reply contradicts an earlier reply: the draft goes against what the customer was already told, by someone on your team or in a reply Overdesk sent.
  • The reply asks again for something the customer gave: the draft asks the customer for something they already gave, without saying why. A reply that gives a reason, such as an expired link or details that were wrong, is not held.
  • The reply has a link not found in your team's material: the draft links to an address that nothing your team gave Overdesk for this draft contains, such as your articles, saved replies or earlier replies. A link that text in a customer's screenshot or linked page asked for waits for a person this way.

Filtering ​

Two menus sit beside the heading. The first shows All rules or one rule on its own, which is how you check what a single rule has been deciding on your real queue. The second shows All decisions, or only Automatic answer, Draft only or Escalate.

Under each rule on the Rules tab, See what this rule decided opens this page already filtered to that rule. The filter stays in the address bar, so the link is worth keeping.

A rule you deleted keeps its rows here, and its name still shows on them.

What the log covers ​

  • Every reply draft, whether it came from the scheduled run, a real-time draft or a re-draft on the Agent drafts page. Only the scheduled run and real-time drafts can be sent; a re-draft never is.
  • One decision per draft, made when the draft is written. Changing a rule later does not rewrite earlier rows.

Follow-up nudges pass through the rules: one decision for each nudge on a workspace that sends nudges, except a nudge in a conversation a teammate forwarded, or on a workspace whose hourly limit is below 2, which is placed unsent and not decided here. Once the day's share of nudges is spent, the conversations still waiting are left for the next run, and nothing is decided for them. A workspace sends nudges when Also send follow-ups is on and sending is on. Without both, a nudge is placed unsent and is not decided here. A nudge has no draft of its own on Agent drafts, so its decision shows here only. See Follow-up nudges.

The same decision shows as a badge on the draft itself, on Agent drafts. See Reviewing drafts and Rules.