Auto-resolve

Auto-resolve lets Aura AI reply to a ticket and mark it solved on its own — when it is confident, grounded, and there is nothing left open. It is at Settings → AI agents → Auto-resolve.

It is off by default and every guardrail fails closed, which means the honest summary is that turning the switch on is not enough to make anything happen. You have to also decide what it is allowed to touch.

Reading needs nothing. Changing needs admin or owner. This screen is a single form with an explicit Save changes, and a Discard changes beside it while you have unsaved edits.

It replies through your Aura AI agent

The banner at the top reports the Aura AI connection, because that is where the replies come from: auto-resolve, web chat and Agent Assist all answer through the same agent. There is deliberately no agent field on this screen — if the workspace has no agent connected, every control here is disabled, admin or not, and the page links you to Aura AI to connect one.

The settings

Auto-resolve — the switch. When on, a reply that scores above the threshold, cites the knowledge base and leaves nothing open is sent to the customer and the ticket is marked solved automatically. Turning it off stops new AI replies from the next inbound message onward; replies already sent are not retracted.

Confidence threshold — a number between 0.50 and 1.00, default 0.70. A second model scores every candidate reply, and only a score at or above this line is even considered for solving. Score alone is never enough: the reply must also cite the knowledge base and leave nothing open. Setting the threshold to 1.00 makes solving effectively impossible, which is a legitimate way to pause the feature without losing your configuration.

Allowlist — two halves, and both must be satisfied.

• Sources: exactly four are offered — Email, Contact form, Web widget, API. These are the sources with two-way traffic today; every other channel is a no-op for auto-resolve whatever you set.
• Tags: a list you build. A ticket qualifies only if it carries at least one of them.

Nothing is allowlisted by default, and an empty allowlist means auto-resolve never fires, whatever the switch says. This is the single most common reason a workspace turns auto-resolve on and sees nothing change. Start with one tag on one source.

Daily call cap — default 200, between 1 and 100000. The most attempts auto-resolve will make in a workspace per day. Once reached, everything further routes straight to a human until the cap resets. Treat it as a blast radius rather than a budget: a misconfigured rule costs you at most this many automatic replies before it stops itself.

Escalate on pushback — always on

There is no switch for this one, and the page says so.

When a customer replies to a ticket auto-resolve solved and the reply reads as unsatisfied, the ticket reopens, routes to a human, and is permanently marked never-again-auto-resolvable. That ticket never gets a second automatic reply, however good a later score is.

That permanence is the point. A customer who has already been failed by an automatic answer does not get a second one; they get a person.

What the customer sees, and what you see

Every automatic reply carries an appended disclosure line telling the customer it was sent automatically by an AI assistant and inviting them to reply if they still need help. It is not configurable and it is not removable.

On the ticket timeline the message is marked Auto-resolved by Aura, and Auto-resolve details expands to show the model, the confidence score and the citations behind it. You can find every automatic reply after the fact and see exactly why it went out.

Turning it on sensibly

The configuration invites a narrow first pass, and it is worth taking it. Pick the one tag you would happily let a new starter answer unsupervised — password resets, opening-hours questions, order status. Allow that tag on one source. Leave the threshold at 0.70. Watch the AI-resolved view in the inbox for a week. Widen from there.