Reads the mailbox. Replies, or hands it over.
Reads every incoming enquiry, classifies it, links it to the correct customer and job, sends routine replies, and escalates unusual cases. Connect a mailbox and the agent works the new mail on a short cycle, replying where your rules allow and escalating the rest.
- Mailboxes
- Google, Yahoo and Microsoft Outlook
- Works in
- Customer Support - Email ( Customer Facing)
- Runs on
- Customer 360 and Workorder Management
- Setup
- A mailbox, a tone and an optional reference document
what it is for
Most of a service mailbox is the same five emails.
A booking. A reschedule. A question the answer to which is already written down. A message that needs a person. Each one still has to be opened, understood, matched to the right customer and appointment, and answered. That reading and matching is most of the work and almost none of the judgement, and it is the part that stops overnight and at weekends.
AI Inbox Triage connects to the mailbox you already use and works the new mail on a short, regular cycle. It decides what each message is asking for, answers the ones it is allowed to answer, books or moves the appointment where it has enough detail, and escalates anything it cannot place. Every reply goes out in the tone you set.
One message, start to finish
Checked on a cycle
Every active mailbox is checked on a short interval, all of them in parallel, so one slow mailbox never holds up the others.
Escalates before guessing
Low confidence, an unrecognised sender, or setup you have not finished yet all go to a person instead of a guess.
Never silently dropped
A message counts as handled only once a reply has actually gone out. If anything fails first, it is left for the next check.
connect the mailbox
Your mailbox, your provider, your tone.
Setup is a mailbox and a handful of choices. Nothing is assumed for you, and an agent without a mailbox stays switched off rather than running on something it should not.
Google, Yahoo or Microsoft Outlook. The agent monitors that one mailbox; it does not need a new address, and it does not touch mail that arrived before you connected it.
You connect your own provider and test the connection before anything goes live. The same agent logic runs whichever one you pick, and it retries by itself if the provider is briefly overloaded.
Friendly, casual, formal or professional, chosen per agent. Every reply is generated in that tone rather than in a house voice you did not pick.
Save an agent without a mailbox and it is stored, not started. It only begins reading once a mailbox is connected and you activate it.
Agent setup
work the new mail
Every new email takes the same path.
Each message runs through one decision process, and the safeguards matter more than the speed: the same mailbox is never checked twice at once, and the same email is never processed twice.
Unread messages that arrived after the mailbox was connected. If a check overlaps the one before it, or the system restarts mid-cycle, a message is still only handled once.
Mail from no-reply and system addresses is passed over rather than answered, because a reply to it goes nowhere.
The existing email conversation is used to work out which appointment is meant. Where that match cannot be made confidently, it is escalated rather than applied to the wrong booking.
A new appointment request missing what is wrong, where, or when gets a reply asking for the missing detail. The appointment is only created once the detail is there, and the customer gets the appointment number back.
This cycle
what it will not do
When it is not sure, it does not guess.
The escalation rules are the point of the agent, not a caveat on it. Anything it is not confident about, and anything that plainly needs judgement, goes to a person with the message attached.
Knowledge questions are answered strictly from the reference document you uploaded. If nothing relevant is in it, the question is escalated rather than improvised from general knowledge.
Where the sender cannot be matched to a customer you hold, the message goes to a person rather than being answered on assumption.
If the configuration it would need is not finished, it says so by escalating, instead of running on defaults you never chose.
Every generated reply is checked before it is sent, including that a confirmation carries the right appointment number. If the check fails, or the provider call fails, a pre-written fallback goes out instead.
What it may answer
how it works
Four steps, then it hands on.
The same path runs on every message, and the last step is the only one that replies.
-
ingests
Connects to inbound communication streams to ingest emails, portal tickets, and service requests the moment they arrive.
-
classifies
Evaluates message intent — identifying whether the communication is a new service request, cancellation, reschedule, complaint, or billing enquiry.
-
links
Matches the sender against existing CRM accounts, service sites, and active job records to maintain a unified communication thread.
-
executes
Triggers approved routine actions — acknowledging the request, drafting standard responses, or creating follow-up tickets for operations.
where the line sits
What it handles, and what it hands over.
Each kind of message has one outcome, and the escalations are as deliberate as the replies.
| Message | What the agent does | Who finishes it |
|---|---|---|
| Appointment request, complete | Creates the appointment and confirms it, with the appointment number in the reply. | Dispatcher schedules |
| Appointment request, incomplete | Replies asking for what is wrong, where, or when, and waits before creating anything. | Nobody, until answered |
| Reschedule | Identifies the appointment from the conversation, moves it, and confirms. | Dispatcher confirms |
| Question covered by your document | Answers using only the relevant part of that document. | Nobody |
| Question it does not cover | Escalates rather than improvising an answer. | Named owner |
| Unknown sender, or unclear ask | Escalates with the message attached, instead of guessing. | Named owner |
you set
Before it runs its first job, three things come from you.
- Connected inbound email inbox or API communication gateway.
- Approved routing criteria and thresholds for automated routine actions.
- Rules defining when an enquiry must escalate directly to human customer service.
None of it is a data migration. It reads the Customer Support - Email ( Customer Facing) records you already keep.
stays with your team
AI Inbox Triage executes strictly approved routine actions. Non-standard enquiries, sensitive escalations, and complex complaints are flagged immediately for human review.
An escalation arrives as the original message, so the person picking it up reads what the customer actually wrote rather than a summary of it.
rollout
Connect one mailbox and watch a cycle.
Most teams do not switch every action on at once, and nothing here requires it. The usual order:
Connect and watch
Connect a mailbox and read what the agent decides on live mail before you let it answer. It only ever looks at mail that arrives after you connect it.
Add the document
Upload the one reference document you want it to answer from. It answers strictly from that, so what you upload is exactly what it can say.
Then let it reply
Turn on acknowledgements first, then bookings, then reschedules. Anything you have not turned on is an escalation, not a judgement call.
works with
What it runs on, and what it hands to.
read next
Where the mailbox meets the rest of the day.
before you switch it on
What teams ask about AI inbox triage.
Only inside the rules you set. Anything outside them - listed under "you set" above - goes to a named person on your team, with the record attached. There is no state where it silently makes a judgement call you have not already approved.
Days, not months. It reads from the records you already keep in Customer Support - Email ( Customer Facing), so there is no separate data migration or new system for your team to learn before it can start.
Yes, instantly, and every action it took stays on the record so you can see exactly what it did and reverse it. Turning it off does not remove or hide anything it has already written.
No. It takes the routine, repeatable part of the job off their desk so they spend their time on the exceptions and judgement calls that actually need a person - the work listed under "stays with your team" above.
AI inbox triage runs on your Customer Support - Email ( Customer Facing) records, alongside Customer 360 and Workorder Management, so nothing needs re-keying between systems.
Mailboxes: Google, Yahoo and Microsoft Outlook. For the AI behind it you connect your own provider and test that connection before the agent runs. The agent logic is the same whichever you choose, and your credentials are stored securely rather than shared.
One reference document per agent, uploaded as a PDF. It searches that document and uses only the relevant part of it to write an answer. If the document does not cover the question, the question is escalated rather than answered from general knowledge.
A generated reply is checked before it is sent, including that a confirmation carries the correct appointment number. If that check fails, or the provider call fails outright, a pre-written fallback is sent, so the customer is never left without a response. The message is only marked handled once a reply has actually gone.
Point it at your own mailbox.
Book a demo
We will connect a test mailbox, set the tone, and show you a cycle running on real mail.
Book a demoSee the other nine
AI inbox triage is one of ten agents on one job record, from first enquiry to paid invoice.
All AI agentsSee the platform
The field service system underneath: scheduling, work orders, assets, invoicing and the field app.
AI Field Service Management