IT support firms on email: sort every request and get the urgent ones to an engineer fast

Why support still arrives by email

Many small firms offer a portal or a ticketing system, and clients email anyway. Email is what they have open, it lets them attach a screenshot, and it feels more personal than a form. For the owner of a ten-person accountancy practice, "email Dave's team" is the support process.

That is fine, as long as nothing gets lost. The risk with email support is not the volume, it is the order. Emails are read top to bottom, so an outage reported at 9:02 can sit behind password questions that arrived at 9:00. An AI agent reads everything as it lands and sorts it by urgency, so an engineer sees the outage first.

Triage first, answers second

The most valuable thing the agent does is not answering. It is deciding what kind of email each one is. Give it simple categories in your own words, such as: urgent (several people affected, a security worry, or the business cannot work), normal (one person, a workaround exists) and how-to (a question your guides already answer).

For urgent emails, the agent replies to say an engineer has been alerted and asks one or two quick questions that help, then hands the email over at once. For normal emails, it asks for the details an engineer will need, such as the device, what the screen says and when it started. For how-to emails, it answers from your guides and offers a person if the steps do not work.

Security questions are never the agent's call

Clients often forward an email and ask "is this real?". The agent must never say yes. An AI agent cannot be sure a message is safe, and a wrong "it looks fine" is exactly what an attacker hopes for. Tell it to reply with your standard safety wording, such as "Please don't click any links or open attachments in that email until one of our engineers has checked it", and to hand the email to an engineer as urgent.

The same goes for anyone who says they clicked a link, entered a password or saw a ransom message. Those are urgent by definition. The agent's job is to calm the person, give your first-steps wording if you have written it, and get a human on it quickly.

What customers ask on Email

• How do I set up my email on my new phone?
• The printer on the second floor has stopped working.
• Nobody in the office can get on the Wi-Fi.
• I got an email from "Microsoft" asking me to confirm my password. Is it real?
• Can you set up a laptop for our new starter on Monday?
• I've forgotten my password and I'm locked out.
• Why is our invoice higher this month?

An example conversation

This illustration shows an urgent email being spotted and handed over within a minute. Names and details are examples.

What the agent answers

• Sorting every email into urgent, normal and how-to, using your own definitions.
• Replying straight away to urgent emails to say an engineer has been alerted, then handing them over.
• How-to questions your guides already answer, such as setting up email on a phone or connecting to a printer.
• Collecting the details an engineer needs: the device, the error message, a screenshot and when it started.
• New starter and leaver requests, by asking for the name, start date and what access they need.
• Confirming your support hours and how to reach the on-call engineer out of hours.

What it always hands to you

• Anything affecting several people, or stopping the business from working.
• Every security worry: suspicious emails, clicked links, entered passwords and unusual files.
• Password resets and any change to accounts or access, which a person must verify.
• Anything that needs an engineer to log in, visit or change a system.
• Invoices, contract questions and requests for new services or quotes.
• Complaints and any client who has emailed about the same problem before.

Never let the agent verify identity

Password reset and access requests are a favourite route for attackers, who pretend to be a member of staff. Tell the agent never to reset, unlock or change anything, and never to treat "I am the managing director" as proof. It collects the request, says an engineer will verify it using your normal checks, and hands it over. Your engineers then verify the person the way they always do, for example by calling them on a number you already hold.

The Email limits that matter here

A new chat here is a new email thread. A reply to an email that arrived in the last week is never held back by the daily limits. Spaces sends one email at a time to one person, never bulk email.

Set it up in an afternoon

1. Write a support note: your urgency categories, your support hours, the on-call arrangement, the how-to guides the agent can share, your safety wording and the topics that always go to an engineer. There is an example below.
2. Connect your support mailbox in Spaces. You sign in to Gmail or Outlook through a secure link and approve the connection, or enter your mail server details for any other mailbox. Use the address clients already email, such as support@, so nothing changes for them.
3. Connect your AI agent using the steps on our agents page. Chat apps such as Claude or ChatGPT reply when you ask them in a chat, so for replies through the day without you asking, use an always-on agent such as Claude Code, Codex, OpenClaw or Hermes. Give it the support note. Spaces does not connect to ticketing, monitoring or remote access tools, so the agent does not log tickets in them or touch client systems. If you want it to create tickets, your AI agent needs its own access to your ticketing tool, which you set up and control separately.
4. Decide how the agent alerts an engineer about urgent emails. Many firms have it tag the email and send a short summary to the on-call engineer's own address, so it is seen even if they are away from the support inbox.
5. Decide when the agent answers. If you offer support only in office hours, tell it in its notes to reply out of hours with your out-of-hours wording and the on-call route, or pause sending overnight.
6. Ask your AI agent to hold replies as drafts for the first week or two, and nothing goes out until you approve it. Read each draft, fix the wording you would never use and add the fixes to the note. Once the drafts need almost no changes, let the routine replies go out on their own and keep drafts for anything sensitive.
7. After a month, look at which how-to questions come up most and write a proper guide for each. Every guide you add is one fewer email an engineer has to answer.

A sample support note

Keep the urgency rules at the top, because they matter most. Here is the shape of one, with example details.

FIRM: Northgate IT. Support hours Mon to Fri 8:00 to 18:00.
Out of hours: urgent only, on-call engineer, call 01234 567890.

URGENT (reply at once, then hand to on-call):
- More than one person affected, or the business cannot work.
- Any security worry: suspicious email, clicked link, password entered,
  strange files, ransom message, lost or stolen laptop.
SAFETY WORDING: "Please don't click any links, open attachments or turn
off the computer until one of our engineers has spoken to you."

NORMAL (ask for details, then hand over):
- Ask: which device, what the screen says, when it started, a screenshot.

HOW-TO (answer from the guides list only):
- Email on a phone, printers, Teams set-up, connecting to the VPN.

NEVER: reset passwords, unlock accounts, change access, say an email is
safe, promise a time an engineer will arrive, discuss invoices.

Before and after: the forgotten password

Here is how the same request reads with and without a note. Names are examples.

The first reply sounds helpful and is dangerous. The agent cannot reset a password through Spaces in any case, but a reply that promises one, or invents one, teaches clients to expect the wrong thing. The second keeps the verification with a person and still answers quickly.

What if a client emails the agent about something it cannot do?

Clients will ask the agent to "just remote in and fix it". The note should give it a kind, plain answer: it cannot access systems, an engineer will pick this up, and here is when. A short, honest reply is better than a vague one. Clients mind waiting far less when they know a person has the problem and roughly when they will hear back.

If a client asks for something outside your contract, such as new hardware or a project, the agent collects what they want and hands it to whoever writes quotes. It should never quote a price or say something is included.

Common mistakes IT support firms make

• Letting the agent comment on whether a suspicious email looks genuine.
• Giving it a vague idea of "urgent", so outages are treated as normal requests.
• Sending urgent alerts only to the support inbox, which nobody watches while they are on site.
• Allowing it to promise arrival times or fixes it cannot know.
• Sending the same "we have upgraded our service" email to every client from the agent. Use your newsletter tool for announcements and let the agent answer replies.

Who you message

In short

For an IT support firm, the biggest win from an AI agent on email is order. Urgent emails are spotted and handed to an engineer at once, how-to questions are answered from your guides and everything else arrives with the details an engineer needs. Security worries, password resets and any change to a system always go to a person.

Write clear urgency rules, decide how the on-call engineer is alerted and start with drafts. Spaces paces every email from your support mailbox, but the agent is only as good as the note you give it and the engineer who picks up what it hands over.

Pricing

Plans from $19 a month

Can the agent log tickets in our ticketing system?

Spaces does not connect to ticketing tools. Your AI agent can reach one through its own access if you set that up, and many firms do. Without it, the agent tags and hands over emails, and an engineer logs the ticket as usual.

Can it reset passwords or unlock accounts?

No. Spaces only gives the agent your mailbox, not your clients' systems, and the note should tell it never to promise a reset. A person verifies the request using your normal checks and does the reset themselves.

How does it know what is urgent?

From the rules you write. Make them concrete, such as "more than one person affected" or "any mention of a clicked link". Review the handed-over emails in the first few weeks and tighten the rules when it gets one wrong.

Will it tell clients whether a suspicious email is safe?

It should never do that. It sends your safety wording, asks the client not to click anything, and hands the email to an engineer as urgent. Only a person who has checked the email should tell the client what it is.

What about out-of-hours emails?

Tell the agent in its notes what to do outside your support hours. Most firms have it reply with the out-of-hours wording and the on-call route for genuine emergencies, and leave everything else for the morning.

Can it send clients a monthly update or newsletter?

Not as a mass mailing. Spaces sends one email at a time to one person and does not do bulk email. Use your newsletter tool for updates, and let the agent answer the replies one by one.

Does it read attachments such as screenshots?

Spaces lets the agent list and download attachments on an email, so an agent that can read images can look at a screenshot of an error. It should still pass the details to an engineer rather than diagnose a fault on its own.

Should clients know an AI agent is replying?

Yes. A short line in each reply saying that an AI agent sorts incoming emails and engineers handle every fix keeps things clear. Clients mostly care that the urgent problem reached a person quickly.