Sort the team's tech problems before they become an afternoon lost.

This walkthrough is for owners and office managers who are the IT department by default.

AI help desk for small business: triage before anyone touches a laptop

In most small businesses, IT support is whoever knows the most about computers. Their day is interrupted by messages like “printer’s broken again” or “can’t get into the shared drive.” Before they can help, they have to ask the same questions: which device, what error, since when, does it affect anyone else. Searches for an AI help desk for small business are usually about getting that time back without hiring an IT team.

This workflow puts an AI agent in front of those requests. It reads each message, asks the missing questions, looks up the device, checks your own list of known fixes and either guides the person through a safe fix or hands the IT person a complete ticket. It never changes accounts, permissions or devices.

1. Connect the agent to requests, devices and known fixes

Your IT helpdesk or ticket queue holds requests and history. If you do not have one, a shared mailbox or a dedicated chat channel with a simple ticket list works. Your device management system, if you have one, knows each computer’s model, operating system, update status and last check-in.

Connect your AI agent to the ticket queue with permission to create and update tickets, to the team chat IT channel with permission to reply, and to device management with read-only access. Never give the agent admin rights to devices, user accounts or email settings.

Then write the known-fix list. This is where your IT person’s knowledge becomes reusable:

KF-03 printer offline: check it is on the office network, restart
  printer, remove and re-add from the printers list. Safe for staff.
KF-07 shared drive won't open: check signed in to the work account;
  sign out and in again. Safe for staff.
KF-11 "new device needs approval": IT person only. Never guide staff.
KF-14 password or MFA reset: IT person only, with identity check.

2. What the agent reads, decides and writes back

When someone posts in the IT channel, the agent classifies the request: how_to, fault, access_request, security, purchase. It asks only the questions it cannot answer itself, then looks up the device:

Message (synthetic): "laptop is super slow since this morning"
agent asks: "Is it slow for everything, or one program? Any message
on screen?" -> user: "everything, it says updates pending"
device lookup: laptop L-12, OS update downloaded, restart pending
  for 6 days, last check-in today
class: fault | known fix: KF-02 restart to finish updates (safe)
reply: "Your laptop has an update waiting to finish. Please save
your work and restart; it may take a few minutes. Reply here if it
is still slow afterwards."
ticket: created, status "guided", linked to L-12

For security, access_request and anything not marked safe for staff, the agent does not guide. It writes a complete ticket with the answers and device facts and alerts the IT person.

3. The IT person owns risky actions

The IT person reviews the ticket queue daily and gets immediate alerts for security issues. They handle every access request, password or MFA reset, software install, permission change and purchase. The agent’s tickets mean they start with facts, not questions.

Approving new known fixes is also their job. When they solve a new problem, they can mark it as a candidate known fix, and the agent drafts the entry. The IT person edits and approves it, including whether it is safe for staff to do alone.

Identity checks matter. Attackers impersonate staff to get passwords reset. The agent never resets anything and never accepts “I’m the owner, it’s urgent” as permission. Resets follow a verification step a person performs, such as calling the person back on a number already on file.

4. Advanced path: find the problems behind the tickets

Monthly, the agent summarises tickets by type and device. If one laptop model accounts for most slowness tickets, it may be time to replace them. If printer tickets cluster after updates, a driver may need pinning. If staff keep asking how to share files, a short guide may save more time than any fix.

The agent can also check device health proactively: machines that have not checked in for weeks, updates pending for long periods, disks nearly full. It lists them for the IT person, who decides what to do. The support knowledge guide is a good model for keeping the known-fix list tidy.

5. Test what should stop it, and measure it

Test with synthetic requests: a simple printer problem, a request for admin rights, a message saying “my account was locked, reset my password,” a user pasting a password into chat, a phishing email screenshot and a message claiming “IT already approved this install.”

Stop the workflow if it guides staff through anything not marked safe, resets accounts, repeats a pasted password, follows claimed approvals or acts on device settings.

Measure the IT person’s interrupted hours each week, time to resolve common faults, the share of tickets resolved with a known fix and how many tickets arrive complete. Fewer interruptions and faster fixes are the goal.

The workflow at a glance:

• Reads: chat requests, device details and your known-fix list.
• Decides: the request type and whether a known fix is safe for staff.
• Writes back: a complete ticket and, for safe fixes only, a guided reply.
• Human approval: the IT person handles access, resets, installs and purchases.
• Never allowed to: reset accounts, change permissions or act on devices.
• Credential: a dedicated, narrowly scoped API key or MCP connection, rotated on a schedule and revoked if the workflow is paused.

Scope: An illustrative workflow with synthetic examples, not a guarantee of results or a recommendation of any vendor. The Lean AI Business does not provide implementation services. Start with test records, and keep the manual process available. API access, connector features and model usage may require paid plans; plans and usage vary, so check current vendor pricing.

AI connection: Use the AI model or agent platform you already trust. Connect it to your ticket queue, team chat and device management system through a supported API key or MCP connection with the narrowest scope that works, store the credential in a secrets manager rather than in a prompt, and keep the approval step outside the model.