Home/Resources/What is AI IT support

Guide

What is AI IT support? A plain-English guide for small businesses

What it actually automates, what it can't touch, what it costs, and how to tell a real implementation from marketing language.

Short answer

AI IT support uses automation wired directly into your identity, device and cloud platforms to resolve routine requests without a technician — password resets, new starters, leavers, licence changes, software installs, patching. Anything it can't resolve escalates to a human. Because automation absorbs the highest-volume work, it's typically priced at $20–$50 per user per month against the $150–$200 a traditional provider charges. The saving is structural, not a discount, and it only applies to the automatable portion of the work.

Why small companies can't get IT support at all

Traditional managed IT costs $100–$300 per user per month, with a small-business average around $150–$200. Providers typically set minimums of 10 to 25 seats, and the per-user price tends to rise as the company gets smaller.

That looks backwards until you see the cost structure. A human helpdesk is a fixed cost. Spread across 200 seats it's affordable; across 10 it isn't. Onboarding a new client — discovering the environment, hardening it, deploying agents, documenting everything — takes roughly the same effort whether the company has 8 people or 80. Below a certain size the account never pays back the setup, which is why the industry has a widely discussed "small client problem" and why so many providers quietly decline sub-15-seat enquiries.

The result is a large, genuinely underserved segment: companies with 5 to 25 staff, no internal IT, and no realistic provider option. They end up with a break-fix number, a friend-of-the-owner, or nothing.

What AI actually automates

This is where specificity matters, because "AI-powered" is applied to everything from genuine autonomous action to a chatbot that reads a help article at you. The meaningful distinction is whether the system can take actions in your environment or merely answer questions about it.

Action-taking automation, connected through an API such as Microsoft Graph, reliably handles:

Identity & access

  • Password resets and MFA re-registration
  • Account lockouts
  • Group, Teams and shared mailbox access
  • File and folder permissions
  • Guest accounts with expiry

Joiners & leavers

  • Full new-starter provisioning
  • Licence assignment, mailbox setup
  • Device enrolment and policy
  • Offboarding and session revocation
  • Data reassignment

Devices & software

  • OS and application patching
  • Installs from an approved catalogue
  • Encryption and policy enforcement
  • Health monitoring with auto-remediation
  • Remote lock and wipe

Cost control

  • Reclaiming licences from departed staff
  • Finding duplicate subscriptions
  • Flagging over-specified licence tiers
  • Usage reporting

These share three properties: high volume, well-defined outcomes, and reachable through an API. That combination is what makes automation reliable rather than risky.

What it cannot do

Any provider unwilling to give you this list is selling you a disappointment.

  • Anything physical. Dead laptops, failed drives, swollen batteries, printers, cabling, Wi-Fi dead spots, a server in a cupboard. Roughly 20 per cent of support tickets involve hardware, and no amount of model quality changes that.
  • Line-of-business applications with no API. The practice management system, the accounting package, the industry tool from 2014 running on a PC under someone's desk. The automation is blind to these.
  • Vendor coordination. Sitting on hold with an ISP, chasing a telecoms port, escalating a case with a software vendor. Someone has to own that and be accountable for it.
  • Judgement calls. Security incident response, procurement decisions, compliance interpretation, anything where being confidently wrong is expensive.
  • Emotional labour. At 15 employees, a meaningful part of "IT support" is reassurance and patience. A product does not absorb blame.

What resolution rate should you expect?

Published figures vary widely, and the honest answer is that it depends almost entirely on your environment.

Source of figureAutonomous resolution
Vendor-reported, general ticket automation10–25%
Industry average, action-taking agents~45%
Mature, well-governed deployments55–65%
Fully cloud-based, no servers, modern devicesHigher end
On-premise servers and legacy applicationsMuch lower

Be sceptical of anyone quoting 90 per cent without qualifying it. That number is only reachable by scoping out the categories that can't be automated — which is legitimate, but only if the exclusions are written into the contract rather than discovered later.

The security questions to ask

You are handing an automated system privileged access to the platform your business runs on. Four questions separate a serious implementation from a risky one:

  1. What exact permissions does it hold, and can I revoke them myself? Access you can't withdraw unilaterally isn't access you control.
  2. What requires human approval? Deleting accounts, changing security policy and removing data should never run autonomously.
  3. Can I see an audit trail? Every action, with what changed and what approved it, readable and exportable by you.
  4. Is my data used to train models? Get the answer in writing. Consumer-tier AI services routinely retain and train on submitted data.

How we answer these four questions →

What it costs

ModelPer user/monthBest for
AI-led support$20–$50Cloud-first companies, 5–40 staff, no servers
Managed IT with AI underneath$80–$12010–75 staff needing a real helpdesk
Traditional managed IT$150–$200Complex or on-premise environments
Full-service with 24/7 and on-site$200–$300Regulated industries, zero-downtime tolerance

Watch for two things. First, whether software licences are bundled — many providers fold Microsoft 365 into a single per-user rate, which hides both the true support cost and the licence markup. Second, whether onboarding is charged separately. A provider with no setup fee has either buried it in your monthly rate or isn't doing the work.

Full cost breakdown, including what onboarding should cost →

Is it right for you?

Good fit

  • 5–40 staff, no internal IT
  • Fully or mostly on Microsoft 365 or Google Workspace
  • Laptops rather than desktops and servers
  • Distributed or hybrid team
  • Frustrated by not being able to get quoted at all

Poor fit

  • Significant on-premise infrastructure
  • A critical line-of-business app with no API
  • Regulated with audit obligations needing provider certification
  • A culture that expects to phone a person for everything
  • Frequent on-site needs

About the author

Raymond Payne is the founder of PCR — Managed IT. Thirty years building and running technology for businesses across more than a dozen countries, and author of Mastering Automation with AI. Last updated 14 August 2026.

Related reading

What managed IT should cost in 2026

Real price ranges by company size, what's usually hidden, and how to compare quotes.

AI-led support vs a traditional MSP

Where each genuinely wins, with the trade-offs stated plainly.

How to switch IT providers

The contract traps to check before you give notice.

Small business IT security checklist

Twelve things to fix, most of them free.

See what this costs for your team

Pick a tier, set your headcount, see the exact monthly and first-payment totals. No quote request required.

From $29/user/monthMinimum 5 users · no bundled licences
See your price