First 90 days at a 25-user accountancy practice

The situation

A 25-person accountancy practice in the East Midlands — a mix of qualified accountants, bookkeepers, and admin staff — had been with the same managed service provider for four years. Their MSP contract ran to £1,800 a month (roughly £72 per user). Not terrible, but the practice manager had started tracking how long ticket resolution actually took. On average: 3.4 hours for anything that required a human response, and over a day for anything that needed an engineer visit.

The team worked almost entirely in Microsoft 365 — Outlook, Teams, SharePoint, and Excel. Their infrastructure was standard: cloud identity via Azure AD, no on-premise servers to speak of. The practice manager read about AI-first support and was sceptical, but the economics were hard to ignore. She signed up for a trial with five of the firm's staff before committing the full headcount.

What we rolled out

Week one was agent deployment. The lightweight agent took under five minutes per machine to install — pushed silently to the five trial machines via the practice's existing M365 Intune setup, so there was no disruption to the working day. The practice manager received a confirmation when each device checked in.

In the first week, the AI ran baseline diagnostics across all five machines: disk health, pending Windows updates, outstanding M365 licence assignments, and MFA enrolment status. It flagged two machines with outdated drivers and one user who hadn't completed MFA setup. All three were resolved before they caused a support ticket.

By week three, confidence was high enough to roll out the full 25 users. The agent deployed to the remaining 20 machines over a Monday night. Tuesday morning, all 25 were active with no reported issues.

What the AI handled

In a typical 90-day window for a firm this size on M365, we'd expect the AI to resolve around 85–90% of tickets without human involvement. For this practice profile, the common ticket types resolved entirely by AI included:

• Outlook configuration issues — shared mailbox access problems, signature formatting, auto-reply conflicts. The AI cross-references the M365 admin tenant and fixes permissions directly. Typical resolution: under two minutes.
• Password resets and MFA issues — locked accounts, lost authenticator app setups, staff returning from leave who'd been locked out. The AI verifies identity via a short challenge flow and resets without a ticket queue.
• OneDrive and SharePoint sync errors — the most common complaint in M365 environments. The AI reads the sync client logs directly from the endpoint agent, identifies the conflict, and resolves it. Staff typically see the fix happen before they've finished describing the problem.
• Performance issues — "my laptop feels slow" is the most common vague support request. The AI checks CPU load, memory pressure, disk I/O, and running processes. In most cases it identifies a specific culprit — Teams caching, OneDrive indexing, a Windows update running in the background — and either resolves it or gives the user a concrete explanation.
• Printer and peripheral problems — driver conflicts, queue jams, USB device not recognised. Resolved via the endpoint agent without the user needing to explain their setup.

In a typical 90-day window, a 25-user accountancy firm generates roughly 60–80 support interactions. The AI would handle around 55–68 of these without escalation.

Where humans stepped in

Around 10–15% of interactions in this type of environment reach a UK engineer. For an accountancy practice, the most common escalation types are:

• M365 licensing and tenant configuration — when a new joiner needed a licence assignment that involved changing a distribution group membership and updating a SharePoint permission set simultaneously, the AI flagged this as a multi-step tenant change and handed to an engineer with the proposed configuration already drafted. The engineer reviewed and applied it. Turnaround: under 30 minutes.
• Specialist accounting software — the practice used a third-party accounts production package that sat outside standard M365. When that software errored, the AI recognised it wasn't a Windows or M365 issue, flagged the error code, and escalated with the vendor's known issue list pre-attached. An engineer called the vendor's support line on the firm's behalf.
• Ambiguous "everything feels wrong" tickets — one senior accountant reported her machine feeling generally sluggish and "not right" after a software update. The AI diagnostics didn't find a single clear cause. An engineer picked up the call, ran a remote session, and identified a background process conflict between two audit tools. Resolved in 45 minutes.

All escalations went to an engineer with the full AI conversation and endpoint diagnostics already attached. No context was lost, and no one had to re-explain the problem.

Outcome

In a typical 90-day window for a firm at this scale and profile, we'd expect:

• Cost reduction of around 85% — from the £72/user/month MSP rate to £10/user/month. For a 25-user firm, that's moving from £1,800/month to £250/month.
• Average resolution time under 4 minutes for AI-handled tickets, compared to the 3.4 hours the practice had measured under the previous MSP.
• Zero dropped tickets — every interaction is logged, escalated if needed, and followed up. The practice manager no longer needs to chase the MSP to find out what happened to a ticket from last Thursday.
• Less stress at year-end — during the January self-assessment rush, when all 25 users are simultaneously under pressure, the AI absorbs the volume spike without any change in response time. No "we're experiencing high volumes" message.

The less measurable outcome: the practice manager stopped thinking about IT. That's the actual goal. IT support should be invisible — something that works, not something you manage.