A 40-user consultancy that outgrew freelance IT

The situation

A management consultancy based in London had been growing steadily for three years. At 18 people, they'd engaged a local freelance IT consultant — a reliable individual who charged £500 a month for ad hoc support: setting up new laptops, fixing the occasional broken VPN, resetting passwords, handling any licensing questions that came up. It worked well at the time.

By the time the firm reached 40 people, the arrangement was visibly straining. The freelancer was a single point of failure: if he was on holiday, on-site with another client, or simply busy, tickets waited. At 40 users, the firm was generating enough IT activity that meaningful resolution delays were affecting chargeable work. A senior consultant spent 90 minutes one afternoon waiting for a password reset. That's not a minor inconvenience at consultant day rates.

The firm also had no visibility. They knew roughly what they paid each month, but had no reporting on ticket volumes, resolution times, or patterns. When the operations director asked whether there were recurring issues that could be fixed at the root rather than patched each time, the honest answer was: nobody knew.

The freelancer was not the problem. He was doing exactly what he'd been engaged to do. The firm had simply grown past what a single person could reliably cover.

What we rolled out

The transition was managed with the freelancer rather than around him — he provided a knowledge handover covering device inventory, existing configurations, any known quirks, and the licensing situation. This took a half-day of his time and meant we weren't starting from zero.

Agent deployment was pushed to all 40 machines using the firm's M365 Intune setup. All devices were enrolled within 48 hours. The AI ran its baseline diagnostics immediately: device health, patch status, MFA compliance, software inventory, and local admin account presence.

The baseline scan surfaced a number of issues that had been quietly accumulating: eleven machines with software more than 30 days behind on patches (including a widely-used browser plugin with a known vulnerability), six users without MFA set up, and four machines with local administrator accounts that no longer corresponded to anyone at the firm. None of these were crises, but they were risks — and they were invisible under the previous arrangement because there was no systematic monitoring.

All eleven patch issues were resolved within 72 hours without user involvement. MFA was enforced across the tenant via Conditional Access policy, applied by an engineer after a 24-hour notice period. The redundant local admin accounts were removed.

What the AI handled

For a consultancy at this size working primarily in M365 with a mix of Windows and Mac machines, the AI typically handles the following without escalation:

• Password resets and account lockouts — the highest-volume support request in almost every organisation. The AI handles these in under two minutes, including identity verification. At 40 users, a consultancy might see 15–20 of these a month. Each one is now resolved without waiting for anyone's availability.
• New starter setup — the firm was onboarding two to three new staff per month during its growth phase. The AI handles licence assignment, device setup verification, MFA enrolment guidance, and a first-day check-in. What previously took the freelancer two to three hours per hire is now largely automated.
• VPN and remote access issues — consultants working from client sites frequently encounter connectivity issues. The AI reads the VPN client state from the endpoint agent, checks for common configuration drift, and resolves the majority without escalation. Resolution time under three minutes versus the "I'll get back to you when I'm free" of the previous arrangement.
• Performance and stability — consultants doing heavy PowerPoint and Excel work on laptops that haven't been properly maintained notice slowdowns. The AI's continuous monitoring means it often identifies and addresses performance issues (deferred updates, background indexing, memory pressure) before the user notices them.
• Printer and peripheral issues on client sites — a specific pain point for consultancies. The AI can walk a user through connecting to an unfamiliar printer on a client's network, troubleshoot USB peripherals, and handle most of the "I'm on-site and something isn't working" scenarios without requiring anyone to be available.

Where humans stepped in

A 40-user consultancy with active growth and client-site working generates a wider range of complexity than a static office environment. UK engineers handled:

• Mac management — around a quarter of the firm's consultants used Macs, and the firm's M365 and Intune configuration had historically been less attentive to Mac compliance. An engineer audited the Mac fleet, enrolled the machines properly in Intune, and configured the same MFA and device compliance policies as the Windows estate. This was a one-off project that took about a week of part-time engineer time.
• Client network access requests — when a consultant needed persistent access to a client's internal systems (rather than just ad hoc VPN), the configuration involved changes to both the firm's Azure AD and negotiation with the client's IT team. The AI flagged this as outside its scope; an engineer managed the liaison.
• A data loss near-miss — one consultant accidentally deleted a SharePoint document library that contained active project work. The AI immediately flagged the deletion and alerted an engineer, who restored from the SharePoint recycle bin within 20 minutes. The consultant didn't lose a single file. Under the old arrangement, they might not have known who to call.
• Offboarding a senior leaver — when a partner left the firm, their account offboarding required care: transferring ownership of documents, ensuring client data wasn't exported, and auditing what had been shared externally. An engineer managed this process, with the AI producing the account activity report that the engineer reviewed before disabling the account.

Outcome

For a consultancy at this size making this transition, we'd expect outcomes broadly as follows:

• Cost comparable to the freelancer at scale — £10/user/month at 40 users is £400/month, versus the £500/month the freelancer charged. At 50 users, the equivalent MSP cost would be £2,000–£3,000/month. The AI-first model scales without the cost jumping proportionally.
• Availability that matches the working pattern — consultants work from client sites, from home, and occasionally at unusual hours before a presentation. The AI is available when they are. The freelancer was not.
• Reporting for the first time — the operations director receives a monthly digest: ticket volumes, resolution times, categories, escalations. Patterns become visible. When three consultants raised VPN issues in the same week, the AI flagged the pattern and an engineer investigated the root cause (a client VPN client update that conflicted with the firm's configuration). Fixed once; wouldn't have been spotted otherwise.
• The freelancer's knowledge preserved — the knowledge handover meant nothing was lost. The transition was professional and the freelancer maintained a good relationship with the firm. They occasionally bring him in for specific physical hardware work that benefits from his local presence.

The outcome that matters most: the firm's senior leadership stopped fielding IT complaints. That's what happens when support scales with the organisation rather than lagging behind it.