How should an MSP or an internal IT team put AI into service delivery?
Start where the work is repetitive and the mistakes are cheap to undo: tier-1 triage, ticket summaries, documentation, runbook drafts. Keep AI away from anything that changes a production system until there is a named approver, a log, and a rollback. That order has held up across a combined twenty years of working alongside MSPs, and it is the order most vendors will try to talk you out of.
- Measure before touching anythingthree weeksNinety days of tickets, categorized and timed. The three or four categories that eat the week become the plan.
- Triage and documentation firstSummaries, categorization, a suggested next step, a draft knowledge-base article per resolved ticket. Every error is caught in ten seconds and undone.
- Governance before productionA named approver, a log, and a rollback on anything that changes a client system. The never-touch list, signed.
- Technicians, on their own ticketsTwo working sessions per tier on the live queue, not sample data.
- Copilot when the tenant is readyData-exposure review first. Then licensing, department by department, with a usage policy and a measurement plan.
- Package itMSPsOnce it works on your own desk, it is a service you sell under your name.
IT & MSP Operations
Why MSPs ask about this service area first
Your clients are asking what you are doing about AI, and your competitors’ websites already have an answer, whether or not anything is behind it. One industry survey put MSP adoption of AI at 2 percent and then 20 percent within a year, and most of that jump was providers adding the word to a service tier rather than changing how a ticket moves (Lumenova).
The gains on offer are real. Across deployments NetSuite surveyed, technician productivity rose 15 to 25 percent and ticket resolution times fell 40 to 70 percent (NetSuite). Those numbers come from taking triage and documentation off people’s plates. They do not come from taking people off the payroll, and a plan that assumes otherwise fails in the first quarter, when senior review becomes the bottleneck.
The wrong first move is the common one: buy the tool with the AI badge from the vendor you already pay, turn it on for everyone, and measure nothing. Six months later the queue looks the same, three technicians have quietly stopped using it, and the client who asked what you are doing about AI has the same question.
What we actually do
Measure before touching anything. Three weeks. We pull ninety days of tickets from your PSA, categorize them, and time them. Every service desk has three or four categories that eat the week, and they are rarely the ones the team names when asked. That list, with minutes per ticket attached, is the plan.
Triage and documentation first. Ticket summaries, categorization, a suggested next step, and a draft knowledge-base article generated from each resolved ticket. Every one of these is a mistake a technician can catch in ten seconds and undo. That property matters more for a first deployment than the size of the gain.
Governance before production. Nothing AI produces changes a client system until a named person approves it, the action is logged, and there is a rollback. We write the list of things AI never touches and get it signed before anything ships: credentials, firewall rules, backup jobs, anything that runs as an administrator. Your clients get a one-page usage and data-exposure policy they can read without a lawyer.
Technicians, on their own tickets. Two working sessions per tier, using the queue as it exists that week. Sample data teaches nothing. A technician who watches AI write a bad summary of a ticket they closed yesterday learns exactly where to trust it and where not to.
Copilot when the tenant is ready. Microsoft 365 Copilot is usually the first thing a client asks an MSP for by name, and usually not the first thing to deploy. Detail below.
Then, for MSPs, package it. Once it works on your own desk, it is a service you can sell. That is its own offer, also below.
Two offers inside this service area
Microsoft 365 Copilot rollout and governance
Copilot surfaces everything a user can already reach, and in most tenants that is far more than anyone intended. A SharePoint site shared with Everyone in 2019 becomes a Copilot answer in 2026. The rollout that works starts with a data-exposure review, not a licensing order.
What we do: license planning, a permissions and oversharing review across SharePoint and OneDrive, a department-by-department rollout with a usage policy per department, and a measurement plan so the renewal conversation has numbers in it.
An AI practice for MSPs
Your clients will buy AI help from someone. We build the practice so that someone is you: an assessment template you can run in a week, governance packages by client size, tiered runbooks, technician enablement, and a pricing model. You own all of it and sell it under your name. We stay behind you on the engagements that need more depth than the template covers.
What you get
- Operations assessment ranked by hours: which ticket categories eat the week, and what a technician actually does in each
- Tooling selection against the PSA and RMM you already run, with a written reason for every choice
- Tier-1 triage and summarization wired into the queue, with a named approver on every action that touches a client system
- Tiered runbooks: what AI drafts, what a technician verifies, what nobody automates
- Technician enablement, two working sessions per tier, on your tickets rather than sample data
- A usage and data-exposure policy your clients can read without a lawyer
- Baselines before and after: first-response time, time to resolution, tickets per technician per day
- For MSPs, a client-facing service design and packaging for an AI-enabled tier
Questions we get
- Do you resell PSA, RMM, or AI tools?
- We hold partner relationships with some vendors and can resell where that is the cheaper route for you. Where a recommendation carries a commercial interest for us, we tell you inside the recommendation. If you would rather we did not resell on your engagement, say so at scoping and we will quote it that way.
- Will this let us run the desk with fewer technicians?
- The published numbers are real: technician productivity up 15 to 25 percent and resolution times down 40 to 70 percent in the deployments NetSuite surveyed. The gain comes from removing triage and documentation load. In practice the constraint then moves to senior review capacity, so plan for that rather than for headcount.
- We are an internal IT team, not an MSP. Does this apply?
- Yes, same sequence. You skip the packaging step, and the client-facing policy becomes an internal one. The oversharing review matters more for you, because nobody outside will ever audit it.
- Does our clients' data go to a model vendor?
- It depends on the tool and the tier. Microsoft 365 Copilot stays inside the tenant boundary under Microsoft's enterprise terms. AI features inside PSA and RMM products vary, and we read the terms before you turn one on. Where data cannot leave at all, we run open-weight models on your own hardware.
- Can we just start with Copilot?
- You can. Most should not. Copilot surfaces whatever a user can already reach, and most tenants have years of accidental oversharing. Do the data-exposure review first; if it comes back clean, Copilot can be the first deployment.
How it starts
Three-week assessment, then implementation sprints of four to eight weeks. You keep everything produced, whether or not there is a next part.