Everything I wish someone had told me about running an MSP.

The three stages

Most MSP owners I know are technical founders who learned business the hard way, me included, and you built something. But a lot of us call ourselves advisors when the advisor meeting is mostly a review of what we already sell. This is the order I’d fix things in, and each stage is a perfectly good place to stop.

1. Stage 1

   Be a good MSP first

   Before you pick an AI role, take a hard look at the business you already run. If it’s a shitty process, AI just makes a faster shitty process, so get your basics solid first.

   • Work doesn’t depend on one hero. Roles and decisions are clear enough that your best tech can take a real vacation.
   • Prove your services work, not just that the licenses are assigned. Patching, backups, EDR and monitoring have evidence and an owner for the exceptions.
   • Projects get scoped, managed for change and closed out, so they don’t vanish into the help desk or turn into an endless favor.
   • Know your retention, margin and capacity, and what is and isn’t included in what you sell. A business that keeps replacing unhappy clients isn’t a healthy one.
   • Use AI inside your own shop: tickets, documentation, quoting, your own inbox. Your margins need it, and you can’t advise on something you don’t use.
2. Stage 2

   Be the advisor

   A real advisor starts from what the business is trying to do, not from the stack you resell. You don’t have to deliver everything, but every part of the client’s technology should at least be in the conversation.

   • Hold a regular planning conversation with a named decision-maker about their business, not only tickets and renewals.
   • For every need that comes up, decide on purpose whether you’ll own it, broker it or partner for it.
   • When you bring in a specialist, huddle privately first, agree on roles, then meet the client together, and you keep the relationship.
   • Brokering isn’t a consolation prize, and for a lot of MSPs it’s exactly the right business model.
3. Stage 3

   Sell AI only where it recurs and fits your business

   Selling AI to clients is optional. If you do, start with the work that looks like what you already do every month. Business process work is consulting work, with different people, a different sales motion and a different support model, so decide that up front, don’t just slide into it because a demo looked cool.

   • Start with AI work that recurs: readiness and access reviews, keeping the approved-tool list current, adoption of what clients already own. Broker or partner on the one-off builds.
   • Put the basic guardrails in first: a client who decides, a short approved-use list, sensible data access and a regular review.
   • Know what it really costs, including the discovery nobody pays for, training, tools and escalation.
   • Only call it a managed service if there’s real recurring work behind it. Don’t sell “AI governance” because the market likes the phrase.
   • Decide what’s repeatable enough to package and what stays custom consulting, and how you’ll measure outcomes without promising AI is right every time.

AI: what you have to do, and what you don’t

Every vendor deck says MSPs have to “do AI.” That’s only half right, and the other half is how good MSPs end up as project shops by accident.

If you do sell it, sell what recurs

The four questions before you promise a client anything

“We don’t do that” isn’t a complete answer for a client, and neither is pretending you do. Ask these in order and the right answer usually shows up on its own.

1. 01Diagnose

   What business problem or friction is actually here?

   Name the problem before anyone names a tool.

2. 02Own

   What can we do credibly with the people and operating model we have today?

   Own the work that fits your practice, and support it for the long haul.

3. 03Broker

   What can we guide and coordinate without delivering it ourselves?

   Advise, coordinate and keep the client relationship.

4. 04Partner

   Where would we create risk by forcing it?

   Bring in a specialist, with clear boundaries on who does what.

“Own” is the part people skip. It means validation, support, exceptions and adjusting the work as the client and their systems change, long after the project party is over.

Know where you are

These aren’t benchmarks, they’re patterns I’ve seen over and over, in my own companies and in the MSPs I talk to, and your numbers will vary, but the walls tend to show up anyway.

1. Around $3MOften 20 to 25 people

   The operational-maturity wall

   Up to here, the founder’s energy carries the business. Getting past it takes real processes, some leadership and a way of running that doesn’t need you in every ticket. Plenty of good MSPs park right here for years.

2. Around $7MOften 50 to 60 people

   The leadership wall

   You run more like a real business by now, but the owner usually still has a hand on most of it. Getting past this one means building an actual leadership team and letting go of the reins, which is harder than any technical problem you’ve ever solved.

3. Past $7MOften investor-backed

   The integration years

   Bigger shops tend to grow by acquisition, and a lot of the work becomes stitching companies together. That’s its own set of problems and a story for another day.

The 10-minute self-check

Tick what’s true today, not what will be true after the next QBR. Most of us grade ourselves a lot nicer than we deserve, me included, so answer as if your service manager is reading over your shoulder. Your answers stay in this browser and nobody else sees them.

Go deeper

• The MSP Playbook seriesEach part takes one idea from this page further, with the mistakes I made along the way. New parts publish every other week.
• AI without the hypeFor the client who asks whether they’re behind on AI, and the practical answer to give them.
• SpeakingTalks and roundtables for MSP owners, if you’d like this conversation in front of your peers.