A Practical Guide to AI for MSPs (No Hype)

Every vendor in the MSP space slapped "AI-powered" on their product this year. Half of them just added a ChatGPT wrapper to their existing tool and called it innovation.

So let me cut through it. I build AI systems for a living. Here's what's real and what's marketing.

What AI is genuinely good at for MSPs

Three things:

• Reading and understanding text. Tickets, emails, chat messages. AI can extract what the person actually needs, categorize it, and route it.
• Finding information across scattered sources. Your docs are in 5 different systems. AI can search all of them at once and synthesize an answer.
• Generating drafts. Responses, summaries, reports, communications. Not final versions — drafts that a human reviews and sends.

That's it. Those three capabilities applied to your specific workflows. Everything else is either a stretch or an outright lie.

What AI is NOT good at

• Replacing your techs. It augments them. A tech with AI is faster. AI without a tech is useless.
• Working "out of the box." Every useful AI deployment requires setup, training on your data, and iteration. Anyone who says otherwise is lying.
• Being 100% accurate. AI operates in the 85-95% range. For many tasks, that's still enormously valuable. For tasks requiring perfection, you need a human in the loop.

How to evaluate any AI project

Three questions:

• Is the task repetitive? AI shines on high-volume, similar-each-time work. One-off creative problems? Humans.
• Do you have data to train it? No ticket history, no documentation, no examples = no AI value. Full stop.
• Is "pretty good" useful? If an 85% accurate draft saves your tech 10 minutes per ticket, that's a win. If you need 100% accuracy (billing, compliance), add human review.

If you can't say yes to all three, the project isn't ready. Walk away.

Red flags from vendors

• "No setup required" — There's always setup.
• "Replaces your L1 team" — It doesn't. It won't.
• "Our proprietary AI" — They're reselling the same models everyone else uses.
• "ROI in the first week" — Realistic timeline is 30-90 days.

The right way to start

Pick one workflow. The most repetitive, highest-volume thing your team does. Test whether AI can meaningfully improve it. Measure the result. Expand only if the numbers work.

Don't sign annual contracts. Don't buy a "platform." Start small, prove value, scale on evidence.