AI inzetten MSP
How do you put AI to meaningful use as an MSP?
AI only produces a return for an MSP once its own data is in order and the process is described in advance. In practice MSPs achieve 20 to 40 per cent time saved on ticket triage, but 90 per cent of AI projects in business fail according to recent MIT research. The difference sits in setting the frame, not in the tool.
MSP Late Night · 6 August 2026
Build your own, buy tools or wait for your vendor?
Three routes come back at MSPs working with AI. Building your own on your own data lake gives a lead in client reporting, but asks for capacity most 40 to 60 person MSPs do not have. Buying separate tools on the Azure OpenAI stack produces quick results on triage and categorisation. Waiting for the PSA vendor prevents wasted development time, because triage functionality arrives in the large platforms themselves within the year.
Where does AI already produce gains?
Ticket triage, prioritisation and filtering noise out of alerts. MSPs that have set this up save 20 to 40 per cent of the time per ticket and categorise around 80 per cent correctly.
Why do so many AI projects fail?
Companies buy a tool, turn it loose on a polluted data lake and expect results. Without clean sources, clear role instructions and a recorded process, AI mainly increases the visibility of the mess.
Does AI replace the service desk?
No. Simple and repetitive tickets move to automation, but contextual escalations and client relationships stay human work. LLMs are not 100 per cent consistent, so validation by an engineer stays necessary.
If your own data lake is not clean, AI mainly brings your mess to light.
Frequently asked questions
What is a data lake and why is it relevant for an MSP?
A data lake is a central store where unstructured and structured data from all sources come together: PSA, RMM, Microsoft 365 scans, vulnerability scans, invoices and contracts. For an MSP this is relevant because AI models only give usable answers when they can talk to one consistent source. Separate silos per tool make role-based reporting to CFOs, IT managers or end users nearly impossible.
How much time saved is realistic with AI triage of tickets?
MSPs using AI to categorise, prioritise and pre-process tickets report on average 20 to 40 per cent less time spent per ticket. Correct categorisation sits around 80 per cent, comparable to what a human service desk employee achieves. The gain sits mainly in filtering out noise and automatically routing sales questions to the right department.
Is it better as an MSP to build AI yourself or wait for the vendor?
For MSPs under around 60 staff, building your own is rarely sustainable: the lead time of your own development is often overtaken by functionality the PSA or RMM vendor rolls out itself. The drawback of waiting is that you build no lead in client reporting and advice. A middle way is deploying separate tools for triage now, and reserving your own development for client-facing insights vendors do not provide.
What role do engineers and service desk staff keep as AI takes over more?
Repetitive work such as password resets, conditional access changes for holidays and standard onboarding flows shifts to automation agents. Engineers therefore get more time for project work, complex incidents and client advice. Validating AI output stays human work, and maintaining relationships and handling contextual escalations weigh more heavily. The role shifts from executing to guiding and judging.