AI voor MSP
Is your MSP really making money with AI?
Most MSPs experiment with AI without it producing margin structurally. Research from MIT shows that 90% of AI projects in business fail, often because the underlying data and processes are not in order before the tool is let loose on them.
MSP Late Night · 31 August 2026
Why do so many AI projects at MSPs fail?
AI only works once the data lake is clean and the underlying process is defined. Without a view of what you want to solve, AI mainly amplifies the mess in your existing systems. Developing it yourself costs capacity an MSP of 60 staff cannot muster against a vendor pouring millions into R&D. Riding along with the platform integrations of your PSA or RMM often produces more than building your own.
Build your own or follow the vendor?
What you build yourself today is often standard in your PSA platform six months later. Triage, ticket classification and prioritisation are already being rolled out by vendors themselves.
Why is data lake hygiene decisive?
Letting AI loose on a polluted data lake mainly brings the mess to the surface. Historical data from tools of ten years ago is rarely usable for current questions.
Where is the human brake?
For validation and sign-off on processes, a person between the AI and the client remains necessary. Full automation of client contact is not yet a realistic scenario, legally or operationally.
If your own process and data lake are not in order, AI mainly brings your mess to light.
Frequently asked questions
What is a data lake and why is it relevant for MSPs using AI?
A data lake is a central store for raw data from different sources, such as PSA, RMM, tickets and client communication. For MSPs it is relevant because AI models only give usable output when the underlying data is structured, current and clean. Polluted or outdated data leads to wrong classifications, wrong priorities and reports clients can do nothing with.
Where is the first concrete gain with AI for an MSP?
Concrete gains usually sit in ticket triage, prioritisation and generating first draft replies based on ticket content. Code assistance for engineers and transcribing meetings or client conversations also save time. These applications work because the input is bounded and the output is validated by a person before it reaches the client.
Is it wise for an MSP to develop AI solutions itself?
For most MSPs, building your own is not sustainable in the long run. An MSP with 60 staff cannot compete on development capacity and maintenance pace with the R&D of a PSA or RMM vendor. Functionality you build yourself today often appears natively in the platform within months. Riding along with vendor roadmaps and integrating through workflow tools is usually more efficient.
Why does a human in the loop remain necessary in AI processes?
A person remains necessary for validation on processes with an impact on clients, contracts or security. AI models still produce errors and hallucinations, and full automation brings liability risks. It is plausible that regulation will in time require human involvement in certain decisions, because societal processes cannot be hung on one model without oversight.