Building your own CMMS with AI?
What AI solves and what it doesn't when building your own maintenance system, the legal obligations you take on if your build sends data through an AI, what sits behind a system in production, and your alternatives: GMAO CLOUD adapted to you, or your own view on its API.
Updated on 14 min read
- Development
- AI
- Costs
- CMMS
With today’s AI tools that write software, putting together a first version that lists assets and work orders is a matter of days. That has made the question come up in many companies: “with AI we’ll build it ourselves, it’s dead easy”. If we can build it, why buy it?
It’s a legitimate question and deserves an honest answer. What follows is what’s easy, what isn’t, what legal obligations come with putting an AI in the middle, what sits behind a system that has been in production for years, and what alternatives there are when what you need is something specific that doesn’t fit.
What AI solves well
It’s worth acknowledging plainly, because it’s a lot.
The basic screens. Adding, listing, searching and editing equipment, work orders or customers. What used to be weeks of repetitive work is now fairly quick.
The initial structure. Where each piece of data is stored, a login with a username and password, and getting it up and running. A usable first version in little time.
The visual side. Decent screens without needing a designer.
If what you need is an orderly record of equipment and jobs for a small, stable operation, it’s a real option. It shouldn’t be ruled out on principle.
What it doesn’t solve, and where the work is
A CMMS isn’t just a set of screens to store and look up data. The hard part isn’t the part that gets generated quickly.
Really working without coverage. It isn’t enough for the app to open offline. The phone has to store what the technician does, send it when coverage returns, know what to do if someone else changed the same thing in the meantime, and warn if something couldn’t be sent, instead of losing a morning’s work without anyone noticing. And maintenance happens exactly where there’s no coverage.
Preventive generation. It looks like a simple task that repeats every so often, and it isn’t: multiple frequencies on the same asset, holidays and technician availability, the task list resolved in cascade by asset, model, subfamily and family, and meter-based triggers when a reading crosses a threshold. Each of those rules comes from a real case someone discovered the hard way.
Access control. Profiles, document visibility by role, what each customer and each supplier sees, controlled by the system itself and not just by hiding buttons. And traceability: who created, changed or deleted each thing, with recoverable deletion.
Integrations. Connecting the CMMS to your management or invoicing software so it works every day isn’t just linking two programs: it’s knowing what to do when something fails, trying again, checking that both say the same thing, and years of odd cases.
History as evidence. For a record to hold up in a claim or an inspection, nobody can be able to alter it without leaving a trace.
The legal side: your data, your customers and AI
There’s a risk that almost never makes it into the budget of an in-house build, and it’s the most expensive one if it goes wrong: complying with the law.
A maintenance system handles personal data all day long: names, phone numbers and addresses of your customers and their contacts, signatures, photos of the premises, the location and hours of your technicians. In some sectors, also particularly sensitive data, such as that of a healthcare facility or a care home.
When a custom app sends that data to a general-purpose AI service — to summarise a report, reply to a customer or “read” a photo — your company is the controller of that processing in every respect. Among other things, that means:
- Having a legal basis to use that data for that purpose, and having informed the people concerned.
- Having a data processing agreement signed with the AI provider and knowing what it does with what you send.
- Controlling whether the data leaves the European Economic Area, and with what safeguards.
- Sending only what’s strictly necessary, not the customer’s full record because it was easier to code.
That’s what the General Data Protection Regulation (GDPR) requires, and it hasn’t changed because the code was written by an AI.
On top of that comes the EU Artificial Intelligence Act, whose transparency obligations have applied since 2 August 2026. If your app has, for example, an assistant that chats with your customers, they have to know they’re talking to an AI. And the people in your company using these tools need sufficient training to do so.
None of this is solved by the first version. It’s solved by someone who has thought it through before writing the first line. This article isn’t legal advice: if you’re going to process customer data with AI, check with your data protection officer.
A team, not a person
In many companies the in-house build ends up in the hands of the company’s IT person: the one who sorts out email, the printer and the laptops, and who now, with AI, also builds applications. They’re usually very valuable, and that’s exactly why it’s worth not loading them with something they can’t sustain on their own.
Think about what the CMMS is for your company. It isn’t just another internal tool: it’s where your customers’ work orders come in, where your technicians work out in the field, and where the signed reports you then invoice come from. If it stops, operations stop.
Having that depend on a single person is a bet worth not making:
- They fall ill or go on holiday, and nobody else knows why the system has stopped receiving the technicians’ work or how to get it running again.
- They change jobs, and the only explanation of why things were built that way leaves with them.
- They can’t be expected to know everything. Working without coverage, security, speed with years of data, backups that can actually be restored, data protection: these are different specialities, and none of them comes from an assistant that writes code.
It isn’t a question of talent, but of structure. What gives continuity to software your operations depend on is a team: several people who know each part, review each other’s work, cover each other’s absences and have spent years solving the same problems for many customers. That’s what you get with GMAO CLOUD. And your IT person is free for what they know best, which is your company.
What sits behind a system in production
AI writes code. What it doesn’t bring is everything around the code that keeps a system working every morning for years. That, specifically, is what we do and what you pay for when you use GMAO CLOUD.
Maintenance experience, not just software. We’ve been building CMMS for more than ten years: some customers have been with us since 2016, started on a custom build and are still with us on the current version. Those preventive rules, those per-customer and per-supplier permissions, come from watching maintenance companies, industry, healthcare and public administration operate. It’s knowledge of the sector, and it’s what an AI can’t make up for you.
Every change recorded and tested in stages. We keep the full history of the software: we know what changed, when, who did it and why. No change goes straight to customers: it first goes through a test version, automatic checks and a pre-release version. And if something goes wrong, we can go back to the previous version.
Review by the most experienced and cross-working. No change is released without someone else reviewing it, and reviews of the sensitive parts — permissions, customer data, working without coverage — are done by someone experienced in that part. Developers deliberately work on what the others have built, so there’s no part only one person understands. It’s the direct remedy for the most serious risk of an in-house build.
Our own error tracking. We run our own system for detecting errors, on our own servers. When something fails on a screen, in the app or in a nightly process, we see it with all the detail we need, without waiting for anyone to call, and it’s often fixed before the customer notices.
Speed with real data. A first version is fast with a hundred work orders. The question is how it behaves with hundreds of thousands, years of history and dozens of technicians sending their work at once. We continuously review and tune the most used lists and searches, because that problem only shows up once a system has been running for a while.
Servers we look after. Each customer’s data and files are kept separate from everyone else’s. We apply security updates as soon as they come out, not when there’s time, and we regularly review the security of the software and of the third-party pieces it uses.
Backups. A daily backup of each customer’s data, stored away from the server where it runs, which retries by itself if it fails and alerts the team if it doesn’t succeed. Files — photos, reports, certificates — are copied, encrypted, to a second data centre in another region. It isn’t something you set up once: it’s monitored.
Continuous improvement. Every issue we fix leaves a written lesson and, where it applies, an automatic check that stops it coming back. The product improves every week with what customers ask for and discover, and that improvement reaches everyone.
The cost nobody calculates
All of the above is what an in-house build has to take on by itself, and it’s what usually gets left out of the budget.
Maintaining the system itself. Security updates, third-party pieces going out of date, Android and iPhone changes that force you to rework the app. It’s permanent work, not a project.
Support. When something breaks on a Friday afternoon and the person who wrote it is on holiday or no longer works here.
Depending on one person. We covered it above, and generating it with AI doesn’t fix it: it makes it worse, because often not even the person who generated it can explain every decision.
The servers. Where the system runs, backups, error monitoring, security updates and keeping it available at all times. And someone who notices when any of that stops working.
Compliance. Reviewing what data goes out, to whom and under what contract, every time someone adds a new feature.
What you don’t know you need. The features that aren’t on your first list because you haven’t yet tripped over the problem that made them necessary.
The two alternatives that do work
What usually pushes a company to build its own is a specific need that doesn’t fit: a workflow, a screen, a requirement from one of its customers. That has two ways out that don’t mean taking on everything above.
GMAO CLOUD, adapted to the way you work
The simplest: ask us. GMAO CLOUD isn’t a closed product. What it doesn’t cover we study with you and, when it’s useful, build into the product, so the improvement is maintained by our team, not your company. That’s how we work with customers who have been with us for years and keep asking for changes. How we adapt GMAO CLOUD to each company.
Your own view, with GMAO CLOUD behind it
If you already work with GMAO CLOUD and want an app for your technicians with your own flow, a portal for your customers or a dashboard your way, you can also build it yourself — with AI or without — connecting it to GMAO CLOUD through its API, the entry point that lets other programs read and save data in the system (how GMAO CLOUD connects). You provide the view; we’re behind it.
What you build is what sets you apart: your screens, your way of working, your way of presenting information.
What GMAO CLOUD keeps solving is what’s expensive to do well:
- Preventive generation, which keeps running even if your app doesn’t know about it.
- Each user’s permissions and scope, controlled by the system itself: your app benefits from that protection without having to program it again.
- The record of every creation, change and deletion, wherever it comes from.
- Backups, error tracking, updates and security.
And if your app fails, your operations don’t stop: the GMAO CLOUD management panel and apps keep working on the same data, and your technicians can carry on with them while you fix it. You can also mix on purpose: your office on your own panel, your technicians on our app and your customers on their portal.
How API access works. It goes by plan, with a usage volume suited to each one, and for high-volume connections we size it with you. That way you know from the start what your app can do and what it costs, and we can make sure the system responds just as well for everyone.
Two recommendations if you go this way:
- Always save data through the API. It’s what checks what comes in, applies permissions and records every change. Putting data in directly underneath the system skips all of that, and that’s what turns a reliable history into one nobody trusts.
- If your app uses AI, decide first what data you pass to it. Whatever leaves GMAO CLOUD for an AI service of your choice is processing you’re responsible for, with everything explained above. If what you need is AI over your own maintenance, look first at what the product already includes.
Tell us what you want to build and we’ll help you set it up properly from the start.
And about AI inside the product
It’s worth separating two things that get mixed up: using AI to build the system, and having AI inside the system.
In GMAO CLOUD, AI does bounded, verifiable things: an assistant that answers about your assets, your plans and your history — in the technician app too; document reading, so a supplier’s PDF report becomes data without retyping it; help writing work orders; and translation of your own catalogues.
It proposes, it doesn’t execute: it doesn’t close work orders or assign jobs. Each feature has its own switch, with most of them off by default, and the documents it consults and the conversations are stored with the rest of your data. It’s covered in detail in AI in GMAO CLOUD.
An assistant over an empty history knows nothing. AI doesn’t replace the record; it relies on it.
When building it yourself does make sense
When your process is genuinely unusual and no product on the market covers it, because that particularity is your business.
When the system is a piece of your product, not an internal tool.
When you have a stable in-house development team, able to maintain it for years and to answer for it to a customer or an inspector.
Outside those cases, what usually happens is that the first version works, gets used for a few months and is abandoned when the first real demand shows up: a technician with no coverage, an inspection that asks for the file, a customer who wants their portal, or a question about where their data went.
How to decide
Make the list of what you really need — not what would be nice — and mark which part is basic screens and which part is the hard kind. If most of it is the latter, the first version won’t save you the work: it’ll postpone it.
Then ask yourself these questions about the in-house system:
- Who finds out if something breaks at three in the morning?
- Where is yesterday’s backup, and who has ever restored it?
- Who reviews a change before it reaches production?
- Who maintains it if the person who built it leaves?
- Which services does your customers’ data go to, and under what contract?
If there isn’t a named answer for each one, work out the cost over three years, not three months, with all of that included. And compare it with the two alternatives: asking us to adapt GMAO CLOUD to the way you work, or building just your view and leaving the rest to us.
If you want to test your specific case, you can write to us or request a demo to see which part of your list is already solved.