Automating maintenance: where to start

Automating maintenance isn’t about the system working on its own. It’s about removing the tasks that today eat up someone’s time without requiring judgment: creating the order that’s due, notifying the right person, deducting what’s been used, and reminding people what’s about to expire.

What’s in what can be automated is the catalog. This one is about order, which matters as much as the list.

The order that works

1. Generate preventive work. Always first, because it frees up the most time and causes the least pushback: nobody misses creating orders by hand.

2. Warn about what’s expiring. Second by effort-to-risk-avoided ratio. Loading the dates of certificates, contracts, and insurance policies is half an afternoon.

3. Deduct material when closing the order. This requires the team to already be comfortable with the app, so it shouldn’t be moved up.

4. Notify status changes. And here, start with few.

5. Assign by rules. Last, because it needs history to know which rules make sense.

Reversing this order is the most common mistake: starting with automatic assignment — which is what sounds most advanced — and configuring it blind.

1. Preventive maintenance, where the bulk of the value is

The checklist gets attached to the asset, its model, or an entire family, the frequency gets defined, and preventive maintenance generates work orders on its own.

Two checks it runs beforehand that set it apart from a calendar reminder: whether the day is a holiday and whether the technician is available.

And a second trigger worth turning on when wear depends on usage: the asset carries a counter — hours, cycles, kilometers — with a limit and a warning percentage, and when a reading that exceeds the threshold gets logged, the order generates automatically.

2. Expiry dates

It’s the automation with the best ratio between how much it costs to set up and the problem it avoids.

Documentation can carry an expiry date, and there’s a daily check of what’s about to expire that triggers the notification. Same for staff personal protective equipment.

Finding out late that a certificate expired is the kind of problem money doesn’t fix.

3. Material

Consumption stops being a follow-up administrative task and becomes a consequence of closing the order: the technician consumes the item from the order itself in the app, and that gesture lowers stock and raises the cost of the intervention.

On top of that, minimum stock per item in warehouses and items warns before you run out of what a job needs.

4. Notifications, in moderation

Statuses on an incident or an order can carry their associated email, so a status change notifies only the person concerned. The technician gets a push and email notification for whatever gets assigned to them.

And what matters just as much: you can mark which statuses shouldn’t notify. It’s much easier to add a notification than to remove one once people have learned to ignore the emails.

Start with two or three statuses that the recipient actually cares about.

5. Rule-based assignment

It exists, and it’s worth being clear about what it actually is: deterministic rules you define, not a model that guesses.

That means three concrete things: you configure them and can turn them off, every decision leaves a record with its reason, and they’re reversible. That combination — configurable, traceable, and reversible — is what makes it acceptable to automate a step someone is accountable for.

It’s left for last because the rules that make sense come from history, not from an initial guess.

And AI, which is a different thing

It’s worth not mixing it with the above, because presenting them together just to sound more modern would do everyone a disservice.

GMAO Cloud’s artificial intelligence features don’t make decisions: they read, respond, and suggest, and what gets applied is applied by a person. They do concrete things — an assistant over your assets and history, document reading, help with order text, translation of your catalogs — and each one has its own switch, with most off by default. Detailed in AI in GMAO Cloud.

Rule-based assignment is a configurable automation. It’s not the same thing, and telling them apart matters when you have to explain why the system did something.

The automation that removes the most administrative work

Two aren’t on the list above because they don’t affect the field team, but they free up plenty of office hours.

Letting work come in on its own. An email inbox that the system empties and converts into incidents, with a different inbox per type if needed. It removes the step where someone reads an email and types it somewhere else.

Letting data flow out to the ERP on its own. Closed orders with their hours and material are exactly what the ERP needs to invoice. In integrations you’ll find the scope of each connector, which isn’t the same across all of them: worth checking before assuming.

Neither one requires changing how the field team works, and both are noticeable from the first week.

What shouldn’t be automated

Priority. What gets attended to first is decided by someone with context in front of them.

Frequency. It’s set by technical judgment, the regulation, or the manufacturer, and corrected with history.

Closing an order. If it can be closed without anyone confirming it was actually done, the history stops working as evidence.

An asset’s criticality. It’s a business decision, not a calculation.

How to know if it’s working

With the reports: executed annual plan, preventive vs. corrective hours, detected anomalies, and cost per equipment.

If executed plan goes up and corrective work goes down, it’s working. If both go up, work is probably being generated that nobody’s actually doing — and that’s useful information too, because it means the plan is bigger than what the team can execute.

If you want to see what would get automated in your case, you can request a demo.