The machine raised an alert. Can anyone actually do the job?

Imagine a production line whose vibration monitoring flags a deteriorating bearing. The maintenance team can see the trend. Production knows there is a problem. The proposed intervention is sitting in the maintenance system.

Then the planner opens the job. The replacement bearing is listed in stock, but nobody has checked its condition. The work requires a lifting arrangement that is already booked elsewhere. The technician who normally performs the alignment is on another shift. Production has a short opening tomorrow, followed by a run that cannot easily be interrupted.

A more accurate warning would be useful. It would still leave these constraints to resolve.

For maintenance AI, one practical measure of usefulness is whether the warning becomes work the site can carry out at the right time.

Follow one alert all the way to the intervention

Choose a recurring equipment problem and follow its current path. Who confirms that the signal warrants investigation? Who specifies the work? Who reserves materials, agrees access and releases the equipment?

Record the pauses between those decisions. A job may spend little time being planned and several days waiting for a production window. Another may be scheduled repeatedly because the parts record says available while the physical item cannot be used.

Those are different problems. A model that summarises history could help the first investigation. A stock reconciliation could remove a planning failure. Neither should be sold as a complete maintenance transformation on its own.

Keep the alert and the work order linked. Otherwise, the team cannot easily tell whether the intervention addressed the observed condition or merely happened nearby in time.

Give the planner a usable preparation pack

An assistant can assemble the relevant history: previous symptoms, work performed, replaced components, inspection observations and unresolved recommendations. It should distinguish a technician's confirmed finding from an earlier suspicion copied into several records.

From that material, it can draft a job scope and identify missing prerequisites. The planner should be able to see which statements came from the equipment record and which require confirmation.

For the bearing job, the preparation pack needs an exact equipment identity, the proposed intervention, required materials, estimated access needs and the specialist tasks involved. It must also preserve the site's approved isolation, permit and return-to-service procedures. Generating fluent instructions does not establish that they are safe or applicable.

Do not let a populated field stand in for a completed check. “Part required” and “part reserved and usable” describe different states. So do “technician named” and “qualified person available during the agreed window”.

Plan around what is ready

A scheduling assistant can compare candidate jobs against real constraints. The useful output explains why a job fits a window, what prevents it from fitting and which other work would be displaced.

The responsible people still approve the intervention and the operational trade-offs. Where the condition could affect safety, the site's competent engineering and safety processes determine the response. A scheduling score cannot authorise continued operation.

There is also value in a clear refusal to schedule. “Waiting for the alignment technician” gives someone a problem to solve. A promising date with hidden assumptions gives production a surprise.

Close the job with information the next planner can use

After the intervention, capture what was actually found and done. A bearing replacement alone does not establish the original cause. If the technician suspects misalignment but has not confirmed it, preserve that uncertainty.

An assistant can turn rough notes into a structured draft, then ask the technician to confirm the facts. It should retain the original record and avoid converting missing information into a reassuring conclusion.

Measure the whole path: alerts investigated, jobs made ready, schedules changed for missing prerequisites, equipment released and recurrent symptoms after intervention. Compare like-for-like work and operating conditions before claiming an improvement. Fewer alerts could mean a healthier machine or a broken sensor.

The first useful exercise is small. Bring one recent alert, its work order and the schedule it eventually entered. Put the waiting periods on the same page.

Use a Strategy Session to examine that maintenance workflow, including the information an assistant could prepare and the checks your site must retain.