Prescriptive maintenance: turning a fault recommendation into a work order that gets done

IBM defines prescriptive maintenance as a strategy that uses AI, IoT and machine learning "to analyze equipment conditions and recommend actions". It goes "one step further" than predictive maintenance "by recommending specific maintenance tasks".

In common maintenance practice, strategies are often described as a maturity ladder. The five steps below are our own framing of that common practice:

1. Reactive: fix it when it breaks.
2. Preventive: service it on a calendar or meter schedule.
3. Condition-based: act when a measured condition crosses a threshold.
4. Predictive: forecast when the asset is likely to fail.
5. Prescriptive: recommend the specific action and when to take it.

In practice, each step relies on the foundations of the one before. IBM states that "prescriptive maintenance depends on predictive maintenance for its core functionality". For the middle of the ladder, see our guides on condition-based maintenance thresholds that raise work orders and predictive maintenance readiness.

IBM draws the line at the recommendation. Predictive maintenance uses operational data and real-time condition monitoring to predict when assets are likely to fail; that is where traditional predictive systems end, and technicians act on the information. Prescriptive maintenance adds a layer that evaluates potential responses and recommends actions.

Summary (our framing, drawing on IBM):

IBM gives a bearing example. If a system finds that a bearing is likely to fail soon, a prescriptive recommendation might be to:

• Replace the bearing within a specific time frame.
• Adjust operational parameters to accommodate the faulty bearing.
• Reduce operating speed to accommodate maintenance tasks.
• Schedule the maintenance during an upcoming shutdown to avoid unplanned downtime.
• Increase lubrication.

Each option needs a decision, a person, a date and a record.

A correct recommendation can still go undone, for ordinary reasons:

• No owner. The alert reaches a dashboard or an inbox, and nobody is named to act on it.
• No due date. The "by when" in the recommendation never becomes a date in the plan.
• Parts not reserved. The bearing is not in stock, or it is in stock but promised to another job.
• Not in the week's schedule. The job competes with breakdowns and preventive work and loses.
• No procedure. The technician has the task but not the steps, the torque values or the checklist.
• Closure not checked. The work order is closed, but nobody confirms that the asset's condition returned to normal.

In our view, without a closed loop from recommendation to work order to verified repair, prescriptive maintenance adds little to what predictive maintenance already told you.

IBM describes a CMMS that automates tasks "such as generating work orders, ordering replacement parts and keeping maintenance schedules up to date", and feeds the outcome of completed work back into continual learning. In practice, the loop runs in eight steps:

1. Receive the recommendation against the right asset in the asset register.
2. Triage it: accept, change or reject, and record the reason. This is a human decision.
3. Create the work order with the task, priority and a due date tied to the recommendation's "by when".
4. Attach the procedure or checklist and reserve the parts.
5. Schedule the job into the maintenance plan or the next shutdown window.
6. Execute the work and record findings, measurements, photos and parts used.
7. Verify the result: check that the condition data returns to normal.
8. Feed the outcome back: record whether the recommendation was accepted or rejected and what the work found, so the next recommendation is better.

Steps 2 and 7 matter most. A recorded rejection shows where the model or threshold is wrong; verification shows whether the repair fixed the cause. For step 4, our guide to spare parts and MRO stock covers reservations and criticality. For step 7, maintenance KPIs show whether closed work is holding.

Prescriptive maintenance sits on top of maintenance basics. Check that you have:

• An asset register and equipment hierarchy you trust, so a recommendation lands on the right asset.
• Preventive maintenance that is actually completed, not just scheduled.
• Work-order history with findings and failure details, not only open and close dates.
• Parts data that shows what is in stock and what is reserved.
• A named maintenance planner who owns the queue of incoming recommendations.

Our guide to predictive maintenance readiness covers the sensors, data and preventive compliance that come first, and our guide to maintenance data teams can use covers keeping the history clean.

Analytics platforms can generate recommendations. The maintenance team still decides what work is done, when, and with what risk.

That is also how Upvantix approaches AI in maintenance. AI assistance in Upvantix CMMS (see Operational AI) is designed to bring asset information, work history, documents and technician knowledge into a reviewable maintenance-assistant experience. As our CMMS page puts it: "AI finds and prepares the context. Your team stays responsible for the decision." Human review is required before operational action.

On 7 October 2026, Tech Mahindra and Infinite Uptime announced a partnership to bring "prescriptive intelligence" to factory operations, as reported by ETEnterpriseAI. As reported, Infinite Uptime's PlantOS analyses equipment, process and maintenance data to identify the faulty component, recommend what needs to be fixed and specify when the repair should be completed. The partners said that identifying problems alone does not improve performance: manufacturers must also "determine the appropriate response and ensure effective execution".

Upvantix CMMS is designed to organise the asset register, schedule preventive maintenance and complete work orders with a connected history for every piece of equipment. It is designed for deployment in your AWS environment.

These capabilities support the work-order steps of the loop above (creating, planning, executing, recording and learning from the work):

• Assets and equipment: asset register, equipment hierarchy, locations, documents and history.
• Work management: work requests, work orders, priorities and status, and completion records.
• Preventive maintenance: PM schedules, maintenance planning, checklists and inspections.
• Parts and knowledge: parts and inventory, procedures, technical documents and maintenance history.
• At the point of work: assigned work, asset lookup and QR access, maintenance history, procedures and checklists, photos, notes and measurements, and parts and time capture.

The connected workflow ends with two stages that matter most here: record, to capture findings, measurements, parts and completion; and improve, to use the asset history to inform future work.

Talk to us about turning maintenance recommendations into owned, scheduled work orders against the asset record.