Every IoT predictive maintenance platform demos well. Clean dashboards, live sensor feeds, an impressive-looking alert. But a demo runs on a curated machine in ideal conditions, and your plant is neither. In Indonesia, where around 80 percent of manufacturers are now putting real budget into smart manufacturing, the risk is not failing to invest. It is investing in a platform that stalls in a pilot and never scales.
Here are eight questions that separate a predictive maintenance platform that reasons from one that only collects. Ask them before you sign.
1. Does it baseline each machine, or apply one threshold?
Two identical machines do not share a signature once foundation, alignment, and load diverge. A platform that sets one generic alarm across assets will nuisance-trip on one and stay silent on another. Ask whether it builds a per-asset baseline, because that single choice decides whether your alerts mean anything.
2. Does it diagnose the failure, or just detect a change?
An alert that says vibration is high is not an answer. The platform worth buying isolates the failure mode and likely root cause, so your team knows what they are walking up to and what the fix will take. Detection is cheap. Diagnosis is the value.
3. What sits behind the sensor?
This is the question most buyers forget. A sensor gathers a signal, but meaning comes from what interprets it. Ask whether the platform draws on a wider library of machine knowledge, or only your own limited history. For example, the Groundup.ai Asset Library, has accumulated health signatures and failure modes of machinery seen across fleets and sectors, which is what lets the system recognise a fault rather than merely flag a number.
4. Can it estimate remaining useful life?
Knowing something will fail is useful. Knowing how long you have is what protects the schedule. Ask whether the platform estimates remaining useful life, so a repair becomes a planned job in a maintenance window instead of an emergency mid-shift.
5. How fast, and how invasive, is deployment?
You cannot take critical assets offline for weeks to instrument them. Ask whether sensors mount non-invasively and whether coverage starts in weeks rather than at the next shutdown. A platform that is painful to deploy is a platform that dies in a pilot.
6. Does it fit the systems you already run?
You have invested in SCADA, historians, and a maintenance system. Ask whether the platform layers onto that stack and pushes recommendations into the work order your team already uses, or demands a rip-and-replace and its own separate dashboard nobody opens.
7. Does it end in an action a person can trust?
The best model is worthless if the operator ignores it. Ask what actually lands in the technician’s hands: a raw alert, or a specific recommended action with a diagnosis attached. Trust is earned the first time the system catches something the spreadsheet missed, and lost the first time it cries wolf.
8. Will it scale past the pilot?
Most IoT programs succeed on a few curated assets and then stall, defeated by legacy systems and messy real-world data. Ask the vendor how their platform handles that mess, and ask to start small on one line where a win is visible. As one Indonesian plant leader put it, you start small and build trust among your coworkers until the solution becomes a standard that helps. Scale should follow proof, not precede it.
The thread through all eight
Notice that most of these questions are really one: does the platform stop at the alert, or carry the reading all the way to a trustworthy action? That is the line between IoT that collects and Cognitive Maintenance that reasons, diagnoses, and guides. Buy the second one.
P.S. Print these eight and take them into your next vendor demo. The questions a platform dodges tell you more than the ones it answers.
