Every manufacturing conference I’ve attended in the past two years has had at least one session on predictive maintenance. The pitch is always the same: sensors detect problems before they happen, you fix things during planned downtime instead of emergency shutdowns, and you save a fortune.

It’s a compelling story. But it’s usually told by vendors selling $200,000 enterprise platforms to BHP and Rio Tinto. The question nobody seems to want to answer honestly is this: does predictive maintenance make financial sense for a small manufacturer with 20-50 employees?

I’ve spent the last few months talking to small manufacturers across Victoria and New South Wales who’ve actually tried it. The answer, as usual, is “it depends.” But the specifics of when it works and when it doesn’t are worth understanding.

What predictive maintenance actually costs

Let’s start with real numbers, because the marketing materials are deliberately vague.

For a small manufacturing operation — say five CNC machines, a couple of compressors, and some conveyor systems — a basic predictive maintenance setup will run you somewhere between $25,000 and $80,000 AUD for the initial deployment. That covers vibration sensors, temperature monitors, the IoT gateway hardware, and the software platform subscription.

The ongoing costs matter too. Most platforms charge $500-$1,500 AUD per month for the software. You’ll need someone trained to interpret the data — either upskilling an existing maintenance person or bringing in a specialist periodically. Training runs about $3,000-$5,000 AUD for a decent course.

So you’re looking at roughly $30,000-$90,000 in year one, then $8,000-$25,000 annually after that.

Is that worth it? That depends entirely on what your downtime actually costs.

When it absolutely makes sense

I spoke with a precision machining shop in Dandenong that invested about $45,000 AUD in vibration monitoring for their three highest-value CNC machines. Each of those machines bills out at roughly $280 per hour. An unexpected failure means not just lost production — it means air-freighting a replacement spindle from Japan at $12,000-$18,000 a pop, plus two to three weeks of downtime.

They had two unplanned spindle failures in the eighteen months before installing the system. After installation, the sensors caught bearing degradation early enough to schedule replacements during their Christmas shutdown. The owner told me the system paid for itself in the first year.

That’s the sweet spot for predictive maintenance: expensive equipment with high replacement costs, long lead times for parts, and significant revenue impact from downtime. If a machine going down costs you $5,000 a day or more in lost production, the math works pretty quickly.

Platforms like SafetyCulture offer inspection and monitoring tools that can complement predictive maintenance systems, especially for the broader workplace safety picture that small manufacturers need to keep on top of.

When it probably doesn’t

Here’s the flipside. I also talked to a sheet metal fabrication shop in western Sydney. They’ve got press brakes, a laser cutter, and some basic forming equipment. Their maintenance guy — one bloke who’s been there fifteen years — already knows what every machine sounds like when something’s off. Replacement parts are mostly standard bearings and hydraulic components, available locally within 48 hours.

They looked at predictive maintenance, got quoted $55,000 for the initial setup, and the owner’s response was: “I could buy a whole spare press brake for that.”

He’s not wrong. When your equipment is relatively simple, replacement parts are cheap and available, and your experienced staff can already spot problems early, the ROI on a sensor-based system just isn’t there.

The same goes for operations where the process itself is simple enough that scheduled maintenance covers most failure modes. If you’re replacing wear parts on a known schedule and your unexpected downtime is already under 2-3% of operating hours, you’re not going to see dramatic improvements from adding sensors.

The data problem nobody mentions

Here’s something vendors rarely bring up: predictive maintenance systems need data to work. Lots of it. The machine learning models that detect anomalies need a baseline — typically three to six months of normal operation data before they can reliably flag problems.

During that learning period, you’re paying for the system but not getting much predictive value from it. For a small manufacturer who needs to show ROI quickly, that ramp-up period can be frustrating.

And the data quality matters enormously. If your production varies a lot — different materials, different speeds, different shifts with different operators — the system has more noise to filter through. One shop told me it took nearly nine months before their system stopped generating false alarms every time they switched from aluminium to steel jobs.

My honest recommendation

If you’ve got equipment where a single failure event costs more than $15,000-$20,000 AUD (including lost production, parts, and emergency labour), and you experience at least two or three unplanned failures per year, predictive maintenance will probably pay for itself within 18-24 months.

If your biggest downtime risk is a $2,000 bearing replacement and your maintenance bloke has things pretty well sorted with scheduled services, save your money. Spend it on training, spare parts inventory, or upgrading that one machine that’s been causing grief for years.

The technology is genuinely impressive. But impressive technology that doesn’t deliver a return is just an expensive hobby. Be honest about your actual downtime costs before you sign anything.