The Real IIoT ROI Numbers Plant Managers Need
Shell turned an $87,000 sensor bet into more than $1 million in returns. Most industrial IoT projects never get that far. Here’s the actual math behind the ones that do.
In 2019, Shell wired up an aging oilfield with $87,000 worth of vibration and pressure sensors. The payout: over $1 million in avoided downtime and deferred maintenance, a return that would make any CFO sit up. That story gets quoted constantly in industrial IoT (IIoT) marketing decks. What doesn’t get quoted nearly as often: 84% of IoT projects never make it past the pilot stage, and more than a quarter of those stay stuck there for over two years.
If you’re a plant manager weighing a predictive maintenance rollout, both of those facts matter. This piece walks through what unplanned downtime actually costs, why so many IIoT projects stall before they scale, and how to build an ROI case that survives contact with a skeptical board.
What Unplanned Downtime Actually Costs You
Start with the number that justifies everything else: unplanned downtime costs U.S. industrial manufacturers roughly $50 billion a year. The average large plant loses about $253 million annually to breakdowns. For Fortune 500 manufacturers, that works out to $2.8 billion a year, close to 11% of revenue, vanishing into machines that stopped when they weren’t supposed to.
It’s not a rare event, either. 82% of manufacturers reported unplanned downtime in the past three years, and the average factory loses roughly 800 hours a year to breakdowns that, in theory, sensors could have flagged in advance.
Here’s where the hardware math has flipped in the plant manager’s favor. Basic industrial IoT sensors now run under $50 a unit, down from over $200 five years ago. A North American plant runs an average of 365 IIoT devices today. Cost is no longer the bottleneck it was in 2019. That changes the calculus: a small, targeted pilot on two or three critical machines is now cheap enough to fund out of an existing maintenance budget, which means you don’t need to wait for a capex approval cycle to test the idea.
Why Most IIoT Projects Never Scale
So if the hardware is cheap and the downtime math is brutal, why isn’t every plant running on sensors already?
Because most projects stall. A McKinsey study, still widely cited years later, found 84% of industrial IoT initiatives get stuck in what the industry now calls “pilot purgatory,” and 28% of those stay there for more than two years. More recent tallies put the failure-to-scale rate above 60%, and Cisco’s older but still-referenced figure puts pilot survival at just 26%.
“It is not purgatory, it is hell!” Manufacturing CEO, quoted by Stephan Liozu, Chief Value Officer at Zilliant and adjunct professor at Case Western Reserve University’s Weatherhead School of Management, via IndustryWeek
Liozu isn’t a fringe voice here. His point, backed by McKinsey’s own data, is that most IIoT programs collapse under the weight of vague mandates like “improve efficiency” instead of a specific, measurable KPI tied to a named line and a real downtime-cost baseline.
Adoption data backs up the stall-out pattern. According to MaintainX’s 2025 State of Industrial Maintenance survey, predictive maintenance adoption among maintenance teams actually dropped, from 30% in 2024 to 27% in 2025. That’s not the smooth upward curve you’d expect from the marketing. Only 46% of manufacturers have deployed IIoT at the facility level at all, and per McKinsey’s more recent numbers, just 25 to 30% of large manufacturers have scaled beyond a pilot to an enterprise-wide rollout.
Watch this number: Global IIoT market-size estimates for 2026 range from roughly $190 billion (Mordor Intelligence) to over $500 billion (Precedence Research), depending entirely on how each firm defines the market’s scope. When a vendor pitches you urgency based on “the $500 billion IIoT market,” ask which definition they’re using. The spread alone should make you skeptical of any single headline figure used to justify a purchase decision.
Jeff Winter, VP of Business Strategy at Critical Manufacturing, put the technical side of the problem bluntly at IIoT World Days 2025: consumer AI succeeds at around 95% accuracy, but industrial models need a near-zero margin for error, 99.5% or higher. That’s a much higher bar than most of the AI hype cycle accounts for, and it’s a big part of why pilots that look great in a demo fall apart on a real production line.
What Actually Works: Two Real Cases
Strip away the invented statistics you’ll find floating around IIoT marketing content (there’s no verified source for some of the dollar figures companies get credited with online, so treat any suspiciously precise claim with caution) and two real, attributable cases hold up.
Shell: $87,000 In, $1 Million Out
Shell’s oilfield asset-monitoring project is the cleanest real-world proof point available. An $87,000 sensor investment on aging equipment generated over $1 million in returns through avoided downtime and deferred maintenance, according to IoT World Today’s reporting. It’s not a hypothetical ROI model. It’s a documented result from a company that had every incentive to keep quiet if the numbers hadn’t worked out.
GE: The Real Numbers Behind the Headline
GE Digital’s Global Electricity Monitoring and Diagnostics Center processes over 200 billion data tags daily from a million sensors across 5,000 assets in power plants in more than 60 countries. Bill Ruh, then-CEO of GE Digital, put the actual, on-record result this way when the platform launched:
“These analytics provide GE Digital with the unique ability to reduce unplanned downtime by up to 5 percent, reduce false alarms by up to 75 percent, and reduce operations and maintenance costs by up to 25 percent.” Bill Ruh, then-CEO, GE Digital, GE News press release, October 2017
Notice that’s 5%, not the round 20% figure that circulates in some secondhand summaries. It’s a smaller number, but it’s GE’s own, and it comes with a false-alarm reduction and O&M cost figure attached that most retellings leave out entirely.
Predictive Maintenance: The Realistic Range
Zooming out from single-company cases, here’s where independent research firms land on predictive maintenance’s actual impact:
| Metric | Range | Source |
|---|---|---|
| Maintenance cost reduction | Up to 25% | Deloitte |
| Uptime increase | 10% to 20% | Deloitte |
| Downtime reduction (upper bound) | Up to 50% | McKinsey & Co. |
| Asset lifespan extension (upper bound) | Up to 40% | McKinsey & Co. |
Treat the upper-bound figures as ceiling cases, not defaults. A well-scoped, single-line pilot is far more likely to land near the lower end of these ranges in its first year.
Building an ROI Case That Survives the Board
Here’s the actual formula, stripped of vendor gloss: ROI equals annual savings minus annual program cost, divided by annual program cost, times 100. Annual savings breaks down into avoided downtime cost plus avoided emergency maintenance spend, measured against a pre-deployment baseline you establish before you install a single sensor.
A few things the data says you need to get right:
- Use your own downtime-cost-per-hour, not an industry average. The $50 billion industry figure is a scale-setter, not an input for your specific calculator.
- Stress-test your assumptions by plus or minus 30% before presenting to a board. This is standard practice recommended in IoT World’s own project-scoping framework, and it heads off the “your numbers were too optimistic” objection before it happens.
- Budget for realistic retrofit costs. End-to-end IIoT retrofits run anywhere from $1 million to over $10 million depending on facility size and complexity, per Emergen Research. A sub-$50,000 sensor pilot is a very different financial commitment than a plant-wide rollout, and conflating the two in a pitch deck is a fast way to lose credibility.
- Set a payback-period expectation that matches your scope. Well-scoped single-line predictive maintenance pilots often show payback in 6 to 18 months. Enterprise-wide rollouts typically take two years or more, and that’s exactly the scope where the pilot-purgatory failure rate climbs.
Our read: the smartest move for most plant managers right now isn’t the big enterprise-wide business case. It’s the small, self-funded pilot on one or two lines, with a hard KPI and a real downtime-cost baseline attached, that proves the model before anyone has to ask a board for millions.
Frequently Asked Questions
How do you calculate ROI for an IIoT project?
ROI equals annual savings minus annual program cost, divided by annual program cost, times 100. Savings combine avoided downtime cost and avoided emergency maintenance spend, measured against a pre-deployment baseline established before installation begins.
How much does unplanned downtime cost manufacturers?
Unplanned downtime costs U.S. industrial manufacturers roughly $50 billion annually. The average large plant loses about $253 million a year, and Fortune 500 manufacturers lose around $2.8 billion annually, close to 11% of revenue.
What percentage of IoT projects fail to scale?
Estimates vary, but a widely cited McKinsey finding shows 84% of industrial IoT projects remain stuck in pilot mode, with more than a quarter stalled for over two years, a pattern the industry calls “pilot purgatory.”
How much can predictive maintenance reduce downtime?
Deloitte and McKinsey benchmarks put predictive-maintenance-driven downtime reduction between 25% and 50%, with maintenance cost savings of 10% to 25%. Results vary heavily by industry, equipment type, and deployment maturity.
What’s a realistic payback period for an industrial IoT project?
Well-scoped, single-line predictive maintenance pilots often show payback within 6 to 18 months. Enterprise-wide rollouts typically take two years or longer, and a meaningful share stall out indefinitely before reaching that point.
What to Watch Next
Three things worth tracking over the next 6 to 18 months: whether predictive maintenance adoption recovers from its 2025 dip or keeps sliding, whether sub-$50 sensor pricing pulls more mid-size manufacturers past the 46% facility-level adoption mark, and whether the gap between market-size hype and the 25 to 30% enterprise-scaling rate starts to close.
None of that changes the math you need today. Get your own downtime-cost baseline, run the small pilot, stress-test the assumptions, and let the numbers, not the vendor slide deck, make the case.
Want the next breakdown like this one delivered straight to you? Subscribe to The Neural Loop for weekly analysis on industrial AI, robotics, and the infrastructure behind it.