Operator Playbook · Operations
Predictive Maintenance That Actually Works
Lindsay Corporation built it on the failure that cost them the customer. GE built it on the failure that was easiest to model. One business now has an $80M MENA subscription deal. The other one wrote down $22 billion.
The two predictive-maintenance stories you need to know
In 2015 GE announced it was going to be a “top ten software company by 2020.” The vehicle was Predix — an industrial Internet-of-Things platform that was supposed to predict equipment failure across power plants, jet engines, locomotives, and refineries before it happened. Jeff Immelt put $7 billion of capex behind it. The Wall Street Journal counted total writedowns and impairments tied to the GE Digital strategy at over $22 billion by 2018. By 2020, GE Digital had been quietly carved out, sold off in pieces, and effectively rebadged as a verticalized OT software business. Predix, the product, is still around. The thesis is dead.
In 2024, Lindsay Corporation — a $620M sprinkler manufacturer in Omaha that the AI press has barely heard of — closed an $80M international subscription deal in MENA based largely on the strength of an AI predictive-maintenance product called Smart Pivot. By Q2 of fiscal 2026, that subscription thesis was the load-bearing narrative on the company’s earnings calls. CEO Randy Wood now talks about Lindsay as a “Tech-as-a-Service” company without flinching.
These are two predictive-maintenance stories, fifteen years apart, both with serious technology, both with real money behind them. One worked. One didn’t. The difference is not in the algorithm. The difference is in what failure each company decided was worth predicting.
If you run a small-cap or mid-cap business with hardware in the field, this playbook is the thing you read before you sign the predictive-maintenance contract.
What GE got wrong
Predix’s pitch was that any industrial machine — a turbine, a pump, a pivot, a press — could be instrumented, modeled, and predicted. Apply machine learning to vibration data and oil chemistry, and you’d know which bearing was about to fail, on which asset, in which week. Avoid the unplanned downtime. Smooth the maintenance curve. Save the customer money. Charge a subscription for it.
The technology basically worked. There are operating papers and case studies showing Predix could detect anomalies that correlated with future failures. That part wasn’t fake.
Three things were fake.
One — the customer benefit was the wrong benefit. Predix was sold as smoothing the maintenance curve, which is an operations-finance benefit. The CFO’s eyes light up. The plant manager’s eyes do not. Plant managers do not lie awake at night worrying about smoothing the maintenance curve. They lie awake worrying about the one specific failure that, when it happens, gets them on a 4am call with the CEO. Predix did not differentiate between failures that smooth the curve and failures that get you fired. Every prediction looked the same in the dashboard.
Two — the ROI required the customer to buy the whole platform. Predix was a horizontal play. It worked best when a customer instrumented their entire asset base, fed everything into a unified data lake, and ran every prediction through one model. That’s a five-year IT project at a Fortune 500. The customer who buys it has to bet his career on it before he sees a dollar of value.
Three — the customer had to surrender his data to GE. Industrial customers, especially the ones running power plants and refineries, do not voluntarily put their highest-value operating data into a vendor’s cloud. They will publicly say they do. They will sign deals saying they do. They will not actually do it. The predictive accuracy of Predix was permanently throttled by the data customers refused to share. GE could not solve that problem because it was the prime contractor, not the customer.
The combination meant that for any given prospect, the value of the platform was theoretical and the cost of adoption was real. The math never closed. The total writedown on GE Digital is in the public record. The strategic exit is in the public record. The lesson is in the public record. Most small-cap and mid-cap operators trying to put predictive-maintenance AI into their business are about to repeat the smaller version of GE’s mistake.
What Lindsay got right
Randy Wood’s team at Lindsay did the opposite of every Predix axiom, on purpose.
One — they picked the failure that costs the customer the customer. Smart Pivot does two things and only two things. It predicts gearbox and motor failures on a center-pivot irrigation machine. And it detects crop-health issues — disease, nutrient deficiency, pest pressure — using leaf-level imagery from Taranis aircraft.
Both of those are the moment a farmer loses his year’s crop. Not the moment maintenance gets expensive. The moment the customer relationship ends. The dollar value of preventing a single mid-July gearbox failure on a center-pivot is not the cost of the gearbox. It is the future revenue from a farmer who otherwise would have switched to Valmont. Lindsay built the AI around that math, not around a smoothed maintenance curve.
Two — they built it around installed equipment, not net-new sales. Lindsay had two pre-existing software products — FieldNET and FieldNET Advisor — that had been in the field since 2007. Smart Pivot was layered on top of that installed base plus new sensors. Customers did not have to replace their pivots to start getting value. They could subscribe and turn on monitoring on equipment they already owned. The cost of adoption was a software signup, not a five-year capital plan.
Three — they bought the data network instead of demanding it. In April 2024 Lindsay took a 49.9% stake in Pessl Instruments, an Austrian company with a million sensors and 50,000 active customers. Lindsay did not ask its farmers to share data. It bought the company that already had the sensors in the ground. The data advantage now compounds for Lindsay every time a Pessl customer turns on a station, regardless of whether Lindsay sold them the pivot.
Four — they aligned with the dealer service network instead of disintermediating it. Smart Pivot does not auto-dispatch a Lindsay-employed technician. It pings the local dealer service tech, who is the relationship the farmer trusts. Lindsay’s revenue model preserves the dealer’s margin. The dealer becomes an active distribution channel for the AI subscription, not a casualty of it. This is the move GE never figured out — predictive maintenance threatens the field service organization’s headcount, and the field service organization will sandbag the deployment unless they’re explicitly cut into the upside.
The result, four years in: Smart Pivot is the load-bearing thesis behind a public valuation rerating. The MENA deal is on the record. International buyers paying a premium for the software layer means there is a real, portable software business inside a steel company. Whether that ARR line ever gets broken out separately on Lindsay’s P&L is a matter of investor relations strategy. The thesis works.
The framework — five rules for predictive maintenance that ships
If you are about to sign a predictive-maintenance contract for your small-cap or mid-cap business, run it through this filter.
Rule one — predict the failure that costs you the customer, not the failure that’s easiest to model. Vibration anomalies are easy to model. The anomaly that loses you the customer is what matters. They are not always the same. Pick the failure based on commercial impact, not modeling convenience.
Rule two — make the smallest unit of value the customer can buy be small. A single asset. A single subscription. A single line in the existing contract. Not a platform. If the customer has to “implement Predix” before they get value, you have not designed a product. You have designed a five-year IT project, and the IT project will lose to the politics inside the customer’s company.
Rule three — the data moat lives in your customer’s plant, not your cloud. Architect the system so the customer’s operating data stays where the customer is comfortable. Push the model to the data, not the data to the model. The companies that figure out edge inference and on-prem deployment are going to outsell the companies that demand a cloud upload, every single time.
Rule four — preserve the field service relationship. The dispatcher, the local dealer, the senior tech — they are the relationship the customer trusts. A predictive-maintenance system that disintermediates them gets sandbagged. A predictive-maintenance system that makes them more effective and more profitable gets evangelized. The field service organization is your distribution channel for AI. Treat it that way.
Rule five — point at the first big subscription win publicly, in plain numbers. When Lindsay closed the MENA deal, Wood did not bury it. He cited the dollar number on every earnings call until analysts could recite it. The narrative drove the rerating. If your predictive-maintenance product earns a real subscription win, you have to evangelize it publicly. The market does not price in narratives it cannot see.
A sentence to take with you
Predictive maintenance is not a technology problem. It is a which failure do you care about problem.
Pick the failure that costs you the customer. Build a small product around it. Put it in the dealer’s hands. Ship it.
How we do this for clients
The Ground-Up Workshop is the alignment step that turns a generic predictive-maintenance pitch into a specific deployment plan. Two days in person with your operations leadership, dealer principals if relevant, and a small group of subject-matter experts. We identify the failure that costs you the customer, score the data moat, and walk you out with a target operating model and a 90-day plan.