Friction consumes about a fifth of the world's energy and wear another three percent, and the people who prevent them have never been paid in proportion to what they save. When a better oil, a sensor or a bearing makes a machine last longer, who keeps the saving: the customer, the supplier of the prevention, or the maker of the parts that no longer need replacing? A model of that bargain, tested against SKF's and Fuchs's own books, and four dated tests that will tell us whether it is right.
The customer keeps most of it, and the company that prevents a failure earns less from preventing it than the parts seller earned from repairing it. That is the finding, and the arithmetic behind it is simpler than the technology. In our worked example a plant that believes the whole of a prevention program's savings can afford to pay the vendor $58,896 a year; the vendor needs $24,638 to break even; the $34,258 between them is the bargain. Let the plant believe only half of the downtime saving, as a manager who has never seen the failure that did not happen usually does, and the bargain shrinks to $4,258. Below 43% belief there is no fee both sides will sign, and walking away is the rational thing to do. Nothing about the technology changed in that sentence. Only the belief did.
A paper on Vista's reading list and a model in its Decision Library say who keeps what is left. Kenneth Arrow's replacement effect, written as a price: an incumbent that sells the parts needs a higher fee than a challenger, by exactly the parts contribution it would lose, $8,000 a year in the example, so a challenger can always underbid it. David Teece's profiting from innovation: the challenger rarely wins anyway, because the assets a prevention service needs to reach a customer, the installed base, the distributors (SKF's Industrial business has more than 7,000) and the service access, belong to the incumbent. The money goes to whoever owns the access, not to whoever owns the chemistry. SKF's own books show the result: a condition-monitoring category of between 2.3% and 2.5% of sales, revenue recognized over time "not deemed material," and about 6,000 metric tons of bearings remanufactured in a company that buys 415,000 metric tons of steel a year. Four tests on SKF, Fuchs, Quaker Houghton and Schaeffler will be scored on April 30, 2027.
This inquiry began with a column. On October 7, 2026, Anjana Ahuja wrote in the Financial Times, under the headline "Tribology makes the economic world go round," that the science of friction and wear "plays a central role in making industry run more smoothly." Tribology, the study of surfaces that rub, is sixty years old as a word and as old as the wheel as a problem. Her column makes the case for prevention. It does not say who gets paid for it.
I read it with my AI research assistants with great interest, because I'm also a director at the Failure Analysis Institute (failureanalysisinstitute.com), which provides experts to sort out failed machines and materials. The column was so beautifully written and so fascinating that I found myself wanting to dive into the financial details and the math (it is a Financial Times piece, after all!): a list of the companies, the savings in numbers, and some sense of when the superlubricity of the laboratories might reach a factory floor. So we built the numbers ourselves: a plant manager's five-year budget for a predictive-maintenance program, the supplier's floor beneath it, the arithmetic of a lubricant maker whose customer suddenly uses half as much oil, and the arithmetic of a monitor that cries wolf. I sent all of it to Vista's Desk and asked it to tell me what was wrong. It found nothing wrong with the numbers and two things missing: a test that could fail, and a model that would say who keeps the saving. This inquiry supplies both.
The question is older than the column, and the people who asked it first deserve their credit. Edward Reiskin, Allen White, Jill Kauffman Johnson and Thomas Votta described in 1999 how a chemical supplier paid per car door painted, rather than per gallon delivered, comes to want its customer to use less chemical, and how gain-sharing splits the saving ("Servicizing the Chemical Supply Chain," Journal of Industrial Ecology). Jose Guajardo, Morris Cohen, Sang-Hyun Kim and Serguei Netessine showed in 2012 that aircraft engines maintained under performance-based contracts were 25% to 40% more reliable than engines maintained by the hour. Henk Akkermans, Rob Basten, Quan Zhu and Luk Van Wassenhove found in 2024 that monitoring services fail to grow for a reason no bargain explains: customers who maintain their machines too carefully never produce the failures a predictor must learn from. What is ours is the bargain itself, written as a ceiling and two floors; the incumbent's floor read as Arrow's replacement effect; and David Teece's forty-year-old question, who profits from an innovation, asked of prevention, which, as far as we could find, nobody has asked it of before.
The largest bearing company in the world was founded by a maintenance man who was tired of replacing the same part.
Sven Wingquist was the maintenance engineer at Gamlestadens Fabriker, a textile mill in Gothenburg, Sweden, where the ground under the mill shifted, the long shafts that drove the looms went out of line, and the bearings on those shafts kept failing. Every failure stopped a line. His answer was a bearing whose two rows of balls ran in a spherical outer race, so that a shaft could tilt a little without binding: the self-aligning ball bearing. He applied for a Swedish patent on February 16, 1907, and received it on June 6, patent number 25406. The same day he applied, the mill set up a subsidiary to make the thing, with Wingquist as managing director: AB Svenska Kullagerfabriken, SKF.
Here is the rub. The company born from one man's refusal to keep replacing a part now earns, by its own account, more than half of its Industrial business's revenues from the aftermarket, from the parts that wear out and the service that follows. "Every machine eventually needs maintenance, repair or optimization," its 2025 annual report says, and about 17,000 distributors, more than 7,000 of them in the Industrial business, stand ready when it does. SKF spends real money and real engineering on preventing failures; it also sells the replacements when prevention fails. Both are true, and the second pays better than the first. When Hamlet says "there's the rub," he is borrowing from the bowling green, where a rub is anything that deflects the ball from the line its bowler intended. In a bearing the rub is friction. In a business plan the rub is the fact that the firm best placed to prevent a failure is the firm that profits when it happens, which, incidentally, is the whole raison d'être of the Failure Analysis Institute!
Leonardo da Vinci wrote down the first law of friction on a page the size of a playing card, 92 by 63 millimeters, in red chalk over a sketch of blocks pulled across a pulley: "friction is of double the effort for double the weight." Ian Hutchings of Cambridge, who went through the notebooks page by page in 2016, dates the note to 1493, in Codex Forster III, one of his notebooks now at the Victoria and Albert Museum, on the front of leaf 72, and shows that over the next twenty years Leonardo arrived at the two laws that bear a French name today: friction is proportional to the load pressing two surfaces together, and it does not depend on how large the surfaces are. From about 1497 he settled on a quarter for the ratio of friction to load; Guillaume Amontons, who rediscovered the same two laws in Paris in 1699 with samples smeared in pork lard, found about a third, and Charles-Augustin de Coulomb separated the friction of starting from the friction of sliding in a prize essay of 1781, printed in 1785, after which the subject carried his name for two centuries. Leonardo's work, Hutchings adds, "had no influence on the development of the subject over the succeeding centuries." He knew what friction cost. Nobody acted on it.
The Romans had the economics before anyone had the physics. Lucretius, in the first book of De Rerum Natura, listed the evidence that matter is made of invisible parts: "A ring upon the finger thins away / Along the under side, with years and suns; / The drippings from the eaves will scoop the stone; / The hooked ploughshare, though of iron, wastes / Amid the fields insidiously." Ovid borrowed all three images in the same order two generations later. The ring, the roof and the plough are the oldest inventory of wear, and the oldest statement of the problem this inquiry is about: the loss is slow, it is invisible day to day, and it is paid for by whoever owns the ring.
The word came much later, and from a committee. In 1964 Lord Bowden, a minister of state at Britain's Department of Education and Science, asked H. Peter Jost, an engineer and industrialist, to lead a working group on the teaching of lubrication. The report went to the minister in October 1965 and was published on March 9, 1966. It needed a name for a field that had none, and Jost, by his own account, took the problem to the editor of the Oxford English Dictionary, who suggested the Greek tribos, rubbing. The Jost Report's number is the one still quoted sixty years later: "a conservative estimate" of savings available to British industry of "about £515 million" a year, "the biggest single item being savings in maintenance and replacement costs of £230 million." British output in 1965 was about £36.9 billion, so the claim was that friction and wear were costing the country about 1.4% of everything it made, and later studies in Germany, America and China settled on 1% to 1.4% of national output as the figure for an industrial country. Jost's estimates were made by a committee of lubrication people, and they have been repeated far more often than they have been re-measured. They are also the best we have had for most of the field's life.
The modern figure comes from Kenneth Holmberg and Ali Erdemir, who added up the world's tribological contacts in 2017 for the journal Friction: about 23% of the world's total energy consumption, 119 exajoules, goes to overcoming friction (20%) or to remanufacturing and replacing worn parts (3%). Better tribology could cut those losses by 40% over fifteen years, which would save 8.7% of the world's energy and about 1.4% of its output. Notice the split. Fifty years ago, Holmberg and Erdemir write, wear was thought to hold 95% of the prize; today they put 74% of the available cost savings on friction, which is an energy bill, and 26% on wear, which is a replacement bill. The Jost Report had replacement at 45% of its total. The prize that matters to an investor, the part that lands on somebody's income statement as a part not sold, is the smaller part of the whole, and the question of who keeps it is the one nobody had worked out.
Begin with the buyer, because the buyer decides. A plant manager considering a predictive-maintenance program, sensors on the bearings, software that reads them, a vendor who answers the phone, has to put five years of savings against five years of costs and one check up front. Our illustration uses round numbers, labeled as assumptions, in dollars: $30,000 a year of energy saved as machines run with less friction; $10,000 a year of maintenance saved; and the big one, the failures avoided, which we write as a 12% chance each year of a failure that would cost $500,000, or $60,000 a year in expectation. Against that: an $80,000 implementation up front; a $30,000 annual fee to the vendor; $15,000 of internal upkeep; and $5,000 a year set aside for the interventions that turn out to have been unnecessary. Five end-of-year periods, a 10% discount rate, nothing left at the end.
The model is two lines. The first is the plant's net benefit in a year:
Net benefit = E + M + c x p x L - F - U - B
E is the energy saving, M the maintenance saving, p the yearly chance of the avoidable failure, L what it would cost, F the vendor's fee, U the plant's own upkeep and B the budget for false alarms. The letter c is the one that matters. It is the share of the downtime saving the plant actually believes, from zero to one, and we will come back to it. The second line turns five years of net benefit into one number today:
NPV = Net benefit x af - K
K is the up-front cost and af is the annuity factor, the present value of a dollar a year for five years at 10%: (1 - 1.1^-5) / 0.10 = 3.79. With everything believed, c = 1, the net benefit is $30,000 + $10,000 + $60,000 - $30,000 - $15,000 - $5,000 = $50,000 a year, worth $189,539 today, and the program clears its $80,000 up-front cost with $109,539 to spare. With half the downtime saving believed, c = 0.5, the net benefit is $20,000, worth $75,816 today, and the program loses $4,184. Nothing in the plant changed between those two sentences; the sensors are the same sensors. What changed is how much of a failure that has not happened the manager is willing to count.
The same two lines give the most the plant can pay. Set the NPV to zero and solve for the fee:
Affordable fee = (E + M + c x p x L - U - B) - K / af
The up-front cost spread over five years is $80,000 / 3.79 = $21,104 a year. So the plant that believes everything can afford a fee of $100,000 - $20,000 - $21,104 = $58,896 a year, and the plant that believes half can afford $28,896. A small table of trial values shows how steeply the ceiling falls with belief:
| Share of downtime saving believed | Expected avoided loss a year | Affordable annual fee |
|---|---|---|
| 100% | $60,000 | $58,896 |
| 75% | $45,000 | $43,896 |
| 50% | $30,000 | $28,896 |
| 43% | $25,742 | $24,638 |
| 25% | $15,000 | $13,896 |
| 0% | $0 | -$1,104 |
The highlighted row is the one the supplier cares about, and we will see why in a moment.
Why would a manager believe half? Because the saving is a counterfactual, a figure that exists only by comparison with what did not happen. The energy bill arrives every month and can be compared with last year's. The failure that was prevented leaves no invoice. Nelson Repenning and John Sterman of MIT gave the problem its best title in 2001: "Nobody Ever Gets Credit for Fixing Problems that Never Happened." A plant with spare capacity can make up lost output on the next shift, so its avoided loss is smaller than the headline; a plant whose maintenance budget and energy budget belong to two different managers can have a program that pays the company and fails both budgets. Those are the two reasons the Desk separated when it reviewed the draft: the first is information, the second is incentive, and the arithmetic below tells them apart.
The same blindness works in reverse, and aviation has the record of it. On January 31, 2000, Alaska Airlines Flight 261 dived into the Pacific off Point Mugu, California, and all 88 people aboard died. The jackscrew that moved the MD-83's horizontal stabilizer turned in a nut whose threads were made to wear before the screw did, and the whole assembly depended on grease. In 1985 Alaska lubricated it every 700 flight hours; by 1996, through a series of separately approved changes, every eight months, about 2,550 flight hours, and the end-play check that measured the wear had stretched from every 5,000 flight hours to about every 9,550. The threads wore through. Each extension saved a little maintenance that showed on a budget, and the risk it added showed nowhere until the day it was the whole of the cost. After the accident, airworthiness directives set the lubrication at every 650 flight hours and the end-play check at every 2,000.
Now the seller. A vendor of prevention has costs of its own: delivering the service, $12,000 a year in our illustration; reinvesting in the software, $2,000; onboarding the customer, $10,000 once; and one cost that most analyses leave out, the parts the vendor no longer sells. If the vendor is the company that used to sell the plant its bearings and oil, a machine that lasts longer is a sale it does not make. Call that lost contribution P, $8,000 a year in the illustration. The supplier's position after five years is:
Supplier NPV = (F - D - P - R) x af - O
D is delivery, R reinvestment and O onboarding. At the $30,000 fee the supplier's NPV is ($30,000 - $12,000 - $8,000 - $2,000) x 3.79 - $10,000 = $20,326, which is a living. Set the NPV to zero and the fee floor falls out:
Fee floor = D + P + R + O / af
That is $12,000 + $8,000 + $2,000 + $2,638 = $24,638 a year for the incumbent, the last term being the $10,000 of onboarding spread over five years. For a challenger that never sold the plant a part, P is zero, and the floor is $16,638.
The $8,000 between the two floors has a name. In 1962 Kenneth Arrow showed that a monopolist gains less from an innovation than a newcomer would, because the innovation replaces profit the monopolist was already earning; economists have called it the replacement effect ever since. In our two lines it is simply a line item. The company that sells the replacement parts must charge $8,000 a year more for preventing their replacement than a company with no parts to lose, and if it charges less it is paying its own customer to buy fewer bearings. A challenger can always underbid an incumbent by the incumbent's parts margin, and the incumbent cannot follow without eating that margin twice.
The purest expression of the replacement effect is a cartel. On December 23, 1924, the leading lamp makers of the world met in Geneva and formed the Phoebus cartel; by early 1925 it had set the life of a household bulb at 1,000 hours, down from the 1,500 to 2,000 hours that had been common, and it fined any member whose bulbs lasted longer. The average life of the cartel's reference bulbs fell from about 1,800 hours in 1926 to 1,205 hours by 1934. Markus Krajewski, who read the Osram archives and the files of the American antitrust case against GE, told the story in IEEE Spectrum in 2014. There is a defense, and it should be heard: Britain's Monopolies Commission concluded in 1951 that "there can be no absolutely right life" for a bulb, since a hotter filament gives more light per watt and burns out sooner, and accepted the makers' evidence that 1,000 hours was the best compromise available. Krajewski's reply is that the archives show the motive was sales. Jeremy Bulow gave the theory its modern form in 1986: a monopolist that sells a durable good, rather than renting it, chooses less durability than its customers would. In Livermore, California, a carbon-filament bulb donated to the fire department in 1901 is still burning at about four watts, with only three interruptions in 125 years, the longest about a week in 1937. It is the longest-running rebuke to the cartel, and it is also a bulb nobody could sell.
Through laziness the roof caves in, says Ecclesiastes (10:18, Berean Standard Bible), and in the hands of the idle, the house leaks. The preacher assumed the owner of the roof would pay for the leak. He did not consider a roofer who sold shingles by the bundle.
Put the ceiling over the floor and the bargain appears. The zone is the affordable fee less the fee floor:
Zone = Affordable fee(c) - Fee floor
At full belief it is $58,896 - $24,638 = $34,258 a year; at half belief, $28,896 - $24,638 = $4,258. There is a belief at which the zone closes, and it can be solved for exactly: the plant must believe enough of the downtime saving to cover the supplier's floor, the up-front cost as an annuity, and its own upkeep and false-alarm budget, less the savings it can see on its bills.
Critical belief c* = (Fee floor + K / af - (E + M - U - B)) / (p x L)
For the incumbent that is ($24,638 + $21,104 - $20,000) / $60,000 = 42.9%. For the challenger with the lower floor it is 29.6%. The chart draws it.
The model gives a rule for reading a market where prevention is not bought. Below the critical belief, no fee clears both sides, and the manager who walks away is right to. The cure is not a better sensor; it is verification, a guarantee, a trial or an auditor who can turn a counterfactual into an invoice. The evidence that verification works is Guajardo and his colleagues' finding on aircraft engines: when the maker was paid for hours of flight instead of hours of labor, reliability rose 25% to 40%, because the maker now had reason to prevent what it had been paid to repair. Above the critical belief, a zone exists, and if no contract is signed there the cause lies elsewhere: a budget split between the manager who pays the fee and the manager who saves the energy; a supplier whose sales force is still paid by the liter; a guarantee whose exclusions, counterparty risk, integration cost or claim on a capital budget already spent make it worth less than its face; or, as Akkermans and his colleagues found at four equipment makers, a predictor with nothing to learn from, because the customer's machines are maintained so carefully that they never fail. Several explanations, then, for the same empty order book, and a low take-up of guaranteed savings does not by itself tell them apart. That needs comparisons in which one thing varies at a time, verification with the budgets held fixed and the budgets with verification held fixed, which plant records can supply and annual reports cannot.
Even the most famous of these contracts stops short of paying for results. Rolls-Royce has sold jet engines by the flying hour since 1962, when Bristol Siddeley, which it bought four years later, offered the service for the Viper engine on a small business jet, and the industry has called it Power-by-the-Hour ever since. Tonči Grubić and Ian Jennions went looking in 2018 for contracts that pay a supplier for the customer's economic result and found almost none; even Power-by-the-Hour sells availability, an engine that is ready, not a flight that paid. Irene Ng and her colleagues explained why from the defense industry: the outcome depends on what the customer does with the engine, and no supplier can price what it cannot control. The contracts in this inquiry are of that kind. They sell uptime and oil-free furnaces. The plant keeps the profit from what it does with them.
David Teece has been my friend for more than thirty years, and the question he asked in 1986 has outlasted most of the industries it was asked about. Why, he wanted to know, do innovators so often fail to profit from their innovations, while imitators and the owners of other assets do? His paper, "Profiting from Technological Innovation," gave three answers. First, the appropriability regime: how well the innovation can be protected, by law or by the sheer difficulty of writing down the know-how. Second, the dominant design: once an industry settles on one, competition turns to price and scale. Third, and this is the one that decides our case, the complementary assets: the manufacturing, the distribution, the service and the approvals an innovation needs to reach a customer, which the innovator may or may not own. His example was EMI, which brought the CT scanner to America in 1973 with a technology nobody else had and was out of the scanner business within about eight years, because GE's medical division owned the hospital sales force, the service engineers and the installed base, and EMI's patents did not hold. A year later Richard Levin, Alvin Klevorick, Richard Nelson and Sidney Winter asked 650 research executives in 130 lines of business what actually protected a new product, and the answers put lead time and sales-and-service effort above patents in most of them.
Now ask Teece's question of prevention. The innovation is a sensor that hears a bearing begin to fail, an algorithm that reads it, a chemistry that does not evaporate at 300 degrees. The appropriability regime is weak: sensors are commodities, algorithms travel, and Teece himself argued in 2018 that enabling technologies, the ones that help everyone, are the hardest of all to profit from because their value spills to whoever owns the platform they run on. There is no dominant design yet. And the complementary assets, the ones the innovation needs to reach a plant floor, are the installed base of bearings, the 7,000 industrial distributors who already call on the plant, the application engineers who know which machine fails how, and the access to the plant's own data. Those belong to the incumbent. The sensor company is EMI. The bearing company is GE. Levin's executives would recognize the result: the lead that protects prevention is not the patent on the algorithm but the service relationship that gets the algorithm onto the plant floor.
So Arrow and Teece pull in opposite directions, and the arithmetic says which wins. Arrow raises the incumbent's floor by $8,000 and lets a challenger underbid. Teece hands the incumbent the channel the challenger cannot buy. Write the division of the zone as a bargain, in the form John Nash gave it in 1950: each side's floor is what it earns alone, the zone is the surplus between them, and the supplier takes a share w of the zone. Nash leaves w as a weight. Teece's complementary assets are what set it, and w rises with ownership of the access:
Fee = Fee floor + w x Zone
At full belief the zone is $34,258. A supplier that owns little of the access, w = 0.2, earns $6,852 a year above its floor, 7% of the plant's $100,000 of gross savings; one that owns the channel outright, w = 0.8, earns $27,407, or 27%. At half belief the same three suppliers earn $852, $2,129 and $3,407 above their floors, between 1% and 5% of what the plant saves. The shares are measured against the plant's gross saving, $100,000 a year at full belief and $70,000 at half. In every case the plant keeps the larger part, and in the realistic case, the half-believing plant, the supplier's reward for preventing a $500,000 failure is a few thousand dollars a year.
| Share of downtime saving believed | Supplier's bargaining weight | Fee | Supplier's profit above its floor | Share of the plant's gross saving |
|---|---|---|---|---|
| 100% | 0.2 | $31,490 | $6,852 | 6.9% |
| 100% | 0.5 | $41,767 | $17,129 | 17.1% |
| 100% | 0.8 | $52,045 | $27,407 | 27.4% |
| 50% | 0.2 | $25,490 | $852 | 1.2% |
| 50% | 0.5 | $26,767 | $2,129 | 3.0% |
| 50% | 0.8 | $28,045 | $3,407 | 4.9% |
That is the inquiry's answer to the column. Prevention is worth a great deal, and almost all of it goes to the owner of the machine, because the owner is the only party who can see the saving and the only one who does not have to prove it to anyone. For a plant that uses it well, the saving can become an advantage its rivals find hard to copy, as long as the advantage lies in the organization built around the sensor, which rivals cannot buy; the sensor itself they can. What the supplier can capture depends on the plant's belief first and on the supplier's ownership of the access second. The incumbent's way out, which Teece also foresaw, is to own the scarce asset before the imitators arrive. In a plant, sensors are easy to buy. What is scarce is access: the bearings already installed, the distributor the maintenance chief calls, the service technician who is on site every week. Whoever owns that access decides whose sensor gets bolted on.
I know this problem from the inside. After I sold my first company, I started an innovation consulting firm with Professors David Teece and Daniel Diermeier, to help large companies innovate. My book records the result without flinching: "extensive effort yielding only modest returns" (The Scholars' Treasure, chapter 16, page 202). If this inquiry's model is right, that should not have surprised us: advice that makes a client better is a saving the client keeps, and the adviser is paid for the hours, not the result. The same page carries a line I have had cause to repeat: "the greatest lesson of history is that we never learn the lessons of history." Prevention keeps proving it. The Jost Report has been quoted for sixty years by people who never re-measured it, and the Phoebus cartel's files sat in a Berlin archive for decades. The lessons were there.
I have since counted how often they were there. For the Failure Analysis Institute I scored a hundred modern catastrophes, from Challenger to Flint, in the institute's first research report, written by yours truly, Known in Advance: A structural analysis of one hundred modern catastrophes and the expertise that was never assembled (August 2026). In 99 of them the knowledge needed to prevent the disaster existed before it happened; in 67, the specific mechanism that killed people had been described in writing, by a named person, before the event, and overruled. One of the report's patterns is the one this inquiry models: the saving that came just before a failure was almost always trivially small next to the loss. At Flint, the corrosion control that was not added to the river water would have cost about $100 a day, by the estimate of Marc Edwards, the Virginia Tech engineer whose testing helped expose the crisis. Catastrophic failure is rarely a failure of knowledge. It is a failure of belief, of budget and of assembly, the same three things that decide whether a plant will pay for a saving it cannot see.
Fuchs, the largest independent lubricant maker in the world, was founded in Mannheim on May 30, 1931, by Rudolf Fuchs, who imported Pennsylvania motor oil and, with his wife Irma, filled the cans at the city slaughterhouse. Its case study from a German insulation plant is the one hard field result in this inquiry, and it is worth reading as the supplier wrote it. The plant cures glass wool in ovens at up to 300 degrees Celsius, on conveyor chains that have to be oiled. The competitor's chain oil evaporated in the heat, left varnish on the links and, several times a month, caught fire. Fuchs supplied a synthetic oil built for the temperature. Since the switch, Fuchs reports no oven fires attributable to the chain oil, and 52% less oil consumed, and Fuchs is planning trials at the customer's plants in the Americas.
Read as economics, the case has three lessons. First, Fuchs won a competitor's customer, so the oil it displaced was somebody else's; its own P, the lost parts contribution in the supplier's floor, was zero, and it entered the bargain as the challenger with the low floor. Second, the fires were worth more than the oil. A plant that loses production to a fire several times a month will pay for a fire-free furnace in a way it will never pay for a smaller oil bill, and the saving it could see, the fires, is what carried the sale. Third, the 52% is a problem for every lubricant maker, Fuchs included, the moment the customer is its own, a problem which I have named the Lubricant Maker's Dilemma: the better the oil, the less of it the customer buys. Here is the arithmetic. Call the old price 1, the old volume 1 and the old cost of making the oil 0.7. The new oil costs 1.0 to make, and the customer needs 0.48 of the old volume. At a price of 1.5, the supplier's revenue is 0.48 x 1.5 = 0.72 of what it was and its gross profit is 0.48 x (1.5 - 1.0) / 0.3 = 0.80 of what it was. To hold revenue, the price would have to be 1 / 0.48 = 2.083, a premium of 108%; to hold gross profit, 1 + 0.3 / 0.48 = 1.625. Or the supplier can replace the lost volume with new customers: at a 1.5 price it needs 1 / 0.72 = 1.39 times as many of them, 39% more, to stand still.
The way out of the Lubricant Maker's Dilemma was written down in 1999. Reiskin, White, Kauffman Johnson and Votta described chemical management services, in which the supplier is paid per unit of the customer's output, per door painted or per thousand circuit boards cleaned, so that the chemical becomes the supplier's cost rather than its revenue, and the supplier wants to use less of it as badly as the customer does. Gain-sharing clauses split what is saved. Quaker Houghton, the company formed when Quaker Chemical of Conshohocken, Pennsylvania, organized in 1918, combined with Houghton International in August 2019, is the publicly traded heir to that idea: most of its sales go through its own employees and its Fluidcare chemical-management programs, and its 2024 case study of an automotive grinding plant claims $300,000 a year saved and 75% less downtime, attributing the saving to the chemistry, the sensing and the on-site people together, which is Brynjolfsson and Hitt's old finding about information technology, that the value depends on the organization built around it, restated with cutting fluid. What the filing says about how Quaker Houghton is paid is the subject of one of the tests below, and it is not what Reiskin described.
The dilemma has a longer history than oil. When Michelin's radial tires reached America through Sears in 1966, they lasted far longer than the bias-ply tires Akron made, whose cords ran at an angle across the tread and flexed the rubber until it wore, and Harvard's Donald Sull, in his history of Firestone, states the industry's problem in one sentence: "Radials' longer life would decrease unit demand in the profitable replacement market." Firestone and Goodyear answered with a half step, the belted bias tire, the old design with a belt under the tread; Firestone committed to radials only when the automakers demanded them in 1972, and its rushed Firestone 500 cost it a $150 million recall in 1978. The durable product won, the incumbents lost share, and nobody captured the longer life in a higher price. The counterexample is a lighter: every Zippo since 1932 has carried the guarantee that "it works or we fix it free," and the company has sold more than 600 million of them, to new users and as gifts. Durability can be given away as marketing, lost as volume, or priced, and those are the three outcomes of our zone.
The Lubricant Maker's Dilemma reaches well beyond oil, to wherever the party best placed to prevent a loss is paid when the loss happens. Walter Stahel argued in a 1976 report to the European Commission that a seller paid for a product's performance, and not for the product, would want it to last, and later called the idea the performance economy. Electric utilities earned more by selling more power, so for decades they had little reason to help their customers use less; in 1982 California's regulators cut the link for the large utilities with a mechanism that let them recover their costs whatever they sold, and in 1985 Amory Lovins named what a utility could now sell instead: the negawatt. A doctor paid by the visit is paid when the patient is ill; a health plan paid a fixed sum a member is paid when the member stays well, and that difference is the whole argument for prepaid care. A franchised car dealer earns more than half of its gross profit from service and parts, by the dealership accountants Presidio-NCM's count for 2025, and a battery-electric car, by Consumer Reports' estimate, costs about half as much to maintain over 200,000 miles, $4,600 against $9,200. In Rome, Plutarch says, Marcus Crassus bought houses that were on fire, and the houses beside them, at a trifling price, and rebuilt them with his own slaves. Two thousand years later, after the Los Angeles fires of January 2025, California had to outlaw unsolicited lowball offers for burned homes. The answer to that kind of incentive came from insurers, the one party that earns most when nothing fails. In 1835 Zachariah Allen, a Rhode Island mill owner, founded a mutual fire insurer that covered only mills that kept its rules of prevention, the ancestor of today's FM, and in 1866 the Hartford Steam Boiler Inspection and Insurance Company began selling inspection and insurance together, the insurance, in its own history's words, "an incentive to inspect."
Does anyone build products to fail on purpose? Economists have argued about it for more than fifty years. Peter Swan showed in 1970 that a monopolist, under simple assumptions, has no reason to make a product wear out sooner than it should. Jeremy Bulow answered in 1986 that a monopolist that cannot commit to its future prices does, because a durable good competes with its own next sale: monopolists, he wrote, "desire uneconomically short useful lives for their goods." Michael Waldman showed in 1993 that the same end can be reached by new products that will not work with the old. The phrase goes back to a 1932 pamphlet by Bernard London of New York, "Ending the Depression Through Planned Obsolescence," and the industrial designer Brooks Stevens made it popular in 1954. The one proven conspiracy is Phoebus, which fined its members for bulbs that burned longer, or shorter, than 1,000 hours. The famous modern suspicion ended without any finding of intent. In December 2017 Apple acknowledged that an update released that January slowed some older iPhones with worn batteries, said it had done so to prevent sudden shutdowns, apologized, and cut the price of a replacement battery from $79 to $29. It later settled a United States class action for up to $500 million and the claims of 33 states and the District of Columbia for $113 million, denying wrongdoing in both, and agreed to pay France €25 million for failing to tell customers, after the charge of planned obsolescence was dropped. Italy's competition authority went furthest, fining Apple €10 million in 2018 for updates that, it found, reduced performance and hastened replacement. France has made planned obsolescence itself a crime since 2015, punishable by two years in prison and a fine of €300,000 or up to 5% of annual sales, and the European Union adopted a right-to-repair directive in 2024.
This reminds me of sixty people at four Harvard teaching hospitals in 1978. Ward Casscells and two colleagues asked twenty students, twenty residents and twenty attending physicians a question: a disease affects one person in a thousand, the test for it has a 5% false-positive rate, and a patient tests positive; what is the chance the patient has the disease? The right answer is about 2%. Eleven of the sixty got it. Twenty-seven said 95%. Daniel Kahneman and Amos Tversky had named the error five years earlier: people neglect the base rate when a vivid signal is in front of them, and a positive test is vivid.
A failure alert on a factory floor is the same problem with a bigger bill. Picture 1,000 machines, of which one in a thousand is about to fail in any month; a monitor that catches 90% of the real failures; and a specificity of 99%, meaning it stays quiet on 99% of the healthy machines. Each month it raises 0.9 true alerts and 9.99 false ones, and 8.3% of its alerts are real. Raise the specificity to 99.9% and the false alerts fall to one a month, and 47.4% of the alerts are real. Then price the alarms. If a false alert costs a $200 inspection, the monitor at 99% wastes $1,998 a month, which nobody notices. If a false alert stops a line for a $30,000 shutdown, it wastes $299,700 a month, which nobody survives, and even at 99.9% the unnecessary shutdowns cost $29,970. Set a true catch at $50,000 of avoided loss and the break-even specificity for a program whose false alarms cost shutdowns is 99.85%; for one whose false alarms cost inspections, 77%. The vendor's brochure counts alerts. The plant's controller counts these.
Two things follow. The false-alarm budget B in the plant's model, $5,000 a year in our illustration, is where the specificity of the monitor meets the income statement, and a plant that sets it should know whether its alarms end in inspections or in shutdowns. And a predictive-maintenance program should be judged by the losses it avoids after verification, the technicians' hours and the interventions it causes, never by the number of alerts it raises. Medicine learned this the hard way: in hospitals, Sue Sendelbach and Marjorie Funk reported in 2013, 72% to 99% of clinical alarms are false, and the staff learn to ignore them. A factory that learns to ignore its monitor has paid for a sensor and bought nothing.
SKF said all of this itself, in 2019, in better prose than most analysts manage. Its annual report for that year, under the heading "How fee-based business models work," put the conflict in one sentence: "In the traditional transaction-based model, suppliers' profits depend on numbers of parts sold and not on improvements in machine performance. This represents a fundamental conflict of interest, as longer component life means fewer sales for the supplier." Its answer was a contract it called Rotating Equipment Performance, in which "the customer pays a fixed fee for service and/or equipment," possibly with "an additional component linked to agreed targets," and "an all-inclusive fee may cover supply of bearings, seals, lubrication and condition monitoring." A panel set "Selling bearings only" against "Selling reliable rotation." A graphic promised that the three million bearings SKF then monitored would become 40 million by 2025 and 150 million by 2030. And a mining customer, with 8,000 sensors on 2,400 critical assets, was saving "almost eight million EUR per year," a figure SKF said it had documented. That is a company that had read Arrow, understood its own floor, and built the contract that was meant to climb over it.
The 2025 annual report, published in March 2026, is a different document. Rotating Equipment Performance does not appear in it. Neither do fixed-fee or performance-based customer contracts; the only performance-based arrangements in the report are the executives' bonuses. Note 1 on revenue recognition says that revenue from service and maintenance contracts is recognized either at a point in time, the way a part is booked when it ships, or over the life of the contract, the way a fee for a service is booked as the months pass, and then: "Revenues recognized over time is not deemed material." The 40 million connected bearings are not mentioned, and no count of connected bearings is given. Missing disclosure is not evidence of failure, and I do not read it as such. The note allows service obligations to be recognized at a point in time as well, so it is a caution about a large continuous-service business, not a count of every recurring or outcome-priced contract: a fee for uptime that had grown to matter would most likely be recognized over time, and SKF's own accounting note says there is not enough of that to deem material. Missing disclosure leaves the capture unproven, not disproven.
What the report does disclose is the EU taxonomy table, which is where the condition-monitoring business shows its size. The taxonomy is the European Union's rulebook of activities that count as environmentally sustainable; a publicly traded company reports the share of its revenue that is eligible, meaning covered by a listed activity, and the share that is aligned, meaning it meets the activity's criteria. The activity "Provision of IT/OT data-driven solutions," the regulation's name for monitoring, accounts for 2% of SKF's revenue; 180 million kronor of it is reported as aligned, and that 180 million is 8% of the activity's eligible amount. Working back, 180 / 0.08 = 2,250 million kronor of eligible monitoring revenue, which is 2.46% of SKF's 91,583 million of sales; allowing for the 8% being rounded from anywhere between 7.5% and 8.5%, the back-out alone gives 2.31% to 2.62%, and the printed 2%, itself rounded from somewhere between 1.5% and 2.5%, cuts the top off: the activity is between 2.31% and 2.5% of revenue, on the usual rounding, which the report does not state. The 180 million is the aligned part, not the services part: the report says that "revenue streams that include both services and hardware have been excluded for cases where a separation has not been possible," which bears on what SKF could show as aligned, not on the eligible figure. So the monitoring category is a business of between 2.1 and 2.3 billion kronor inside a company of ninety-two, and the part SKF can show as aligned is two tenths of a percent of sales.
The materiality arithmetic my AI assistants and I set up, and the Desk confirmed, follows. Take the category at 1.5% to 2.5% of sales, grow it 20% a year and let 30% of the growth fall to operating profit, both labeled assumptions. In one year that adds 82 to 137 million kronor of operating profit, or 0.7% to 1.2% of 2025's adjusted operating profit of 11,673 million. It is a materiality test, not a forecast. Compound the same 20% for five years and the fifth year's profit runs 613 to 1,022 million above 2025's, 5.3% to 8.8% of that profit: a yearly figure in the fifth year, not a sum over the five, and before any parts sales the monitoring cannibalizes. That is the horizon at which the business would begin to show in the group's margin. Against the aftermarket it is small on any horizon: the 2025 report says the aftermarket represents "more than half" of the Industrial business's 65,614 million of revenues, at least 32,807 million, so a monitoring category of about 2,250 million is at most 7% of it, and by the mid-2026 report "Aftermarket and service" is 39% of the whole group's sales. In the other direction of prevention, SKF remanufactured or refurbished "over 6,000 tonnes of bearings" in 2025, in a year when it bought 414,893 metric tons of metal; the two figures give the scale and not a share, since the metal SKF buys is not the stock of bearings in service that could come back for remanufacture. Its RecondOil oil-recovery service "is not yet reported separately."
| SKF, 2025 | Figure | Where |
|---|---|---|
| Net sales | 91,583 MSEK | Annual Report 2025, p. 152 |
| Adjusted operating profit | 11,673 MSEK | p. 152 |
| Industrial net sales | 65,614 MSEK | p. 109 |
| Aftermarket | more than half of Industrial revenues | p. 12 |
| Monitoring, taxonomy-eligible revenue (back-out) | 2,250 MSEK, 2.46% of sales; 2.31% to 2.5% allowing for the rounding of both printed figures | p. 43; 180 / 0.08 |
| Monitoring, aligned (services only) | 180 MSEK, 0.2% of sales | p. 43 |
| Revenue recognized over time | "not deemed material" | p. 107 |
| Bearings remanufactured or refurbished | over 6,000 metric tons (414,893 metric tons of metal bought in the year) | p. 67 |
So the company that named the conflict in 2019 and designed the contract to resolve it reports, six years on, a monitoring business of between two and two and a half percent of sales, a contract business too small to deem material, and an aftermarket of more than half. Teece would not be surprised, and neither would Levin's executives. The saving is being captured, where it is captured at all, through the assets SKF already owned: the bearings, the hardware, the application engineers and the distributors, more than 7,000 of them for the Industrial business alone. On December 1, 2026, SKF plans to hand its Automotive business, renamed SKF Vertevo, to its shareholders and list it separately, five SKF shares to one Vertevo share, and the Industrial company that remains has told investors to expect an adjusted operating margin above 17% in the medium term and above 19% in the long term. Those targets rest on the aftermarket. The prevention business, in the model's terms, is the incumbent's way of keeping its access to the customer.
None of this is a recommendation; it is where the arithmetic points, with the evidence behind each hypothesis and the fact that would prove it wrong. The five companies are priced, on October 6, 2026, as follows, in the roughest useful way: the market value of all the shares over the last twelve months' owner earnings (net income plus depreciation and amortization, less capital spending), and the growth in those earnings, forever, that each price pays for at a 9% cost of equity.
| Company | Price, October 6, 2026 | Market value of the shares (millions) | Owner earnings, twelve months (millions) | Multiple | Growth forever the price assumes at 9% |
|---|---|---|---|---|---|
| SKF | SEK 280.20 (B) | SEK 127,540 | SEK 5,500 | 23.2 | 4.5% |
| Fuchs | €43.70 (pref.) | €5,342 | €333 | 16.0 | 2.6% |
| Schaeffler | €6.36 | €6,009 | €248 | 24.2 | 4.7% |
| Timken | $119.54 | $8,284 | $349.8 | 23.7 | 4.6% |
| Quaker Houghton | $161.08 | $2,772 | $110.7 | 25.0 | 4.8% |
The growth figure comes from one line, the price as a growing perpetuity: P = OE x (1 + g) / (r - g), with r the cost of equity and g the growth rate, solved for g. For SKF, (127,540 x 0.09 - 5,500) / (127,540 + 5,500) = 4.5%. At 8% the same price needs 3.5%; at 10%, 5.4%; the model's output has the same two cases for all five. Owner earnings are a rough instrument, and Schaeffler's and Quaker Houghton's are depressed by last year's write-downs, which inflates their multiples; the point is the range. Every one of these prices pays for a mature industrial business growing a little faster than the economy. None pays for a prevention windfall. That is the one price that carries the theme: the market has already decided that the supplier keeps little of the saving, and the question is only whether it is right.
SKF: the price pays for the aftermarket, and prevention defends it. At 23 times owner earnings and 4.5% growth forever, the price values the franchise the 2025 report describes, a recurring aftermarket of more than half of Industrial revenues, with the monitoring business as a defense of it rather than a line of its own. Evidence: the 2019 report's own description of the conflict and the contract; the 2025 report's silence on that contract; Note 1; the taxonomy row. Falsified if SKF's annual report for 2026 deems revenue recognized over time material, or reports the monitoring activity's eligible revenue at 2,750 million kronor or more, which is 3% of 2025's group sales and the top of the 2025 band grown 20%. The test is set in kronor because the Automotive separation on December 1, 2026 shrinks the denominator: the 2025 figure is already 2.8% of Industrial sales with no new monitoring sold. Either would mean the contract route is capturing the saving after all.
Fuchs: the challenger with the low floor. Fuchs enters most of its prevention bargains as the challenger in our model, winning a competitor's customer by making it use less, so its own lost contribution is zero and its floor is the lowest in the room. Its gross margin rose to 34.9% in 2025 "as a result of reduced material usage," and its price, at 16 times owner earnings and 2.6% growth, assumes the least of the five. The hypothesis is that a lubricant maker whose products cut consumption holds its margin through price and mix when volume growth fades: the first half of 2026 grew 11% on volume, part of it pre-buying during the conflict in the Middle East and competitors' supply failures, which will not repeat. Falsified if Fuchs's gross margin for 2026 falls below 34.9%.
Quaker Houghton: the servicizing model's publicly traded heir, paid the old way. Quaker Houghton's Fluidcare programs are the closest thing on an exchange to the chemical management services Reiskin and his colleagues described, and its 2024 case study claims $300,000 a year saved for an automotive grinding plant with 75% less downtime. But its 10-K says it is paid as a principal at negotiated prices or as an agent for an administrative fee, with service revenue measured "by labor costs and time incurred," and nowhere mentions a fee tied to the customer's output or savings; the only sentence about consumption is a risk factor, that customers may "reduce consumption of the specialty chemicals that we produce." Grubić and Jennions found that contracts paying for the customer's economic result barely exist, and this filing agrees. Falsified if Quaker Houghton's 10-K for 2026 reports revenue tied to customer output or documented savings.
Schaeffler: the replacement division out-earns the prevention division. Schaeffler's automotive aftermarket, Vehicle Lifetime Solutions, earned a 14.8% margin before special items in 2025 on €3,038 million of revenue, "a reliable earnings driver with stable demand"; its Bearings & Industrial Solutions division, which sells the bearings and the condition-monitoring systems that prevent their replacement, earned 7.5% on €6,368 million. That is Arrow's effect written across two segment lines: the business that sells the replacement earns twice the margin of the business that prevents it. Falsified if the gap between the two margins closes to less than five points in 2026.
Timken is the quiet case. Its 10-K for 2025 never uses the words "condition monitoring," states its strategy as capturing "subsequent equipment replacement cycles" through independent distributors, and sells 40% of its output to distribution and end users. It is priced like SKF. We have no test for it that the model would stand behind, and we say so.
The laboratory results that give tribology its current glamour should be kept in a separate room from the investment case, and the column that started this inquiry did not always keep them there. In March 2026 Wan Wang and nine co-authors reported, in a paper posted to arXiv, a friction coefficient of about 0.008 sustained over 100,000 sliding cycles at a contact pressure of 12.7 gigapascals in air at 40% humidity, using laser-patterned arrays of diamond-like carbon, a hard amorphous carbon coating, on molybdenum disulfide, reinforced with a MXene, one of a family of two-dimensional ceramic sheets: superlubricity, the near-vanishing of friction, "under engineering-relevant conditions," in their words. It is a real advance. It is also, at the 6.37 hertz and 5 millimeter stroke of the test rig my AI assistants found in the paper's methods, about 4.4 hours of sliding and one kilometer of distance. A bearing in a paper mill runs for years and travels thousands of kilometers under oil at temperatures the test never saw. Fuchs took part in a German research consortium on superlubricity, CHEPHREN, with Fraunhofer institutes, BMW and others, from September 2021 to February 2025; its annual report does not mention it, and it has no revenue.
Joel Mokyr, one of my favorite teachers at Northwestern, draws the line this inquiry needs. There is propositional knowledge, what is true about the world and why, and prescriptive knowledge, how to make something work, and the two advance on different clocks. The superlubricity papers add to the first. A furnace that has not caught fire in two years because the oil stopped evaporating is the second. Leonardo had the first in 1493 and, as Hutchings found, it changed nothing for two hundred years. The bargain in this inquiry is struck entirely in the second kind of knowledge, between a plant that can see a saving and a supplier that can deliver it, and the papers will join it only when they can be sold by the liter or the fixed fee.
Vista makes no buy or sell calls. The analysis still points in clear directions.
In any story about prevention, ask first who owns access to the customer. The chemistry, the sensor and the algorithm are the innovation; the installed base, the distributors and the service relationship are the complementary assets, and the second set earns the money. A start-up that prevents failures in machines it did not sell is EMI with its scanner: EMI, the British record company, invented the CT scanner, the first clinical scan was made in 1971, and within a decade General Electric, with the hospital sales and service network EMI lacked, had the business, bought from Thorn after Thorn took over EMI in 1979. The exit most prevention start-ups will find is a sale to the company that owns the access, and the history of condition monitoring is a list of such sales. Emerson bought CSI, a vibration-analysis company in Knoxville, in 1997 for about $160 million; General Electric bought Bently Nevada in 2002; Rockwell Automation bought Entek IRD in 2000 and Fiix in 2020; SKF bought Baker Instrument in 2007 and Presenso, an Israeli predictive-maintenance start-up, in 2019; Schaeffler bought autinity systems in 2017; Fortive's Fluke bought Prüftechnik in 2019 and Azima DLI in 2023; the Japanese bearing maker NSK bought Brüel & Kjær Vibro in 2021; and Siemens bought Senseye in 2022. The independents to watch are companies such as Augury, valued at more than $1 billion in its 2025 round and in partnership with Rockwell since July 2026, and Tractian, which raised $120 million in December 2024. Naming them is no recommendation; each is the kind of company whose natural buyer owns the access, and the incumbents' acquisitions are the signpost to watch.
Value an incumbent's prevention business as a defense of its aftermarket. The $8,000 in our model, the parts contribution the incumbent gives up, is what it pays to keep the customer; on SKF's scale the monitoring category will not move the margin for years on any growth rate the filings support, and what it can move, retention, the filings never show.
Watch the customer's belief, because it is the variable a better sensor does not move by itself: the technology moves the savings, the costs and the alternatives, and each of those moves the zone, but the share of a saving a plant will pay for is set by what it can verify. The suppliers that will capture more of the saving are those who can turn a counterfactual into an invoice: a documented €8 million at a mine, a fire-free furnace, a performance guarantee.
The theme's price is the five prices in the table, and they are modest. If contract capture appears, the upside is in the multiple, because nobody has paid for it; if it does not, the status quo is already priced. That is the asymmetry, and it is not large. And if the owner of the machine keeps the saving, the beneficiaries of better tribology are the machine-heavy operators, the miners, paper and cement makers, utilities and railroads, whose maintenance and energy bills shrink without their having to prove anything to anyone. Where that shows first is a question for the next inquiry.
The signposts, in order: SKF's third-quarter report on October 21, 2026; Fuchs's quarterly statement on October 30; Schaeffler's nine-month results on November 3; Timken's and Quaker Houghton's third-quarter reports in late October or early November; SKF Vertevo's capital markets day on November 17 and its first trading day on December 1; the annual reports and 10-Ks of February and March 2027, which carry the tests below.
The model is also a business plan, and in the age of AI a business plan can be started in the time it takes to write one down. Read backward, the arithmetic says where a prevention company makes its money.
Build the sensor, the software or the oil that prevents failures on machines you did not sell, so that Arrow's penalty, the parts margin you would give up, is zero and your floor is the challenger's. Sell it first to the biggest users, the plants whose failures are so expensive that even a doubting manager will pay. Then sell the company to the incumbent that owns access to those plants, as the start-ups named above did. Rinse and repeat.
The better version adds one feature the model values above any improvement in the sensor: a ledger that shows the customer what it saved. Every verified catch, the bearing changed on a Tuesday morning instead of failing on a Saturday night, entered with the downtime it would have cost and signed by the plant's own maintenance chief. That ledger moves the belief c, and c is the variable that opens or closes the whole bargain: at full belief the zone is $34,258 a year, at half belief $4,258, and below 43% there is nothing to share. Even an occasional true alarm, shown this way, pays for many quiet months. A company that can turn a counterfactual into an invoice can charge for prevention. A company that cannot is selling alerts.
Three more businesses fall out of the same arithmetic. An independent verifier of avoided failures, which certifies what a monitoring program prevented, so that the plant, its insurer and its supplier can agree on c. An insurer for the sensor age, in the tradition of Hartford Steam Boiler, which earns most when nothing fails and so can pay for prevention itself. And an uptime contractor, paid by the hour a machine runs, which owns the cost of failure the way an engine maker does under Power-by-the-Hour, and needs the balance sheet to carry it.
Entrepreneurial readers should move quickly. I am already busy starting prevention businesses of my own.
Likely, we can get some of it from primary research, which is what Vista is all about. The people who set the fees, signed the renewals and sold against the incumbents already know what the filings will take years to show.
As a matter of general past practice, what share of a documented saving did customers accept when a contract was renewed, and did the parts margin the contract displaced enter the fee?
The answer that would change the view"Customers paid for the whole documented saving, the fee was set from it, and the parts margin never came up."
In general, how much of the vendor's avoided-downtime estimate did your finance department accept, and did the maintenance and energy savings sit in the same budget?
The answer that would change the view"Finance took the vendor's number without a haircut, and we renewed on it."
As a matter of general past practice, how often did the incumbent's distributor or service relationship decide the sale, and where did the company's exit lead?
The answer that would change the view"Plants bought from us over the incumbent's head, and service access never mattered."
In general, what happened to pricing and to your commission when a product cut a customer's consumption by half, and was the price per liter raised to hold gross profit?
The answer that would change the view"We priced the saving in, and margin per customer rose when volume fell."
In general, how were the fees actually set, by product price, by administrative fee, by labor and time or by the customer's output, and how often did a gain-sharing clause exist and pay?
The answer that would change the view"Most contracts shared measured savings, and the clause paid every year."
For a client engagement, Vista combines this report with that primary research.
We will score this inquiry on April 30, 2027, after the five companies' annual reports and 10-Ks for 2026 are published, and publish the result on the scorecard whether it flatters us or not. Each test sides with the reading here, that the customer keeps most of the saving and the supplier captures it, if at all, through the assets it already owned:
All four passes and the reading here gains ground; three is too early to say; two or fewer, and the saving is moving to the supplier faster than the model allows.
Sven Wingquist solved a maintenance problem and built the company that sells the replacements. The Phoebus cartel fined its members for building a bulb that lasted. Firestone sold a half-step tire for six years rather than one that lasted longer, and lost its lead doing it. The rub in a bearing is friction, and engineers have been reducing it since a page in red chalk in 1493. The rub in the business is that the firm best placed to prevent a failure is the firm that is paid when it happens, and no sensor fixes that. A contract might, once the customer believes.
The bet. Anyone holding the shares of a bearing or lubricant maker at these prices is betting on a mature aftermarket growing 2.6% to 4.8% a year forever, which is what the prices pay for. Anyone betting on a prevention windfall, a supplier that captures the saving its products create, is betting that two things move at once: the customer's belief in a saving it cannot see, and the supplier's ownership of the access that lets it charge for the saving.
The payoff. If contract capture appears, over-time revenue deemed material at SKF, a fluid manager paid on savings, the upside is in the multiple, because none of the five prices includes it. If it does not, the prices already assume the status quo, and the holder earns the aftermarket's growth and dividends, SEK 7.75 a share at SKF this year, and nothing more.
Our read. The evidence sides with the customer keeping the saving. SKF named the conflict in 2019, built the contract to resolve it, and six years later the contract is too small to deem material while the aftermarket is more than half of Industrial revenues; Fuchs wins share by making customers use less and holds its margin by price; Quaker Houghton is paid for product and time; Schaeffler's replacement business earns twice the margin of its prevention business. Arrow explains why the incumbent's floor is high, and Teece explains why the incumbent wins anyway.
What settles it, and when. The 2026 annual reports and 10-Ks, between February and March 2027, carry the four tests; we score them on April 30, 2027. The first conversations we would have are with the people who priced fee-based contracts against a parts business, and with the plant managers who renewed them, because between them they know the one number the filings will never print: how much of the saving a customer believes.
Mostly the customer. In the inquiry's worked example a plant that believes the whole saving can pay a vendor $58,896 a year, the vendor needs $24,638 to break even, and even a supplier that owns the whole channel takes about 27% of the plant's gross saving above its floor; at half belief the supplier's profit is a few thousand dollars a year. The saving is a counterfactual the plant alone can see, so the plant keeps most of it.
Often because they are right not to. Below a critical level of belief in the avoided downtime, 42.9% in the example, no fee satisfies both the plant and the supplier, and walking away is rational; the cure is verification, not a better sensor. Above that level, an unsigned contract points to split budgets, a sales force paid by the liter, or, as Akkermans and his colleagues found, machines maintained so well that a predictor has no failures to learn from.
Kenneth Arrow showed in 1962 that an incumbent gains less from an innovation than a newcomer because the innovation replaces profit it already earns. In a prevention contract that is a line item: the parts seller's fee floor is higher than a challenger's by exactly the parts contribution it loses, $8,000 a year in the example, so the challenger can always underbid. SKF's own 2019 report called it "a fundamental conflict of interest, as longer component life means fewer sales for the supplier."
Teece's 1986 paper says the profit from an innovation goes to whoever owns the complementary assets it needs to reach customers. For prevention those are the installed base of bearings, the distributors (SKF's Industrial business has more than 7,000) and the service access, which the incumbent owns. The sensor start-up is EMI with its CT scanner; the bearing maker is GE. The money goes to whoever owns access to the customer, not to whoever owns the chemistry.
SKF's 2025 EU taxonomy table reports the monitoring activity at 2% of revenue, with 180 million kronor aligned, which is 8% of the activity's eligible amount; working back, about 2,250 million kronor, or 2.46% of sales (2.31% to 2.5% allowing for the rounding of both printed figures). Revenue recognized over time is "not deemed material," and the aftermarket is more than half of Industrial revenues.
At 48% of the old volume, Fuchs's chain-oil case, a price 1.5 times the old one leaves revenue at 72% and gross profit at 80% of before; holding gross profit needs a price of 1.625 and holding revenue 2.083, a 108% premium. Fuchs won that customer from a competitor, so it lost nothing of its own, and its gross margin rose to 34.9% in 2025.
Because failures are rare. With 1,000 machines, one in a thousand about to fail each month, 90% sensitivity and 99% specificity, a monitor raises 0.9 true alerts and 9.99 false ones a month, so 8.3% of alerts are real; at 99.9% specificity, 47.4%. If a false alert costs a $30,000 shutdown the program pays only above about 99.85% specificity. Judge a program by the losses it avoids after verification, never by the number of alerts.
Every figure was computed in code (model.py, standard library only) and each public input was checked against its source. The customer's model: net benefit = E + M + c p L - F - U - B with E = $30,000, M = $10,000, p = 0.12, L = $500,000, F = $30,000, U = $15,000, B = $5,000; NPV = net benefit x af - K with K = $80,000 and af = (1 - 1.1^-5) / 0.10 = 3.790787; the affordable fee is the F at which NPV is zero. The supplier's model: NPV = (F - D - P - R) x af - O with D = $12,000, P = $8,000 (zero for a challenger), R = $2,000, O = $10,000; the fee floor is the F at which it is zero, $24,638 and $16,638. The zone is the affordable fee less the floor; the critical belief solves zone = 0; the split is Nash's bargaining form on the net surplus: a share w of the zone above the floor, with w of 0.2, 0.5 and 0.8 as illustrations of a supplier's bargaining weight, which Teece's complementary assets raise. The lubricant arithmetic indexes the old price, volume and cost to 1, 1 and 0.7 and the new volume and cost to 0.48 and 1.0. The alert arithmetic uses 1,000 assets, a 0.1% monthly prevalence, 90% sensitivity and the stated specificities; precision is true alerts over all alerts, and the break-even specificity sets false-alert cost equal to the value of true catches. SKF's materiality applies a 20% growth rate and a 30% incremental margin to a category of 1.5%, 2.0% and 2.5% of 2025 sales, for one year and for the fifth year of compounding (a yearly figure), against adjusted operating profit of 11,673 MSEK; the taxonomy back-out divides 180 MSEK by 8% (and by 7.5% and 8.5%) and keeps the result inside the printed 2%'s own rounding band, below 2.5%. What the price assumes solves P = OE (1 + g) / (r - g) for g with r = 9% (8% and 10% shown in the model's output), using October 6, 2026 closes, shares outstanding from the latest filings and owner earnings for the twelve months to June 30, 2026 as net income plus depreciation and amortization less capital expenditure, each from the companies' reports. Currency conversions use the European Central Bank's reference rates for October 6, 2026. The economy-wide figures divide the Jost Report's £515 million by British output of £36,906 million in 1965 (the Office for National Statistics' figure at market prices) and take Holmberg and Erdemir's shares as published. Every historical date, quotation and founding was checked against the primary document or two independent sources, and the superlubricity figures against the arXiv abstract; the 6.37 hertz and 5 millimeter stroke come from the paper's methods as reported in the draft and are not in the abstract.
| Input | Value | Source |
|---|---|---|
| Customer model (illustration): energy and maintenance savings; failure chance and loss; up-front cost; fee, upkeep, false-alarm budget; rate, years | $30,000 and $10,000 a year; 12% and $500,000; $80,000; $30,000, $15,000, $5,000; 10%, five years | Russ W. Rosenzweig's draft, 7 October 2026, assumptions not forecasts |
| Supplier model (illustration): delivery, lost parts contribution, reinvestment, onboarding | $12,000, $8,000, $2,000 a year; $10,000 once | Same |
| Fuchs chain-oil case: oil consumption, furnace fires | 52% lower; several a month before, none attributable since | Fuchs case study page (2026) |
| Alert model (illustration): assets, monthly prevalence, sensitivity, specificities, costs | 1,000; 0.1%; 90%; 99% and 99.9%; $200 or $30,000 per false alert, $50,000 per catch | Russ W. Rosenzweig's draft; the catch value is ours |
| SKF net sales, adjusted operating profit, Industrial net sales, 2025 | 91,583; 11,673; 65,614 MSEK | SKF Annual Report 2025, pp. 109, 152 |
| SKF aftermarket; distributors; application engineers | more than half of Industrial revenues; about 17,000 (more than 7,000 Industrial); around 600 | SKF Annual Report 2025, pp. 8, 9, 12, 34 |
| SKF taxonomy row CE 4.1: eligible share, aligned turnover, aligned in eligible | 2%; 180 MSEK; 8% | SKF Annual Report 2025, pp. 42, 43 |
| SKF revenue recognized over time | not deemed material | SKF Annual Report 2025, Note 1, p. 107 |
| SKF bearings remanufactured or refurbished; metal bought | over 6,000 tonnes; 414,893 tonnes | SKF Annual Report 2025, p. 67 |
| SKF 2019: the conflict, the REP contract, connected bearings, the mining customer | p. 27; 3 million today, 40 million by 2025, 150 million by 2030; almost €8 million a year | SKF Annual Report 2019, pp. 26, 27 |
| SKF Vertevo distribution and listing | record date November 27, 2026; first trading day December 1, 2026; five SKF shares per Vertevo share | SKF press release, October 1, 2026 |
| Fuchs 2025: sales, EBIT, gross margin, employees; first half 2026 growth | EUR 3,563 million; 435 million; 34.9%; 6,879; sales +11%, volume-driven | Fuchs Annual Report 2025, pp. 17, 56, 59; Half-year report 2026, pp. 4 to 7 |
| Schaeffler 2025: Vehicle Lifetime Solutions and Bearings & Industrial Solutions revenue and margin before special items | EUR 3,038 million, 14.8%; EUR 6,368 million, 7.5% | Schaeffler Annual Report 2025, segment note |
| Timken 2025: net sales; sales to distribution and end users; over-time revenue | $4,581.8 million; 40%; about 9% | Timken Form 10-K for 2025 |
| Quaker Houghton 2025: net sales; Fluidcare compensation | $1,888.6 million; principal at negotiated prices or agent for an administrative fee; service revenue by labor costs and time | Quaker Houghton Form 10-K for 2025, Note 5 |
| Closing prices, October 6, 2026; shares outstanding; owner earnings, twelve months to June 30, 2026 | SKF SEK 280.20 (455.4 million shares, SEK 5,500 million); Fuchs EUR 43.70 and 37.85 (131 million, EUR 333 million); Schaeffler EUR 6.36 (944.9 million, EUR 248 million); Timken $119.54 (69.3 million, $349.8 million); Quaker Houghton $161.08 (17.2 million, $110.7 million) | Yahoo Finance closing prints; company interim reports and Forms 10-Q (facts-prices.md) |
| Exchange rates, October 6, 2026 | USD 1.1269 per EUR; SEK 11.2425 per EUR | European Central Bank reference rates |
| Jost Report (1966): annual savings, maintenance and replacement share | about £515 million; £230 million | Lubrication (Tribology): Education and Research, HMSO, March 9, 1966, as quoted by Jost (2006) |
| World energy lost to friction and wear; long-term saving | about 23% (119 EJ): 20% friction, 3% wear; 40% over fifteen years, 8.7% of total energy | Holmberg and Erdemir, Friction, 2017 |
| Superlubricity result | friction about 0.008 over 100,000 cycles at 12.7 GPa and 40% humidity | Wang et al., arXiv 2603.15089, March 16, 2026 |
The model (model.py), its output and the facts files behind each input are kept with the inquiry's working files; the arithmetic in the text was checked against them.
This inquiry is research, not investment advice. It states what prices and models assume under labeled assumptions; nothing here is a recommendation to buy, sell or hold any security, and the decision belongs to the reader. As of October 4, 2026, Russ W. Rosenzweig, Vista's founder, owns no shares of SKF, Fuchs, Schaeffler, Timken or Quaker Houghton, and no shares of Rolls-Royce or Michelin, which appear in the history. He owns shares of Apple, which appears in the history of planned obsolescence, and of GE Aerospace, GE Vernova and GE HealthCare, the successors of the General Electric named in the history, and holds options on Apple and GE Aerospace; he owns no shares of Emerson, Rockwell Automation, Siemens, Fortive or NSK. He is also a director of the Failure Analysis Institute, which provides experts on why things fail, and he is starting businesses in failure prevention. Holdings through mutual funds and exchange-traded funds are not counted. One detail was corrected on October 7, 2026: the column's headline is "Tribology makes the economic world go round"; the first version added "How" to it. How this inquiry was made: written by Russ W. Rosenzweig with Vista's AI research desk. It began with Anjana Ahuja's column in the Financial Times; Russ W. Rosenzweig's question, corrections and models followed, drafted with the help of another AI assistant, and the desk challenged them before this inquiry was written; the models were rebuilt and every figure computed in code, each public source was checked against its primary document, and the draft was fact-checked against those documents again before he read it. I read it and listened to it before it was published. Why I write with AI.