On 22 October 1938, in an apartment in Astoria, Queens, Chester Carlson produced the first xerographic image in history. It read "10.-22.-38 ASTORIA". The technology for copying onto plain paper was solved that day.
It found no buyer. More than twenty companies turned down funding it between 1939 and 1944; IBM among them. Carlson ended up handing management of his patents to the Battelle Memorial Institute in 1944-45, and in December 1946 a small photographic paper company called Haloid bought a non-exclusive licence for $10,000. That $10,000 was ten percent of everything Haloid had earned in 1945.
Thirteen years later the machine existed and still would not sell. In 1958 IBM commissioned the consultancy Arthur D. Little to assess the device Haloid was preparing. The report was negative and IBM passed, convinced the market topped out at around 5,000 machines.
On 16 September 1959 Haloid unveiled the 914 at the Sherry-Netherland hotel in New York, live on television. It weighed about 650 pounds — 294 kilos — made seven copies a minute and could reach a hundred thousand copies a month. It cost $29,500. At that price, almost nobody bought it.
So Haloid stopped selling it. It leased it for $95 a month, with the first 2,000 copies included and four cents for every copy beyond that. The average office copier of those years made 15 to 20 copies a day. The 914 ended up averaging 2,000 a day: customers burned through a whole month's allowance in a single working day.
The rest are well-known figures that almost nobody connects to the one above. In 1955 the world made about 20 million copies a year, nearly all of them with carbon paper. In 1964, five years after the 914, it made 9.5 billion. The commercial copying business went from $40 million to $400 million between 1952 and 1962. In 1965 the 914 alone generated $243 million, two thirds of the company's revenue. Total revenue went from $32 million in 1959 to more than a billion by 1970.
The first comfortable reading is the inventor who never gave up: twenty-one years, more than twenty closed doors, and in the end he was right. It is a lovely story and it explains very little. Carlson did not get in by insisting: he handed his patents to a research institute and took a royalty — around one sixteenth of a cent per copy between 1956 and 1965, with which he ended up donating more than $150 million. What changed between 1938 and 1959 was nobody's determination. The technology was the same. The determination had been there for twenty years.
The second comfortable reading is the opposite one, and today it is the more repeated: the market study got it wrong, IBM made fools of themselves, consultants understand nothing. It is also weak, and in a more interesting way. Arthur D. Little did not mismeasure. Five thousand machines was a reasonable read on a $29,500 device doing a job offices already did with carbon paper at fifteen copies a day. Under those conditions, the report was probably right.
What both readings share is that they locate the explanation in people: a stubborn man, some short-sighted analysts. And the difference between 1938 and 1959 is not in any person. It is in one line of the contract.
This is not about perseverance, and it is not about how little market studies are worth. It is about where the invention lives when the product is already finished and still will not move.
1. The obstacle was not the price: it was the estimate the price demanded. To sign for $29,500, the buyer had to answer how many copies they were going to make. Their honest answer was fifteen or twenty a day, because that is what they did with the carbon paper they had. Nobody signs a large cheque against a forecast they know they cannot make, and whoever does make it makes it low. The purchase price was not asking for money: it was asking for certainty. And certainty was precisely what did not exist.
2. Haloid did not cut the price. It changed who carries the uncertainty. Ninety-five dollars a month was not a discount; with the machine averaging two thousand copies a day, the monthly allowance ran out on the first day and everything after it came in at four cents. The arithmetic of those two published figures — two thousand copies daily, four cents beyond two thousand monthly — produces bills on the order of fifteen hundred or two thousand dollars a month, not ninety-five. Which is to say: customers ended up paying considerably more than the $29,500 they had refused to pay. They did not say yes because it was cheap. They said yes because they no longer had to be right.
3. The meter is a claim the seller cannot fake. A brochure saying 'you will use it far more than you think' costs nothing to write. A contract where you only get paid if they actually do use it is the same sentence backed by your company's balance sheet. Haloid had been making that kind of bet since 1946, when it paid for the licence with ten percent of the previous year's earnings, and in 1959 it risked all of its assets to launch the 914. The payment structure told the customer something no demonstration could: we lose if this goes unused.
4. The twist: the forecast was correct and still useless. Here is the counter-intuitive part. Arthur D. Little's five thousand machines were not a measurement error; they described the behaviour that was possible under the contract that existed when the measurement was taken. The nine and a half billion copies of 1964 were not hidden, waiting to be discovered: they did not exist yet. Nobody made two thousand copies a day because nobody could, and a survey cannot capture an appetite that has never had the chance to show itself. Every demand forecast measures possible behaviour under current conditions. When the thing you are changing is precisely the conditions, the forecast becomes an exact description of a world you are about to withdraw.
Take your offer and find the impossible estimate: at what exact point does your customer, in order to say yes, have to predict something about themselves they cannot know? How much they will use it, how much it will save them, how many people on their team will adopt it. That point is your $29,500, and you have probably been reading it as a price objection.
And the uncomfortable one. If tomorrow you charged purely in proportion to what your product ends up doing — nothing if they don't use it, a lot if they use it a lot — would your revenue go up or down? Answer that before you start listing the reasons you can't do it. Because if the honest answer is 'down', you have just discovered that your current price is not held up by the value: it is held up by the customer being unable to check.
If the missing invention can be in the contract rather than in the product, the work is not to build more. The layers:
Walk your sales process and mark the moment the customer has to forecast their own future behaviour. It is almost always hiding inside one word: annual, licence, rollout, plan, minimum. While that stays there, every discount you make buys a little bit of yes and never touches the cause, because the problem was never how much you ask for — it was what you force them to know.
These are two different levers and they get confused constantly. A discount reduces what you take in and leaves the buyer's uncertainty untouched. A structure that charges in proportion to use or to outcome leaves what you can take in untouched — in the 914's case it multiplied it — and removes the uncertainty. If you can only test one thing this quarter, test the second: it is the only one of the two that can raise your revenue and your conversion at the same time.
Haloid learned what the world's real copy consumption was because it was billing for it copy by copy; no study would have told it, and in fact the study said the opposite. If you charge by usage, instrument what the billing teaches you: which customers run ten times the average, which specific task drives the volume, at what point in the month. That is market research that pays for itself and that measures behaviour instead of intent.
Moving the risk means you are the one who takes it, and this same mechanism has a seat on the other side: it is exactly what today drains AI budgets in months. Charging by usage only works if your marginal cost per use is small relative to what you charge — a copy cost Haloid cents — and if you have the cash to finance the product while the meter fills up. Haloid risked all of its assets to carry a fleet it had not yet been paid for. If your marginal cost is high, or if you cannot absorb the lag, the contract that saved Xerox is the one that sinks you. The question is not 'should I charge by usage?'. It is 'who can afford to carry this uncertainty, me or my customer?', and answering it honestly decides both things.
Twenty-one years, more than twenty rejections, and a technology that did not change in all that time. The only thing that changed was to stop asking for $29,500 up front and start asking for ninety-five a month plus four cents a copy. With that, twenty million copies a year became nine and a half billion in under a decade. The useful part is not that Xerox made a lot of money. It is that for twenty-one years the problem looked technical, then it looked like price, and it was neither: it was that the buyer had to guess and did not want to. Before you go build one more feature or sign one more discount, look at where your offer asks somebody to be right. Sometimes what is missing is not in the product. It is in the line that says who carries the risk if you are wrong.