In 1923 the B.F. Goodrich Company started selling rubber galoshes that closed with a slide fastener instead of a row of metal buckles. You pulled a small tab up the front, and the boot shut. Goodrich called them Zipper Boots, and they sold. Within a few years people were using Goodrich's word for the fastener rather than the boot, and they still do.
The fastener had been invented for boots thirty years earlier. On boots, it had failed.
The claim
Plenty of small and mid-size businesses tried an AI use case around 2023, watched it make something up or break on the tenth try, and crossed it off. Many of those were the right use case tried with a tool that wasn't ready yet. The need was sound, and the tool jammed. Tools like this don't stay jammed. If you wrote off an AI use case two or three years ago, the list of things you gave up on may now be the most useful list you have.
Key point: A failed tool is not a verdict on the need. Keep the need, and test the tool again.
The boot that failed first
In August 1893 Whitcomb Judson, a Chicago engineer, was granted patents for what he called a "clasp locker," a chain of hooks and eyes that a slide pulled together. The patent was for shoes. High boots of the day closed with a long row of hooks or buttons, and people did them up one at a time, every morning, often with a buttonhook.
Judson's device didn't work well enough to replace that routine. He and his backers set up the Universal Fastener Company and spent more than a decade trying to make it sell. Their 1905 version, the C-curity, went onto skirts, and it had a serious flaw: under strain the hooks popped apart, whatever the slider was doing. A fastener that opens on its own is worse than a row of buttons, because you can't trust it. Few people bought one.
The fix came from Gideon Sundback, a Swedish-born electrical engineer the company hired in 1906. By 1913 he had given up hooks altogether. His "Hookless No. 2" used small cupped teeth, each with a bump on one side and a dent on the other, which nested into their neighbors as the slider pressed them together. He raised the count from four fastening elements per inch to ten or eleven. His patent for a "separable fastener" was granted in March 1917.
Then he solved the part that made it a business. Sundback built a machine, called the S-L or "scrapless" machine, that cut the teeth from shaped wire with no waste and clamped them onto cloth tape in a continuous strip. The first machines turned out a few hundred feet of fastener a day. A device that had been fiddly to assemble became something a factory could make cheaply and the same way every time.
Even then, it didn't go straight back onto boots. The company lived on small, forgiving uses. In 1917 a New York tailor began sewing the fasteners into money belts for sailors, whose uniforms had no pockets, and nearly all of the company's 24,000 sales that year were money belts. In 1918 the Navy ordered 10,000 for aviators' flying suits. In the early 1920s the fastener went onto tobacco pouches. None of these was a big market. Each was a place where a failure cost little and a success could be seen.
Goodrich came next. The company tried the fastener on galoshes, and its president, Bertram Work, wanted an action word for them, so he called them Zippers. The boots sold, and clothing followed in the 1930s. Children's clothes came first, sold as a way for children to dress themselves. In 1937 the zipper won what the Smithsonian calls the "Battle of the Fly," when Esquire called the zippered trouser fly the newest tailoring idea for men.
| Year | What happened |
|---|---|
| 1893 | Judson patents the clasp locker, for shoes |
| 1905 | The C-curity version pops open under strain and sells poorly |
| 1913 | Sundback's Hookless No. 2 replaces hooks with interlocking teeth |
| 1917 | The separable fastener patent is granted, the scrapless machine runs, and money belts sell |
| 1918 | The Navy orders 10,000 for flying suits |
| Early 1920s | Tobacco pouches |
| 1923 | Goodrich sells Zipper Boots |
| 1930s | Children's clothes, then trouser flies in 1937 |
The boot was the first use case, and it failed. Thirty years later, with better teeth, a cheap machine and a good name, the boot was the use case that made the zipper famous. The need never changed. The tool did.
Five lessons from one fastener
- Separate the need from the tool. People wanted to stop fighting their boots in 1893, and they wanted the same thing in 1923. When a first attempt fails, write down which part failed, the need or the thing you tried to meet it with. Most teams record only "tried it, didn't work," and that sentence erases the difference.
- Dependable beats new. The zipper didn't catch on when it became possible. It caught on when it stopped popping open. A tool crosses over to ordinary buyers when it becomes boringly reliable, because an ordinary buyer has no time to watch over it.
- Cost is part of the invention. Sundback's machine mattered as much as his teeth. Many ideas wait on a cost curve rather than on a better insight, and a use case that didn't pay at one price can pay at a tenth of it.
- Live on small uses while the big one matures. Money belts and tobacco pouches kept the company alive and gave it somewhere to learn. A pouch that opens spills some tobacco. It doesn't lose a customer.
- The name can open the door. Goodrich sold Zipper Boots, not hookless slide fasteners. Packaging is part of adoption, and it often comes last, from someone other than the inventor.
The same arc, three more times
Tablets. Apple sold the Newton from 1993 to 1998, and its handwriting recognition became a punchline. Microsoft pushed Tablet PCs from 2002, with pens and a version of Windows that still expected a keyboard. The need, a computer you hold in one hand and touch, was right both times. The iPad worked in 2010 because touchscreens, batteries, chips and a mobile operating system had caught up with it.
Video calls. AT&T showed the Picturephone at the 1964 New York World's Fair and launched commercial service in Pittsburgh in 1970, at $160 a month. It peaked at a few hundred subscribers and was shut down within a few years. People did want to see each other while they talked. They wanted it at a price they could pay, with someone on the other end to call, and that took cheap cameras, broadband and a laptop on every desk. Zoom, founded in 2011, made video calls ordinary, and the pandemic made them universal.
Grocery delivery. Webvan raised about $800 million, built automated warehouses, and went bankrupt in July 2001. Instacart launched in 2012 with no warehouses at all. It used the stores that already existed, shoppers carrying smartphones, and gig drivers. The need was the same. The logistics underneath it had changed.
Key point: In each case the first product failed, and people read the failure as proof the need was wrong. Each time the need came back and found a better tool.
The AI use case you gave up on
Here is the pattern I see with small and mid-size businesses. In 2023 someone tried having ChatGPT draft customer replies, pull line items from invoices, or summarize contracts. It invented a clause, dropped a line, or wrote an email the owner would never send. The owner concluded that AI doesn't work for this business and moved on. At the time that was a fair reading, because the tools often weren't dependable. In June 2023 a federal judge in New York fined two lawyers and their firm $5,000 in Mata v. Avianca for filing a brief that cited six cases ChatGPT had made up. That was the clasp locker popping open in public.
The teeth have changed since then. One concrete example: in August 2024 OpenAI shipped Structured Outputs, which forces a model's answer to match a data format you define. On OpenAI's own test of following a complex format, the new model scored 100 percent, and the GPT-4 model from mid-2023 scored under 40 percent. If you tried to pull invoice fields into a spreadsheet in 2023 and got garbage on one invoice in ten, that is the difference between a clasp locker and a zipper. The cost curve has moved too. The price of a fixed level of model capability has been falling several times over each year, and models now read a long document whole, point to where they found an answer, and check their own work with tools.
None of that means every use case you tried now works. Some needs were never there, and some tools still aren't ready. The point is that you can't tell which is which until you test again, and most businesses never do.
Principles of Disruptive Innovation
Every truly disruptive innovation ultimately solves a contradiction.
Every solved contradiction was once an un-solved contradiction.
When you solve a contradiction, express the contradiction that you solved — as a contradiction.
Matrix Morphology framework from David Quimby & Innovation Radiation Associates.
Keep a list of failed but valid use cases
The fix is a habit rather than a project.
- Write down each use case you gave up on, with its test. Record the need in one sentence, what you tried, and the exact inputs where it broke: the five invoices it misread, the three customer emails it answered badly. Those inputs are your test set, and they're the most valuable thing on the list.
- Retest when the tools change. Two or three times a year, or when a major model ships, run the same inputs through the current tools. It takes an afternoon. Hold the output to the bar you'd hold a new employee to.
- Move a passing case into a low-stakes pilot. Put it where a mistake is cheap and easy to see, such as an internal draft that a person reviews rather than a message that goes straight to a customer. That's your money belt.
- Widen it only when it's boringly dependable. When the person reviewing the output stops finding problems, give it more to do.
While the list waits, build confidence on the uses where a mistake costs little: first drafts, internal summaries, sorting and tagging. Those are your tobacco pouches. They teach your people what the tools do well and where they break, so the team can judge the retest when a bigger use case comes up again.
What I'd do this month
Go back to the galoshes. Goodrich didn't discover a new need in 1923. The need had been on people's feet since 1893, waiting for a fastener that didn't pop open. Your list of abandoned AI experiments is probably full of needs like that one.
So I'd find the notes from those 2023 trials, or ask the people who ran them, and write each one down with the inputs that broke it. Then I'd retest the three that mattered most. At Netrii this is work we like doing with owners: sorting the needs that were never there from the tools that weren't ready, and retesting the rest on your own documents.
The open question is timing. Judson's need waited thirty years, and today's tools change in months. How often should a small business look again at what it gave up on? My working answer is whenever the tools change enough that your own test set would notice, and the only way to know that is to keep the test set.
The ruts people fall into when they do adopt AI are the subject of The Faster-Horse Trap in AI Adoption. The falling cost of routine AI work, measured against offshore labor, is the subject of AI shoring.
