The Barcode Bargain
What a fifty-year-old argument about price stickers has to say about AI adoption.
On 26 June 1974, a ten-pack of Wrigley's Juicy Fruit passed over a scanner at a supermarket in Troy, Ohio. The machine beeped, the register showed 67 cents, and a packet of gum entered history, complete with what it was missing: the price sticker.
To the supermarket, the sticker was a cost. Somebody had to put one on nearly every item, change it when the price changed, and key the number in at the till. Across American grocery that was a great many people doing slow, repetitive work. To the shopper, the sticker was a promise. It put the price on the product, in plain sight, while the product was still in your hand.
Scanning made it possible to remove most of that work. It also moved the price of record, from something printed on a sticker to a number held in a computer the shopper could not see. The same sticker was a cost to one side and a check to the other, but the business case was written by the side that saw the cost.
The barcode needed a village
Getting to that first beep had taken an unusual amount of cooperation. The grocery industry picked one standard out of several competing designs, which meant rivals agreeing on something none of them controlled and no regulator was going to impose. Manufacturers had to print it on their packaging. Retailers had to buy scanners and computers, build price files, rework checkout routines and train their teams.
By the time the gum crossed the scanner, an enormous amount of genuinely difficult work had been done. But adoption still depended on whether the arrangement made sense to the person on the other side of the counter, and shoppers had a fair question. If the price now lives in the machine, how do I know it is the price I was offered?
Scanner accuracy, and the removal of the individual price marks that let people check, became a real consumer and public-policy issue. Stores were picketed by shoppers. Consumer groups were vocal opponents. By 1976 six states had passed laws requiring stores with scanners to keep pricing every item, which is how a number of grocers ended up paying for the machines and doing the labelling anyway. The shoppers had a vote, and they used it.
Every technology changes an exchange, whether or not anyone writes one down. Something gets faster, cheaper or easier. Something else becomes less visible, less local, or less under the control of the person on the other end.
Scanning was sold partly on efficiency, and it delivered. Later research estimated that early scanners raised grocery store labor productivity by around four and a half percent in their first few years. In a low-margin business that is worth having. It is also a solid single rather than a home run, and the same research found the short-run gains were small against the fixed costs of putting the system in. Plenty of grocers looked at the economics and waited.
The shopper's side of the ledger was thinner. The queue moved a little faster. The benefit did not arrive as a visible reduction in the weekly grocery bill. And the thing they had been using to check what they were charged had moved into a computer.
Both sides were reading their own position correctly. The grocers were solving a real and expensive problem. The shoppers had lost something they used. The exchange was uneven, and it stayed that way for years.
The value left the checkout lane
A faster checkout was helpful. A structured record of every item sold was a different order of thing.
Each scan tied a physical product to a digital system. At scale, that let retailers see what was selling, manage inventory, replenish shelves, forecast demand, plan promotions and coordinate with suppliers. The barcode stopped being a quicker way to enter a product at the till and became the data layer for retail. While the barcode application was at the checkout, the benefits were felt across the whole organization.
The barcode itself did not need to become more impressive. The system around it became more capable.
Some of that found its way back to the shopper. Over time, the same system made prices easier to check, made products more likely to be in stock and, eventually, put scanners in shoppers' own hands. My dad's favorite part of a grocery trip is reviewing the receipt. He goes through the whole list afterwards, checks every saving and announces the total with no small celebration. The receipt gives something back. The system has done its work and the shopper can inspect the result.
Getting the barcode adopted took coordination, capital, adequate returns and enough public acceptance to carry on. Becoming infrastructure happened more slowly, as its value broadened beyond the checkout lane and began appearing in forms shoppers could see and use.
Ubiquity could not be commanded. The barcode became useful enough, to enough people, that eventually using it stopped being a question anyone raised. You can design the conditions for that to a certain extent, but you cannot produce it on demand.
Trust was a response to that bargain, not an attitude toward the machine. The problem was never simply convincing people that scanners worked. It was creating an exchange that made sense.
Why this matters for AI now
A great many organizations have now made a visible investment in AI. The tools are bought and installed, access has been rolled out, the training has been run, the guidance has been written. Those were not easy things to do, and people are now waiting for results.
That is roughly where grocery stood at the end of the 1970s. So what has it returned? It bought capability, and it bought availability. It did not necessarily buy broad returns. Those arrive when the capability is built into work that matters and people choose to rely on the resulting system. That choice belongs to them. The visible deployment can be finished while the operating system around it is nowhere near finished.
People are already forming a view, and they express it through what they use, what they check and what they quietly avoid. That judgement is rarely about the underlying technology in isolation. It is about the arrangement being built around it, and the question is the same one the shopper asked. Is this better for me, and how would I know?
Trust in that sense is not a feeling about technology. It is a judgement about an arrangement, assembled out of what somebody can see, what they cannot see, what it costs them, what they believe it costs them, and what the experience around it is actually like. Being different from before does not settle it. Neither does being better for the provider, even when the provider's benefits are real, carefully measured and honestly reported.
You will probably run both for a while. The system's answer and the human check, side by side, for longer than anyone budgeted. Sometimes that is a choice, and sometimes a regulator or a customer contract makes it for you. It is easy to read as failure and it is not. Grocers ran barcodes and price stickers together for years, in six states because the law required it and elsewhere because customers expected it. Belt and braces is what a transition looks like from the inside, and it is better planned for than treated as a setback.
Some of the return may also turn up somewhere you were not measuring. The barcode was applied at the checkout and paid off across the whole business, and the same pattern is likely here. The application sits on a task. The larger benefit may turn up somewhere else, often outside the line of the business case that funded it.
Running work through a system can produce a usable record of how the work actually runs, where it breaks, what people are really asking for, and where the documented process and the real one part company. Much of that existed before, scattered through tickets and email and spreadsheets and the heads of experienced people. What changes is that it can now exist in a form that is structured, comparable and worth building on, and that record may point at opportunities nobody could have specified in the original case.
The larger value from the barcode did not come from making the scan fractionally faster. It came from noticing what the scan made visible and building something on top of it, which retailers had to choose to do.
Whether any of this becomes ordinary is a separate question. Widespread use is not the goal. It is the outcome. What you can design are the conditions that make it possible. Get them right and you make ordinary use possible, which is not the same as producing it. Aim only at the usage number and you can end up with compliance that looks healthy while the value underneath stays shallow.
Which gives three questions worth asking of any proposal. What gets better immediately? What else gets better if this works? And what changes for the person being asked to rely on it, both what they gain and what they are being asked to give up? A proposal that cannot answer the third has not yet earned the reliance it needs.
The barcode did not become essential because the scanning kept improving. It became essential because a narrow efficiency tool grew into infrastructure that made the whole system work better, while returning enough value, visibility and control to the people around it.
The same test applies to AI. A saved minute is a reasonable place to start. What you are looking for is a system people would notice if it went away.