Case study — Retail & consumer · 12 min read · 5 questions
Why self-checkout didn't replace cashiers
The thesis
Retail is a brutal, low-margin business and everyone in it knows why. Sixty to seventy percent of every dollar goes to buying the merchandise, leaving about twenty points of gross profit, and another fifteen to thirty goes to labor and rent. What is left is 1.8% at Albertsons, 2.5% at Kroger, 4.5% at Walmart. That thinness had made retailers conservative for decades.
Then Amazon opened one store. Just Walk Out promised computer vision that eliminated the checkout aisle entirely, turning an unavoidable variable cost into a one-time capital investment. It was a proof of concept and everybody believed it. Retailers could not match Amazon's engineering, so they bought the nearest thing from NCR and Diebold and got customers scanning their own groceries. It was never about the economics — it was about not looking like a dinosaur to a market paying tech multiples.
The technology was not real. Amazon had more than 1,000 people in India manually reviewing transactions — roughly 30% of purchases were genuinely automated — and has since closed most of its Go stores. The retailers who copied it have nothing to show either: Walmart's labor costs went from 19% of revenue to 21%, Kroger has more staff per store than before, and Dollar General found that removing the human at the front multiplied shoplifting, hit a sixteen-year profit low, and ripped the machines out.
How do you think about this? 5 strategy questions this case raises and does not answer.
Read the comments, or add yours →
The statistics
By the numbers — swipe or use arrows
Staff per store, selling/general/administrative costs as a share of revenue, cost of merchandise and operating margins from Walmart, Kroger, Albertsons, Sprouts, Dollar General, Dollar Tree, Target, Home Depot and Lowe's annual reports and 10-K filings for the years shown; NCR and Diebold revenue and operating income from company filings; Just Walk Out staffing and manual-review figures as reported by The Information in May 2023
Key takeaways
Retail runs on almost nothing. Average operating margins over the last decade: Albertsons 1.8%, Kroger 2.5%, Costco 3.2%, Walmart 4.5%, Sprouts 5.4%, Target 6.3%, Dollar Tree 6.3%, Dollar General 8.9%. These are the biggest and best-run retailers in the country.
Merchandise eats it. Cost of goods runs 87% of revenue at Costco, 78% at Kroger, 75% at Walmart, 72% at Albertsons and Village Super Market, 71% at Target and 69% at Dollar General. Even specialists sit in the same band — 67% at Sprouts, 66% at Home Depot and Lowe's, 65% at Tractor Supply, 64% at TJ Maxx, 62% at Whole Foods.
Then the market changed what it rewarded. In the zero-interest 2010s investors wanted upside rather than safety, and tech delivered it — a $1,000 investment held from 2010 to 2019 became $41,255 in Netflix, $13,799 in Amazon, $9,781 in Apple and $5,396 in Meta, against $2,200 in the S&P 500.
The gross margin gap was the whole story. In 2019, Meta ran 82%, Shopify 55%, Google 56%, Netflix 38% — against 22% at Kroger, 25% at Walmart and 13% at Costco. Retail and tech were structural opposites and the market priced them accordingly.
So retailers started buying a tech multiple. Price-to-earnings ratios climbed even though the underlying business did not change: Costco went 17, 18, 24, 28, 28, 32, 34 and 55 across 2011 to 2025. A grocery wholesaler on 55 times earnings is a story about narrative, not margins.
Amazon supplied the narrative in 2016. Just Walk Out promised computer vision, deep learning and smart sensors that would eliminate the checkout aisle — no queues, dramatically lower labor, and an unavoidable variable cost converted into a one-time capital investment. Every retailer was spooked, and there was no catching Amazon's engineering.
The technology was not real. A 2023 report found Amazon had more than 1,000 people in India manually reviewing Just Walk Out transactions, running about 700 human reviews per 1,000 sales against an internal target of 20 to 50. Roughly 30% of purchases were genuinely automated by computer vision.
It did not scale physically either. A single store needed thousands of state-of-the-art cameras and fibre internet. Amazon has since laid off hundreds from retail, closed most of its Go stores in the US and Europe, pulled the turnstiles out of Whole Foods, and returned to conventional checkout with human cashiers.
The profits never came at all. NCR operating income ran $232M, $353M, $135M, $599M, $691M, then $191M, $221M, $26M, $136M, $37M; Diebold went $165M, $59M, −$160M, −$94M, −$326M, −$26M, $24M, $137M, −$212M, $87M. Even the shovel sellers lost money on this gold rush.
Walmart is the clearest test because it went first and hardest. Hourly workers per store went 234, 217, 204, 201, 192, 190, 187, 181, 178, 175 — then back up to 184, 199, 181 and 188 by 2024. Fewer cashiers, not fewer staff.
Its labor costs went the wrong way. Selling, general and administrative expense ran 19% of revenue from 2011 to 2015, then 20% and 21% every year from 2016 to 2023 — the exact period in which self-checkout went nationwide.
And so did the profits. Walmart's operating margin went 6.1%, 5.9%, 5.9%, 5.9%, 5.6%, 5.0%, 4.7%, 4.1%, 4.3%, 3.9%, 4.0%, 4.5%, 3.0%, 4.2%. Inflation cannot explain it either: cost of merchandise held between 74.4% and 75.9% of revenue across the whole period.
Its numbers moved the wrong way on every measure. Workers per store went 108 in 2005 to 158 in 2021 and 144 in 2023 — more staff, not fewer. Labor and administrative costs are at their highest in six years at 17.5%, up from 15.4%. Merchandise costs are *lower* than a decade ago at 77.8%, so inflation is not the culprit. Operating margin sits at 2.1%.
Sprouts is the one chain that genuinely hired fewer people, going from 91 workers per store in 2018 to 75 by 2023 while opening more stores. Labor costs still rose, from 21.7% to 28.5%, and operating margin fell from 6.7% to 5.1%. Fewer staff, higher labor costs, lower profits.
When profits hit a sixteen-year low in 2024 it reversed the entire program. Margins went 9.9%, 9.4%, 9.5%, 9.4%, 8.6%, 8.3%, 8.3%, 10.5%, 9.4%, 8.8%, 6.3% — and Dollar General ripped the machines out, restored staffed aisles and now runs 8.8 workers per store, more than at any point in its history, with SG&A at an all-time high of 24.0%.
The single exception is home improvement. Home Depot and Lowe's cut labor costs from 23% and 22% to 17% and 17%, and margins improved materially — Home Depot from 11.6% to 14.2%, Lowe's from 6.2% to 10.5%.
Common questions
Did self-checkout actually save retailers money?
Almost nowhere. Walmart's labor costs went from 19% of revenue before self-checkout to 21% after, and its staff per store went from 234 in 2011 to 188 in 2024 — fewer cashiers, not fewer people. Kroger has more workers per store than it did in 2005 and its highest labor costs in six years. Sprouts genuinely hired fewer people and its labor costs still rose from 21.7% to 28.5%. Dollar General's went to an all-time high of 24.0%. The one exception is home improvement, where Home Depot and Lowe's cut labor costs from around 23% to 17% and materially improved margins.
Was Amazon's Just Walk Out technology real?
Not in the way it was presented. A 2023 report found Amazon had more than 1,000 people in India manually reviewing Just Walk Out transactions, running roughly 700 human reviews per 1,000 sales against an internal target of 20 to 50 — meaning only about 30% of purchases were genuinely automated by computer vision. The system also did not scale physically: a single store needed thousands of cameras and fibre internet. Amazon has since laid off hundreds from its retail teams, closed most of its Go stores in the US and Europe, removed the turnstiles from Whole Foods and returned to conventional checkout with human cashiers.
Why did every retailer install self-checkout at the same time?
Because Amazon opened one store in 2016 and Wall Street demanded an answer. Just Walk Out was a proof of concept, but Amazon's track record and the market's enthusiasm for tech meant it was believed, and when Amazon bought Whole Foods for $13.7 billion a year later it was clear it intended to run supermarkets rather than license software. Retailers could not match Amazon's engineering, so they bought the nearest available thing from NCR and Diebold. For Walmart, Kroger and Target it was mostly about not looking like dinosaurs to a market paying tech multiples — Costco's price-to-earnings ratio went from 17 in 2011 to 55 in 2025 without the underlying business changing.
Why did Dollar General remove self-checkout?
Theft. It installed self-checkout in 2019 and found that removing the human at the front turned shoplifting into a consequence-free crime of opportunity — people walking out through the machines with bags of merchandise. When profits hit a sixteen-year low in 2024, with operating margin down to 6.3% from 10.5% in 2021, the company ripped the machines out of most stores, restored staffed checkout aisles and invested in cashiers. It now runs 8.8 hourly workers per store, more than at any point in its history, with SG&A at an all-time high of 24.0%.
Why does self-checkout work at Home Depot but not at a supermarket?
Because theft at self-checkout is a crime of opportunity and home improvement merchandise resists it. The expensive items are too big and bulky to hide, the small high-value SKUs are already locked up or under extra surveillance, and order values are far higher, so a single missed scan matters less relative to the basket. Home Depot and Lowe's have also spent decades and millions on loss prevention, where grocery chains are only beginning to invest. Their labor costs fell from around 23% and 22% of revenue to 17% each, and Home Depot's operating margin went from 11.6% to 14.2%.
Did the companies selling self-checkout machines make money?
Barely, and then not at all. NCR and Diebold had built their businesses on ATMs and spent the 2000s hunting for growth beyond banking — NCR even bet on DVD kiosks after Blockbuster's bankruptcy, which Netflix erased. Self-checkout was meant to be their second act. Revenue did climb, NCR from $4.5 billion in 2006 to $6.9 billion and Diebold from $2.9 billion to $4.6 billion, before falling back to $3.8 billion each by 2023. Operating income was erratic throughout and frequently negative at Diebold, including losses of $326 million in 2018 and $212 million in 2022.
Is inflation the reason retail margins fell?
No, and the filings rule it out directly. Walmart's cost of merchandise stayed between 74.4% and 75.9% of revenue across the entire self-checkout period, and Kroger's is actually lower today at 77.8% than it was a decade ago. What rose in both cases was labor and administrative cost — 19% to 21% at Walmart, 15.4% to 17.5% at Kroger — which is the opposite of what installing machines to replace cashiers was supposed to produce.
Discussion
Net margins run 1.8% at Albertsons, 2.5% at Kroger, 4.5% at Walmart. What does that thinness do to a company's appetite for risk, and why did one Amazon store override decades of it?
No answers yet — be the firstJust Walk Out promised to convert an unavoidable variable cost into a one-time capital investment. Why is that trade so seductive, and how often does it actually work?
No answers yet — be the firstIt was a proof of concept and everybody believed it. What separates a proof of concept from a product, and who inside a retailer should have been asking?
No answers yet — be the firstSelf-checkout shifted labor to the customer rather than eliminating it. When do customers accept that trade, and when do they punish it?
No answers yet — be the firstYou run a grocery chain now. What do you do about checkout, given that both the old way and the new way have failed to deliver what was promised?
No answers yet — be the first
Related case studies
Retail & consumerWhy marijuana legalization never paid offEvery public cannabis company loses money. Three self-funded LA shops do not. Being unable to lose money is the advantage.11 min read
Retail & consumerWhy gun prices make no senseThe most lucrative consumer market in America is also the only one with no prices, no ads and no data.12 min read
Retail & consumerWhy UNIQLO makes more money than GucciSame playbook, three outcomes. The only variable was whether the boss knew what stage he was in.12 min read