Every major loss driver AP fights was chosen by someone else. Offenders choose to steal. Employees choose to sweetheart. Crews choose your stores.
Self-checkout is the exception. It is the only structural shrink driver in modern retail that the industry selected, purchased, installed, and defended — on purpose, with a business case.
Three weeks ago, that business case finally got audited. ECR Retail Loss published the largest study ever conducted on self-checkout and loss: 39 retailers, more than €1 trillion in combined turnover, seven of the top 30 retailers on earth. It is the first loss dataset at this scale that was not produced by someone selling the fix.
The numbers say the quiet part out loud: the machine works exactly as designed. The design just never included the loss line.
The labor math that built the machine
Self-checkout was never a customer-experience project. It was a labor arbitrage: fewer cashier hours per transaction, in exchange for trusting the customer to scan. That trade scaled faster than almost any technology in store operations:
96% of grocery retail now runs some form of self-checkout — and in stores that have it, an average of 54% of all transactions go through it (ECR Retail Loss, 2026).
217,000 new terminals were delivered globally in a single year, and the market is projected to roughly double from USD 6.9B to USD 13.5B by 2030 (Statista; Fortune Business Insights).
The machines take transactions, not revenue: 54% of transactions but only 41% of sales — small baskets move to self-checkout while the big baskets stay with people (ECR Retail Loss, 2026).
The deal was explicit and the savings were real. What was never priced into the model is what the industry is only now able to see: the cost of the trust itself.
The audit arrived eight years late
The first serious measurement — Adrian Beck's 2018 study for ECR — covered 13 retailers. The 2026 update triples the sample and, for the first time, measures before-and-after at scale. The findings:
Stores with self-checkout lose 33% more than stores without it — a difference of +0.42 percentage points of shrink (ECR Retail Loss, 2026).
Installing self-checkout raises loss by an average of 22% in the first year, measured across before/after studies covering more than 1,300 test stores (ECR Retail Loss, 2026).
Every additional 1% of transactions pushed through self-checkout adds 0.030–0.048% of loss — three to five times the rate reported in 2018 (ECR Retail Loss, 2026).
Run that last number against your own fleet. A store that moves from 40% to 60% self-checkout penetration adds up to nearly a full point of shrink. On a business with 2–3% net margin, that is not a checkout format decision. That is the P&L.
And the vendors knew where the loss lived. As we covered in Edition #02, NCR Voyix's own patent filings state that roughly one-third of total retail shrink occurs at self-checkout lanes. The industry's business case and the vendor's patent application were describing two different machines.
Nobody can tell you if it's theft
Here is the part of the report that should stop the "crime wave" narrative cold. The retailers in the study cannot agree on how much of this loss is even intentional:
Estimates of how much self-checkout loss is malicious range from 6% to 80% — the report's own conclusion is that the methods retailers use to estimate intent are not reliable (ECR Retail Loss, 2026).
Missed scans are the most frequent loss event (1–4.8% of transactions), but walkaways are the most expensive — €88 (about USD 103) per sustained incident, typically a soft-declined card discovered after the customer has left (ECR Retail Loss, 2026).
Every 10,000 self-checkout transactions generate roughly 1,000 staff help calls and 90 walkaways — the "self-service" machine runs on a human intervention layer nobody staffed for (ECR Retail Loss, 2026).
This ambiguity cuts in both directions. Dollar General pulled the machines from 12,000 stores and credited the removal with 62 of its 105 basis points of margin expansion (Edition #01). Maybe that attribution is right. But the ECR report's closing warning applies to everyone: business cases for self-checkout "continue to rest on assumptions rather than evidence."
So do the business cases for ripping it out.
A perspective from the field
I have sat in the meetings where self-checkout gets approved — in four countries. The deck is always the same: hours saved per transaction, payback under two years, one customer-experience slide to make it feel strategic. In none of those meetings did I ever see a loss projection on the same page as the labor savings.
The loss conversation happened later, in a different room, with different people, after the machines were bolted to the floor. By then the capital was spent, and the question was no longer "should we do this" but "how do we defend it." The most expensive line in any business case is the one that was left out of it.
What changed my mind about self-checkout was not the theft. It was the inventory. Once the machines corrupt your on-hand accuracy, every downstream system — replenishment, availability, e-commerce promising — inherits the error. The shrink is what you see. The corrupted data is what you pay for.
Three practical moves for the next 90 days
Put a loss number on your own penetration curve — this quarter, not at refresh time. Take ECR's coefficient (0.030–0.048% of loss per point of self-checkout penetration) and run it against your store-level penetration and shrink data. You are looking for the stores where the math says the machine stopped paying for itself. That list, not a chain-wide opinion, is the deliverable.
Stop reporting intent you cannot measure. If the industry's estimates of malicious loss span 6% to 80%, any "theft vs. error" split in your reporting is fiction with a decimal point. Instrument events instead: missed-scan rates, walkaway counts, help-call volume per terminal. Report what happened. Let the pattern, not the adjective, make the case.
Treat removal as a capital decision with the rigor installation never got. Before pulling machines, run matched-store pilots and measure shrink, labor cost, queue times, and sales transfer together. And take the one free move the data supports: do not turn off the weight scales — the report found that switching them off measurably increases loss.
Closing note
Self-checkout is not a mistake. For thousands of stores the labor math still clears, and the ECR data suggests the losses are manageable when the right controls — exit gates, personal display monitors, missed-scan detection — are actually deployed.
The mistake was the accounting. Retail booked the labor savings in year one and left the loss discovery to a research group working eight years behind the rollout. The next decade belongs to the leaders who refuse to let any technology — installed or removed — skip the loss line twice.
I'd like to hear from leaders who have run the store-level math on their own self-checkout fleet. Did your loss curve match ECR's coefficients — or did your stores tell a different story? Reply with anything you can share, anonymized if you prefer.
Forward this to one LP or AP leader who should be reading it.
— Gabriel
The LP Brief is a weekly intelligence read for senior loss prevention and asset protection leaders. Free. No vendor noise.
Not yet subscribed? thelpbrief.com