The Self-Checkout Redesign That Cut Labor Costs and Multiplied the Losses It Was Meant to Prevent

Retailers didn't remove self-checkout because customers hated it. They removed it because customers, a meaningful share of them, used it exactly as designed and just didn't pay.

Walmart has been quietly pulling self-checkout lanes out of select stores through 2026, most recently at a South Philadelphia Supercenter, and replacing them with staffed registers. A company spokesperson framed the change as a customer service upgrade, citing feedback from shoppers and associates and a goal of more personalized, attentive checkout.

The framing leaves out the number that actually drives the decision. A recent industry survey found that 27% of self-checkout users admit to stealing while using it. Walmart has reported losing close to $6.5 billion to inventory theft and scanning errors in a recent reporting period. And the company isn't acting alone. Dollar General removed self-checkout from 12,000 stores in 2024. Sam's Club is replacing self-checkout with staffed, AI-assisted scan-and-go technology. Costco has started placing employees at the door to check cart contents before customers reach the register.

One data point makes the mechanism unusually easy to see. At a Walmart in Shrewsbury, Missouri, police calls to the store dropped from 509 to 183 over a comparable five-month period after self-checkout was removed. Theft arrests at the location fell from 108 to 49 over the same window. Nothing else about the store changed. The technology at the front of the building did, and the numbers moved with it.

A single retailer rolling back a checkout format is an operations story. Four of the largest retailers in the country independently reaching the same conclusion, at the same time, is a behavioral pattern, and it's now spreading into policy. Lawmakers in California, Connecticut, Massachusetts, New York, Ohio, Rhode Island, and Washington are pursuing legislation to regulate self-checkout deployment, including minimum staff-to-kiosk ratios.

That's the detail worth sitting with. This didn't stay contained to loss prevention departments quietly adjusting store layouts. It became a multi-state legislative response, which only happens when a mechanism is consistent enough, and costly enough, that regulators conclude the market isn't going to correct it on its own.

Self-checkout was built on a labor-cost calculation: replace the variable cost of a cashier's wage with the fixed cost of a kiosk. That math treated the cashier's other function, simply being a visible person watching the transaction, as free to remove. It wasn't.

Routine activity theory: A framework from criminology holding that an opportunity for theft requires three things to converge: a motivated offender, a suitable target, and the absence of a capable guardian. Removing any one of the three removes the opportunity. Retail loss prevention has traditionally focused on the first two. Self-checkout quietly eliminated the third.

The cashier wasn't just processing transactions. Their presence was the capable guardian in the equation, and removing it didn't only affect people who arrived already planning to steal. It changed the calculus for people who wouldn't otherwise think of themselves as dishonest at all.

Moral hazard: A situation in which a party is insulated from the consequences or observation of a behavior, and as a result, behaves differently than they would under full exposure to those consequences. The behavior isn't caused by a change in character. It's caused by a change in what the situation makes visible.

This is the part that makes the 27% admission rate make sense. Most of that group is not a population of committed shoplifters who found a new venue. It's a population whose behavior shifted because the specific condition that had been quietly restraining it, a person watching, was engineered out of the transaction in the name of efficiency. The self-checkout kiosk didn't create dishonest people. It removed the one variable that had been keeping a share of otherwise honest people honest.

Notice what Walmart's public statement does and doesn't say. Customer feedback, associate feedback, personalized service, shorter lines. Not shrink. Not theft. Not the $6.5 billion figure sitting in the same reporting period.

That's a deliberate choice, and it's a reasonable one from a legal and reputational standpoint. Publicly stating "we built a system that made theft too easy and are now reversing it" invites scrutiny that "we're investing in customer experience" does not, especially with state legislatures already watching. But the gap between the stated reason and the actual reason isn't free. Store associates who process the shrink numbers know the real driver. Customers who've had a receipt checked at the door, a practice already generating its own backlash at other chains, correctly sense they're being treated as a suspect population even while corporate messaging talks about personalized service. When the stated rationale and the operating rationale diverge this visibly, the people closest to the change stop trusting the next stated rationale too, whatever it turns out to be.

The self-checkout kiosk didn't fail at scanning. It failed at the one job nobody put in the product spec: making dishonesty feel observed.

The market is already running several versions of the fix in parallel, and they're not equally well targeted.

• Reintroduce the guardian where the risk actually concentrates, not uniformly. Full removal treats every product and every shopper as equal risk. Shrink is not evenly distributed. Staffing or monitoring calibrated to high-shrink categories, alcohol, meat, over-the-counter medication, does more with less than a blanket return to staffed lanes everywhere.

• If a scan-and-go or AI-assisted model replaces a human guardian, verify it actually functions as one. Routine activity theory was built around a human presence that carries social and observational weight. A sensor array or an app-based system may deter theft through a different mechanism entirely, likely detection risk rather than felt observation, and retailers should treat that as an open question to test, not an assumption to build a rollout on.

• If a receipt check or monitoring policy is going to be part of the fix, make it visible, consistent, and explained before the customer reaches the door, not applied selectively at the exit. A policy applied unpredictably reads as suspicion of the individual. The same policy applied consistently and explained in advance reads as a store rule, which is a materially different experience for the honest majority being asked to comply with it.

• Separate the labor-cost decision from the loss-prevention decision when evaluating any checkout format going forward. Self-checkout's original business case never priced in the guardian effect because it was never treated as a cost. Any future automation decision, in retail or elsewhere, should explicitly model what oversight function is being removed, not just what labor cost is being saved.

• The fix reintroduces the exact cost the technology was built to eliminate, at the worst possible time to absorb it. Grocery prices were already running 2.7 to 2.9% higher year over year through 2026. Adding back cashier labor costs industry-wide lands directly on margins already under pressure from food inflation, and that tension doesn't resolve just because the theft numbers justify the reversal.

• Blanket removal punishes the majority for the minority's behavior. Roughly three-quarters of self-checkout users, by the same survey, are not stealing. Removing the format entirely makes checkout slower and less convenient for that majority to solve a problem caused by a minority, which is a real tradeoff even when the theft numbers make the decision financially rational.

• The AI scan-and-go fix is a genuine unknown, not a proven substitute. Sam's Club and Costco are effectively betting that algorithmic monitoring reproduces the deterrence effect of a human guardian. Routine activity theory offers no strong guarantee that it does. If detection risk without felt observation doesn't move behavior the same way, these retailers may be rebuilding the same vulnerability with better cameras.

• State-mandated staffing ratios apply a uniform standard to stores with very different risk profiles. A legislative minimum ratio doesn't distinguish between a location with Shrewsbury-level shrink and one with almost none. Retailers forced to staff to the mandate rather than to the actual risk are paying for guardianship in places that may not need it, while the highest-risk locations get the same ratio as everywhere else.

If any part of your organization runs on an honor system, expense reporting, remote work output, self-reported numbers of any kind, genuine question: when did you last check whether removing the oversight also quietly removed the thing that was keeping people honest, and whether you priced that risk in before or after it showed up in the numbers?