Summary. Retail technology spending divides into two categories that behave very differently: tools that let the customer serve themselves, and tools that make the associate more capable. The industry has just run an enormous, expensive experiment on the first category, in public, and the results are worth reading before the next budget cycle. This paper sets out what the reversal proved, answers the case that it was an execution failure rather than a strategy failure, and gives you a way to price the two options against each other.

The most useful lesson available for free

Self-checkout was deployed at scale, then partly reversed by some of the largest retailers in the country, and they explained why on the record.

Dollar General removed self-checkout entirely from 300 stores and limited it to five items or fewer at a further 4,500. Chief executive Todd Vasos said the company had relied “too much” on it, and that “we should be using self-checkout as a secondary checkout vehicle, not a primary.” He described the changes as having “the potential to have a material and positive impact on shrink.”¹

Target limited express self-checkout to ten items or fewer across most of its roughly 2,000 US stores.¹ A year on, it reported what the change produced: average transaction times across both self-checkout and staffed lanes fell by nearly 8 percent, its net promoter score for wait times rose 5 points and for interactions rose 3 points, and more customers chose fully staffed registers. Target also gave store managers control over the ratio of cashiers to self-checkout lanes through the day, and invested in staff training alongside the change.²

Read that last paragraph twice. The fix for a technology problem was partly a staffing and training decision, and it improved both speed and satisfaction at the same time, which is the combination the original automation was supposed to deliver.

It is rare to get evidence of this quality. Two large operators, at fleet scale, publishing the direction of the change and the result, in a category where most claims are vendor-supplied. Whatever else is true, this is not a small pilot in a friendly store.

What the reversal actually demonstrates

This is not an argument that self-service is bad. It is an argument about where the work goes.

Automation that removes the associate does not eliminate the labor in a transaction, it transfers it to the customer, and it removes the person who was also doing several other jobs at the same time: noticing the item that did not scan, answering the question that would otherwise become an abandoned basket, and recognizing a regular. When those side effects are priced, the saving is smaller than the business case assumed.

The corollary is the useful part. Technology that makes the associate faster, better informed or more confident does not have that failure mode, because it adds capability to the person already standing there rather than removing them.

“That was bad execution, not a bad strategy”

The strongest objection is that self-checkout was deployed clumsily, in the wrong formats, at the wrong basket sizes, and that the correct conclusion is to do it better rather than to do less of it.

Partly true, and worth holding onto. But three things in the record cut against treating it as only an execution problem.

The correction was toward the associate, not toward better machines. Target’s reported improvement came with manager discretion over lane mix and investment in staff training.² If the diagnosis had been purely technical, the fix would have been technical.

Both speed and satisfaction improved together. That combination is the tell. If the original deployment had simply been over-extended, you would expect to trade throughput for experience when pulling it back. Getting both suggests the model itself was mispriced rather than merely oversized.

The failure mode was invisible in the business case. Shrink and abandoned baskets do not appear in a labor saving calculation, which is exactly why the deployment looked good on paper for years before anyone reversed it. A strategy whose costs are systematically outside its own business case is a strategy problem, not an execution one.

The concession is real and should be stated plainly: the answer both retailers arrived at was fewer lanes and tighter limits, not zero. Self-service at the right basket size is a good tool. The error was treating it as the default rather than as one lane type among several.

The test to apply before committing budget

Ask one question about any proposed spend: does this reduce the number of interactions, or improve them?

Both can be legitimate. But the answers carry different risks, and the risks are asymmetric. Reducing interactions saves a known amount of payroll and puts an unknown amount of revenue and shrink at stake. Improving interactions costs a known amount and puts a smaller, better-understood upside in play. In a business where the store is the profitable channel, systematically betting on the first category is a strange strategy.

Putting the two options side by side

The comparison is usually run as a labor saving against a capital cost, which flatters the customer-facing option by leaving out everything it puts at risk. A fairer version takes four steps.

1. Price the labor saving honestly. Hours removed times loaded cost, less the hours that come back as assistance, exception handling and queue management around the machines.
2. Price the leakage. Your own shrink differential between assisted and unassisted lanes, if you measure it, and your own abandonment rate when nobody is available to answer a question. If you do not measure either, that absence is the finding.
3. Price the associate-facing option on the same page. Cost of the capability change against the gap between your best and worst quartile on whatever metric the interaction drives, whether that is attachment, conversion or return rate.
4. Compare the downside cases, not the base cases. If the self-service option underperforms, you discover it in shrink and in customer scores, slowly. If the capability option underperforms, you discover it in adoption data within a quarter and can stop.

Most technology decisions are made on the base case. In this category the difference between the two options shows up almost entirely in how badly each can go wrong and how long it takes you to find out.

Five questions for a technology proposal

1. Who does the work after this ships? If the answer is the customer, say so explicitly in the business case, and price the abandoned baskets and the shrink.
2. What happens when it breaks on a Saturday? Customer-facing hardware fails in front of customers. Associate-facing software fails in a back room.
3. Does it need a trained person to work properly? Most customer-facing technology quietly does, and the training is usually funded last.
4. Can we measure the result within a quarter? If the benefit only appears in a three-year model, it is a bet rather than an investment.
5. Would we buy this if it produced no press coverage? A surprising amount of in-store technology exists to be seen rather than used.

Where customer-facing technology genuinely wins

A paper that argued against all self-service would be wrong, and easy to dismiss.

Small baskets are exactly what self-checkout is good at, which is precisely the conclusion both retailers above reached rather than abandoning it. Inventory visibility, endless-aisle ordering, appointment booking and digital receipts all remove genuine friction without removing a person from a conversation that needed one. The distinction that matters is not the presence of a screen. It is whether the technology replaces a judgment an associate was making, or supports one.

A useful way to sort a roadmap: for each item, name the judgment involved. If there is no judgment, automate it without hesitation. If there is one, the question becomes whether the machine makes it better than a well-briefed person would, and in a store the honest answer is usually no.

Where to point the money instead

The associate-facing equivalent is unglamorous and cheap by comparison: product knowledge that arrives before the product does, short content the floor can actually finish, comprehension that is checked rather than assumed, and results the store manager can see and coach against.

None of that will feature in a trade press round-up of store innovation. It changes what happens in the conversation that decides the sale.

It also compounds in a way hardware does not. A device fleet depreciates from the day it lands. A floor that has been taught how to handle the objection on this season’s range carries that into next season, and the capability transfers to the next launch, the next format and the next initiative.

The returns argument is the sharpest version of this case, because it puts a number on what a well-advised purchase is worth. Or model it directly on your own figures.

References

1. Modern Retail, Why retailers like Target and Dollar General are limiting self-checkout, March 20, 2024. Dollar General removed self-checkout at 300 stores and limited it to five items or fewer at 4,500 more; comments from chief executive Todd Vasos. Target limited express self-checkout to ten items or fewer at most of its roughly 2,000 US stores. modernretail.co
2. Retail Dive, reporting Target’s results from its express self-checkout change, May 9, 2025. Rolled out March 2024; average transaction times across self-checkout and staffed lanes down nearly 8 percent; net promoter score for wait times up 5 points and for interactions up 3 points. retaildive.com