Self-checkout participation and Loss: what every percent really costs

Self-checkout participation and loss: what every extra percent really costs

Self-checkout participation and loss now have a measurable relationship, and it is the most useful thing to come out of the Self-Checkout Loss Report 2026. Professor Matt Hopkins of the University of Leicester built the study on data from thirty-nine retailers, making it the largest examination of fixed self-checkout loss yet attempted, and an update to the global study that has anchored this debate since 2018.

The headline is not that self-checkout generates loss. Everyone knew that. It is that the loss scales predictably with participation. For every one percent rise in the share of transactions going through self-checkout, shrink rises by between 0.03 and 0.048. Average participation across the study was fifty-four percent, so a business targeting seventy now has a shrink figure it can budget for rather than a target its stores quietly fail to hit.

Matt also shows that risk is not uniform. Stores in high-crime or high-density locations, and those serving younger customers with higher disposable income, cluster at the top of the shrink range. That points at a different operating model for those sites rather than a uniform rollout. More on the wider picture at the self-checkout hub.

The full conversation follows.

How self-checkout participation drives loss: Colin Peacock and Matt Hopkins in conversation

COLIN:

On this podcast I’m with Professor Matt Hopkins of the University of Leicester. About two years ago we commissioned him to do an update on a report around the scale and nature of the problems of loss at the fixed self-checkout: the ones where you take your products and scan them yourself, normally at the front of the store as you leave. At least, that’s the intention.

This report was an update on a publication that Professor Adrian Beck originally published in 2018, a colleague of Matt’s. That paper was important for a number of reasons. It gave benchmark data that retailers were able to use in the business to start building the business case and start to get a right-sized shrink number.

It also gave the loss prevention team hard evidence to go in and talk about the choices organisations have in terms of whether to put in interventions, whether that’s technology, design or people, that might have a chance to reduce some of the losses. So it was a really important paper in 2018. We got asked by numerous retailers, can we do another report?

And hallelujah, we have one, and it’s available to download. I’m with Professor Matt Hopkins, and I thought on this podcast we’d explore some of the key findings and what retailers can do next.

So let’s start with you, Matt. We’ve covered off what we’re trying to do with this report. Let’s focus on the findings. What do we find in this report?

The biggest study yet on fixed self-checkout loss

MATT:

Hi, Colin. I think we were really lucky with this study. We got some quite decent data from a number of retailers. We had thirty-nine who submitted data for us on this. It’s the biggest study that’s ever been done on fixed SCO loss, so that in itself is really encouraging.

In terms of the core findings, there are a number of things we can draw out. The first is around the magic question that everybody always wants to know, around scale: does it generate loss? The straight answer is yes. But no surprise to anybody across the industry, we know that SCO generates loss.

What we were able to do in this report is get some reasonable data on what the scale of those losses are. A number of studies have been done by retailers that have looked at before and after in stores. Measured shrinkage is the key way that retailers are measuring SCO loss, obviously. So where they’ve measured shrinkage before implementation of SCO and then afterwards, in the year after you implement SCO you get about a twenty-two percent increase in losses in those stores.

When you just look at before and after, we also had some data where retailers compared stores with SCO to those without, and the measure there tends to be a plus thirty-three percent increase in shrinkage in those stores where SCO is implemented. So we know there are some losses associated with this.

I think it’s not time to panic about this when we look at it compared to the twenty-eighteen study that Adrian Beck did. Adrian estimated between thirty-three percent and one point four seven percent increase in losses in stores. So a core message for the industry is, yes, it increases losses, but I don’t think those losses are out of control. And I think the industry knows those losses are not out of control.

We also had data on how SCO generates loss, and some indicators around the types of interventions which seem to be promising in terms of reducing those losses.

COLIN:

Well, first of all, as an academic, I know we spoke a lot about this. On that sort of before and after, with or without, you gave me the sense that you were quite happy that the data all seemed to sit quite closely together, which I guess suggests that it’s not as random as we initially found, but it’s clustering around a broad set of twenty percent to thirty percent.

MATT:

Yes, I think the data there is quite reassuring. What we did, we initially got data from thirty-nine retailers, but those who were giving us data that we thought was robust enough to present in the report, the sample actually shrinks quite significantly.

But if you look through those that have submitted what we think is robust enough data to report on, the numbers are quite reassuring. We’re not seeing these massive ranges in increases in shrink that have been previously observed in this area. So it suggests to me that our measures are not wildly inaccurate.

Every one percent of participation adds measurable shrink

COLIN:

Now, we did some really interesting stuff. One of the most useful statistics coming out of the previous report was from Adrian. I think it was based off maybe just one retailer, but he was able to work out that for every one percent of transactions that went through self-checkout, there was something like a one basis point increase in losses.

That was helpful because back in twenty-eighteen a lot of retailers were looking to expand their self-checkout proposition and have greater participation. In this study the average participation was fifty-four percent. Of the retailers on average who submitted data, fifty-four percent of their transactions were going through fixed self-checkouts.

I think we did some really good work in this study around participation that helped us be sure about that progression, that every percentage of transactions you put through self-checkouts will lead to X percent more loss. You’ve got the numbers, so you can reveal the numbers. But I know the one percent number.

MATT:

We were really lucky here. In the twenty-eighteen report Adrian did a pretty sterling job, in all honesty, with fairly limited data. We were lucky because we got some indicators from retailers who’d done some work around utilisation, generally saying that where you have greater utilisation you tend to get greater loss in store. That’s no great surprise. Most retailers would recognise that.

But we were really lucky because we had four data sets that were given to us by three retailers, where we were able to plot out on scatter plots the relationship between utilisation and loss. That enabled us to do a bit of regression analysis on the data, which enabled us to predict, for every one percent increase in participation or utilisation, what you’re seeing in terms of increases in loss and shrinkage.

The figure we came out with was that there seems to be a range here of between zero point zero three and zero point zero four eight increase in loss for every one percent increase in transactions. One of the great things about this is we’ve got data from four data sets across three retailers, and again those figures are not wildly different.

I think it gives us a sense of validation of the data, in the sense that the data seem to be fairly reliable. Obviously, the more data you get and the more consistent that data seems to be, the more confident you can be in it. The ranges we’ve got here, I don’t think, are wildly inaccurate. I think it’s a fairly accurate summary. Most of the retailers we’ve been to with this data seem to be saying, yeah, we think this is about right.

So I think that is reassuring. What we need to do in the future is keep trying to get this sort of data and keep trying to see what those numbers are looking like.

COLIN:

It’s a very practical data point, because if your participation rate is the average of fifty-four percent and the business says we need to get to seventy percent, it would make sense that the store shrink budget increases by that amount, so that the stores are not fighting a fight they can’t win. And often self-checkout does require central interventions, be that design, people allocation or technology.

MATT:

Yes, I think that’s exactly what it does. You can say with a certain amount of confidence, look, if we’re going to look to increase our utilisation by this, this is what is likely to happen.

Why self-checkout risk is not uniform across an estate

MATT:

There are some caveats to that. One of the things this study did, which is a real step forward in this area, is we started to tease out some of the locations, some of the characteristics of stores where you might start to see stores implemented which drift into those higher brackets of loss or shrink. That was a little bit exploratory, the analysis we did there, but it did reveal some quite interesting things.

We were lucky, we got data from two retailers where again we were able to do regression analysis on the data. What it showed us, and it was statistically significant, was that where you’ve got stores in high-risk locations, so that might be stores where there’s high crime risk or high population density, and where you have a customer base with a high proportion of young singles with high disposable income, in those SCO stores they tend to push towards the higher end of the shrink bracket. We looked at above-median shrink as an indicator for those stores, as our outcome variable.

So the data suggests that your risk with SCO is not uniform across the whole industry, across all stores. There are certain stores where your risks are going to be higher. I think this is something for the industry to think about as well. Most people working in retail, in loss prevention or operations, would know where their high-risk stores are.

COLIN:

Yes. One of the retailers in the group pointed out this was a great piece of analysis, because often in this course of push for productivity, the push for self-checkouts, there is this sort of let’s get to as many as we can. But as Dollar General found out when they rolled out, they had pushed self-checkouts to their three hundred ultra-high-risk stores.

MATT:

That’s kind of business suicide in some ways. You’ve got to be really careful where you push SCO out. This is not to criticise the industry. The industry is right in many ways. More and more customers like SCO. There is obviously risk associated with SCO, but in some ways we do have to be a little bit careful thinking about what the loss generators are when we push technology out.

The loss pyramid: what missed scans and walkaways actually cost

COLIN:

Great segue. So we have this pyramid in the report. You broke out, or got data from retailers, in terms of what percentage of transactions would include these types of loss generators or loss drivers, walkaways or missed scans. Take us through that data.

MATT:

Yes. We were able to identify from the data that we had what seems to be driving SCO loss, and we broke this down into a number of categories. We asked about missed scans, walkaways, items left in trolley. We were able to identify the frequency of loss: the proportion of transactions, so if you had one hundred transactions, how would this type of loss feature in those hundred?

To give you an example, for missed scans we managed to estimate that between one and four point eight percent of all transactions going through SCO would include a missed scan. We were able to identify that around zero point zero two to zero point one percent of transactions would involve a walkaway.

But what we also did that was really interesting, we managed to put financial data to this, so what the cost of the loss was in relation to this. Where you have a missed scan, the average cost per event was about two euros fifty. If you compare that to a walkaway, it’s eighty-eight euros for a walkaway on average, which is unsurprising, because obviously with a walkaway nothing is going to have been paid for. With a missed scan you might have one or two items per basket which are missed.

I think this takes us a little bit further from what we were able to do in twenty-eighteen. Going forward, what we need to try and do with this sort of data is get more and more data that can enable us over a period of time to keep going back to the pyramid and seeing where we are. If the loss prevention measures are any good going forward, you should expect to see some changes in the pyramid. You should expect to see that frequency or prevalence of loss come down over a period of time.

COLIN:

You’re right. There are a lot of war stories that go on here in terms of, oh look, this happened, barcode switching, product stacking, walkaways. I think for the first time your pyramid captures the extent to which these are happening according to retailers, and then the impact when it does happen. That was a really good piece.

Which interventions have evidence behind them

COLIN:

Now let’s go on to some of the interventions, because we do need to wrap up here. The interventions data was really interesting. You divided it into things around people. You looked at adding more SCO hosts, training the SCO hosts. We looked at design, use of exit gates, use of mirror monitors. We looked at technology. And we also looked at process to a certain extent.

Maybe just call out what you think the top three or four interventions were, where we had really strong evidence that these are things you might want to consider for your business if you’re not already doing them.

MATT:

There are a couple of things to say about interventions. The first is that obviously there’s huge interest in interventions across the industry and a lot of interventions are being pushed out at the moment. One of the things about interventions we need to think about going forward is how they’re tested, how robustly they are tested. This is a real key thing for the industry to think about.

We did look at design, technology and people-type interventions, so I’m going to give you something from each which I think is promising.

Design: huge interest in exit gates at the moment. Some retailers are pushing out exit gates and are testing them. We found that seven studies we identified from seventeen retailers suggested there was some sort of positive impact of exit gates, so exit gates would reduce loss in some way or help recovery.

In terms of technology, product non-scan identification seemed to be the thing. Seven retailers there out of fourteen that have done some work around this suggested there was positive impact of product non-scan.

In terms of people, people is quite interesting in terms of its impact here, because lots of retailers were suggesting that supervisor training or SCO host training is becoming really important in this area. Having hosts who know how to deal with customers who may be getting frustrated going through the system. Though there’s not a great deal of hard evidence in terms of numbers around the impact of that.

But I think there is a piece here in how you integrate design, technology and people going forward in relation to interventions.

Nudges, and the malicious versus non-malicious question

COLIN:

Let’s close out with this discussion around what we think might be going on and what the role of nudging can be, and perhaps how we can create better behaviour, maybe increased accuracy.

You put together a really interesting framework when we think about how crime happens: the opportunity, how easy are the sanctions, and so on. But you added two other factors which are going on here creating loss, which is simply that sometimes the barcode doesn’t scan, but also there’s frustration, or there’s justification.

What we saw, and what we’re seeing in some of the evidence, is that a simple nudge, whether that’s “Hi Matt, good to see you back at the self-checkouts again, I see you bought another bottle of Jack Daniels, go to till number three”, or the mirror monitor, or the computer vision idea, “we noticed you didn’t scan that last item, please rescan it”.

So if you could just touch on the malicious versus non-malicious, and perhaps the role of nudging and simple behavioural technique changes that we can instigate through the queues, the design, the people, the technology.

MATT:

Yes. In terms of malicious versus non-malicious, this in many ways is the golden question for the industry. We identified that the estimates around what is malicious behaviour in SCO vary massively. We went from something like some retailers suggesting about six percent where there was non-scan activity they thought was malicious, up to about eighty percent in some cases. So there are huge variations here in what we think is happening.

COLIN:

To be fair, it’s hard to measure, right? I mean, when you look at it.

MATT:

Well, yes. This is not a criticism of the industry at all here. It’s really difficult to measure. How do you know what’s going on in someone’s head?

We know that SCO provides camouflage for crime, for want of a better term, where people can mimic scanning activity. In some ways, across the industry, we’ve told ourselves that real shoplifters don’t use SCO. I’m not overly convinced by the argument. I think some people are quite happy to pay for some items and then not pay for some items, and I think they are using SCO as camouflage for crime.

In some cases there are lots of genuine errors through SCO as well, obviously. But identifying what is malicious versus non-malicious is very difficult indeed. I think the interventions that we previously highlighted are really important here.

One of the things we did identify seems to have quite a bit of efficacy: nudges. Where somebody doesn’t appear to scan, or maybe they’re not weighing items properly, then you intervene with a nudge on the screen: look, do you want to rescan the item? Or maybe where you get playback video technology showing what you’ve done. They seem to be very effective in terms of changing that behaviour of customers.

So if you do have customers who would normally pay for items, but maybe get a bit frustrated and use techniques of neutralisation to justify why they might not then rescan an item, I think those are the types of customers you can pull back through nudges quite easily. More work needs to be done around nudges, but the limited evidence that we have in this report suggests they could be incredibly effective at pulling a lot of customers back from non-scan behaviours.

A problem that responds to management

COLIN:

This is a really great report, Matt. It’s really well put together. As you say, we got very lucky. No one ever did a study as big as this one on self-checkouts, so it’s fairly authoritative.

I think it’s encouraging for the industry. What we’ve learned from this report is that this problem of loss at self-checkouts does respond to good management. It does respond to cross-functional engagement and cross-functional work, where all parts of the organisation, from IT to operations to human resources and people and technology, can come together to come up with interventions, down to that sort of, is it high risk, do we have different interventions or a different approach, a different design, different people for high-risk versus low-risk stores, urban versus rural.

I think this problem can be solved. I think the industry can deliver a shopping experience, certainly at the front end, that consumers seem to like and enjoy, with the minimum amount of disruption.

So, a really encouraging report, really well produced. Thank you very much.

And it’s free to download from the website. Thank you very much, Matt. Thank you.

Download the research

The Self-Checkout Loss Report 2026 is free to download. For the wider body of work on this topic, start at the self-checkout focus area.