Flooid thinks: People and process, as well as technology

Eric, can technology alone stop shrink at the self-checkout?

As a technologist, it’s easy to imagine a new ‘age of the machines’; a dystopian future where every in-store transaction is carried out by a self-checkout or a robot, and technology automatically acts to prevent theft in real-time. But the situation is a lot more nuanced, and technology must be considered in line with the human context. Today technology used for fraud detection informs, but humans need to act. Technology is only part of the overall solution.

Can you explain more?

There’s a lot of buzz around AI use cases in retail, particularly from commerce providers eager to highlight their cutting-edge solutions. Technologies like image recognition, facial recognition, and AI-driven pricing and promotions are often touted as game-changers. While these capabilities are undeniably valuable, discussions tend to focus on the technology itself, with little attention to how it integrates with people and processes in the fast-paced, complex retail environment. It makes me wonder if technology providers have fully considered the many customer-associate interaction scenarios that unfold during busy periods in-store.

Consider a frontline retail associate today. There is significant pressure on them. They might be responsible for overseeing anywhere from four, eight, ten or even twelve self-checkouts, frequently stepping in to help guide shoppers, verify age, or troubleshoot issues like printer jams or shopper confusion over which code applies to which product. And now – because of the rising occurrence of theft at self-checkout – associates are also asked to monitor for fraud. The ‘help’ side is already stressful, but adding the monitoring and handling of fraud and theft is a real challenge.

Now, technology can present these associates with a lot of information. It can alert an associate to suspicious activity, even in real-time. Image recognition might spot that a customer has scanned a packet of gum, but that an object like a packet of razor blades was passed on top of it at the self-checkout. An app like Flooid Empower can send an immediate message to a handheld device, to let the retail worker know there is potential theft happening at this self-checkout.

But when the associate sees that information, what do they do? Ask the customer to rescan every item to be sure? What happens if the technology is wrong or misinterpreted? What happens if the technology is right, but the customer is abusive, or even violent? In this context we can see that technology informs the situation, but it doesn’t necessarily resolve the whole problem and can even create bigger issues for the retailer.

Are there additional technological solutions to solve these issues?

Let’s say a retailer has a “do not approach” policy on suspected theft at self-checkout. An associate could use technology to capture and save CCTV footage, . The associate could press an alert button that calls over a security guard at the first sign of unusual activity at a SCO. Or perhaps even trigger a ‘blue screen of death’ to stop the self-checkout working during item scanning, meaning the suspected shoplifter needs to re-scan every item at a new checkout or manned register. But this is not enough alone. Again, the human context – and human intuition – may necessitate some shopfloor specialised training and decision-making, and that’s where technology is limited.

How can technologists keep the human dimension in mind when developing new products?

Firstly, understand everyone who may be impacted by the technology. In the example above, a technologist might think about the store associate who receives the alerts, but the technology has a wider impact; on the associate, on the security guard, and of course on the retailer’s customer or customers. 

Now consider the customer may not be a thief, but a good customer who has made a genuine mistake, perhaps is mortified to think they are suspected of theft. Many retailers are now adopting a mindset that no good customer should ever be put in a bad position. This raises important questions for technology design: How can interventions at self-checkout avoid alienating genuine customers? Could a solution involve initiating a seemingly random check or a prompt that encourages the associate to assist the customer rather than accuse them?

Ultimately, technology should be designed to support people—both customers and staff—by enabling better service and a more positive work environment. This also means that alongside the technology, policies and employee empowerment need to be in place to ensure the customer experience remains respectful and seamless.

Tell us more…

Technology, and AI in particular, is often based on iterations. What we do at Flooid is have weekly reviews where we have both the product managers and the engineers together, and we try to play through every scenario with every persona. We are constantly revisiting these scenarios. This is not Quality Assurance, as in testing, but rather true role-playing. We consider everything that happens in the ‘journey’; are the messages correct in the app, are we sending too many notifications, what happens if internet connectivity goes down in-store etc. Our Loss Prevention engine is rules-based, and we’re learning and re-iterating all the time, from day one of product development, to proofs of concept, to well after deployment, based on feedback and analytics from real-world engagements. We also constantly test the interactions between our POS engine, Empower App, and Loss Prevention engine on different messaging bursts in the cloud.

There is a lot of unpredictability with in-store transactions. Technology may provide absolutes in black and white, but real-life self-checkout transactions are an area with a lot of grey space. Technologists need to ensure a process whereby store associates can give feedback on the performance of the system, and potentially over-ride the decision of a system. Never forget the associate knows better than the technologist about what is truly happening during the transaction.

Can you tell us about false positives?

Technology can be wrong. AI must constantly learn and relearn and there are factors in the real-world that can confuse the digital world. There are many examples of false positives. AI image recognition is all based on learning. The more supervised learning data you have, the better, but things can still happen. Weigh scales, for example, can set off an alert about a mismatch on item recognition and expected weight. But in reality, it could be that the supplier of the bottle of milk changed their packaging in January, so the item now weighs 10g less, but the system has not been updated. How do you remedy this situation, without knowing the cause of the error alert?

How can you ensure technology simplifies tasks, rather than complicates them?

The key is to be close to our customers. I spend a lot of time looking at micro front-ends and architecture diagrams but I also love doing store visits. I’ll take an engineer to a store and just sit and observe.

For example, we may spot that the cashier on the mainline POS barely looks at the screen while scanning items. But without seeing the cashier in action, an engineer may not realise the way they use the screen and may move a button on the screen UI to a more aesthetically-pleasing, but less-intuitive place, interrupting the scanning flow and increasing the burden on the cashier and the duration of the transaction. It’s the kind of thing you must witness to understand.

The UX goes beyond the user, it goes to the mental context. You must think like Boeing when they design a cockpit, making sure there is no additional stress in the environment. It may not be a natural inclination to deduce that this type of thinking would take place in retail, but it is a reality, and it’s exciting to be part of that process.

 

If you have a problem with self-checkout shrink or customer/employee experience, speak to the Flooid team. Our composable, unified, cloud-native SaaS platform will function without friction, whatever your mix of endpoints and desired customer journey.