Podcast (english): Play in new window | Download (Duration: 11:50 — 16.4MB)
Subscribe: Apple Podcasts | Spotify | Android | Blubrry | Email | RSS
Our reporters attended the Paris Retail Week 2024 event, a trade show of which we are media partners, to take stock of fraud and the role of AI. We collected a lot of valuable feedback on threats (both in-store and e-commerce) and the countermeasures proposed by artificial intelligence. To do so, we interviewed Gilles Bijaoui, head of CX at Fujitsu (Customer Experience being the name of the retail division chosen by the Japanese company). During this discussion, which took place at the opening of the show, he gave us a wealth of information on fraud and described the AI solutions designed to combat it. Such solutions have now been deployed for almost two years in retail chains of all sizes.
When AI fights fraud in-store and on websites
Isn’t AI-driven fraud reduction deja vu?
Gilles Bijaoui. Indeed, this has been a recurring topic over the last five years, whether at NRF or Paris Retail Week. But it wasn’t really implemented in the field. But over the last two years or so, that has completely changed. Post Covid, we find these solutions in production in many retailers of all sizes.
There are several types of AI involved, including generative AI. This has been rolled out by several large retail chains and the feedback is fantastic. Be it regarding performance, reducing employee theft and shoplifting, whether for physical retail or e-commerce.
What kind of fraud are we talking about?
GB. There are two types of fraud in retail and e-commerce:
- Firstly, in-store fraud, just over half of which is linked to theft in the shop and around 40% is due to shop assistants themselves.
- When it comes to e-commerce, it’s more likely to be linked to electronic flows of information, either payment fraud or identity theft (or phishing). It’s more complex and also more damaging.
Fraud on e-commerce platforms accounts for almost 20% of e-commerce sales worldwide. That’s a huge amount! If you compare this figure with in-store theft or fraud, it’s down to only 2-3%.
We’re definitely on a different scale with e-commerce. It’s also linked to the volume of transactions, which soared during and after the Covid crisis.
But the good news is that, thanks to AI technologies, we’ve seen a reduction in the growth of these fraudulent activities over the last two years or so. Previously, electronic fraud was growing at a rate of around 20% a year, but thanks to AI we’re now down to less than 15% per annum. And it’s getting even better with the education of individuals and businesses.
Are there any regional variations regarding fraud in retail?
GB. Not in Europe, all countries are pretty much the same. Only one country is an exception, and that’s the UK, which has even more fraud, as in the US, whether it’s physical or e-commerce.
That’s a difference of around 10 points over France. What’s more, the figures are reversed. Physical retail theft in the United States and Great Britain is higher than in the rest of Europe. And conversely, e-commerce is better controlled there than on the continent. Most likely because they decided to deploy these solutions before other countries.
On the other hand, given the massive in-store fraud in the United States, many everyday consumer products are under lock and key.
Let’s now talk about combatting fraud with artificial intelligence
GB. For physical stores, there have been two eras, and two different systems.
Previously, artificial intelligence was based on data. The more data we had, the more artificial intelligence systems learned, and were able to identify recurring patterns. This has changed to the extent that now, we’re relying more on the definition and identification of behaviours and movements.
Some of these modern solutions have recorded hundreds of human behaviours that can help determine a person’s intentions. Based on what we call “patterns”, we are able to tell, with the in-store camera system and artificial intelligence, whether a person intends to buy, steal, or just potter around. This is a starting point for AI to trigger personalised in-store promotions.
Personalisation With AI
GB. For example, if you’re standing in front of the aisle dedicated to formula milk in a supermarket and you’re hesitating, artificial intelligence will advise you based on the information available. It can also provide you with recipes based on the food you are buying or suggest that you virtually try the garments you are about to purchaseThe B2B purchasing process is the result of a long life cycle often linked to a contract as there are many people to convince..
We can also offer promotions tailored to the person based on their loyalty cards, and instantly offer a coupon on an item a consumer is looking at.
These marketing and merchandising efforts are no longer restricted to online commerce. Now, they are also available at the point of sale. And all this, thanks to artificial intelligence.
How can one avoid hallucinations on such recommendations?
GB. We have to fight this idea that humans will be disappearing because of AI. These systems are not autonomous. They need to be controlled. Certain words must be banned, for example. The way we address people has to be calibrated so as not to offend or be too intrusive. All this has to be strictly supervised.
Marketing messages must be constantly calibrated, both by us and our customers.
Let’s take an example. We’re currently working on fitting rooms. A lot of fraud is happening there, both by consumers and shop assistants. And RFID doesn’t really help fight this issue.
We have therefore combined several solutions so that we can identify that if a person comes in wearing red and goes out wearing black, something is clearly wrong. Similarly, if that person goes in wearing a size 12 and comes out with size 24, something isn’t quite right either. Behaviours are also analysed, as I pointed out earlier. At the end of the day, though it’s always the customer who decides where to draw the line. There are also laws governing the use of these technologies. For example, one isn’t allowed to recognise faces in continental Europe.
Is there a good business case you implemented with a retail chain?
GB. We’ve worked with a well-known German hard discount chain for which we have deployed anti-theft solutions at the checkout. The solution makes it possible to identify fruit and vegetables in a relevant way. If you take a bottle of wine and you change the label or the barcode is not the right one, the system is able to detect it.
With this solution, we have succeeded in reducing thefts by 60%. Into the bargain, improving in-store communication, both towards shop assistants and customers, also helps to reduce thefts by around 20%.
How does it work in practice?
GB. For automatic checkouts, a camera is placed on top, coupled with sensors that can measure not only the weight, but the typology of the product, its firmness, size, colour and even its density. The progress made with cameras is considerable. We’re able to spot the right product with around 90% success rate.
And it’s the same at the physical checkout, in other words, it also prevents certain employees – let me remind you that this accounts for 40% of fraud – from passing the wrong products, or forgetting to pass a product. If this happens, an alarm rings.
And once a certain number of alarms have been triggered, human intervention is required. We also know that most checkout fraud occurs in the last two hours before closing time. The pace picks up enormously at that point, because there are longer queues and dishonest employees figure it’s going to be easier to get away with theft.
Yet it is precisely where controls will be tightened, alarms will be more frequent, there will be greater vigilance over data reconciliation and a speeding up of the process. What is very important to bear in mind at all times is that all these systems are controlled and validated by human beings.
In the event of a mistake or problem, human intervention will either trigger apologies or reverse an incorrect identification. Artificial intelligence is an ally of commerce, but it is not inhuman if it is properly implemented.
What about e-commerce?
GB. On the e-commerce systems that we deploy, we observe that fraud often comes from the same IP addresses and the same geographical areas. Fraud is often linked to an initial failure to enter a credit card number or make a payment. This triggers alerts accordingly.
In response, we will either block the payment or notify the end customer’s bank. The computing power is such that fraudsters can hardly get away with it. And banks have also made efforts in this direction, with 3D secure and dual authentication of purchases via mobile applications, email or text messages.
All online businesses, even the smallest merchant sites, are equipped with these solutions, which help to prevent attacks, the theft of customer files and phishing.
Could fraud ever disappear with AI?
GB. We’re not that far off. Especially as we develop new technologies, including fingerprinting, retinal scanning and even palm scanning. Probably a lot of e-commerce sites or personal computers will soon be equipped with them.
In partnership with Ingenico, we have developed a system that recognises only the palm of the hand. The venous system is unique to each individual. And you can’t reproduce it with a photograph. Add to this heat measurement and you have an ultra-secure system that will further reduce fraud.
About Fujitsu
Fujitsu is a company that is both well-known and little known. It is a large Japanese technology company that started out over 100 years ago with telecommunications and switched to services in the 2000s. It employs just over 130,000 people worldwide, and has a turnover of 33 billion dollars. Its retail division (aka Customer Experience) accounts for just over 20% of the Japanese company’s worldwide sales. Its activities are now focused on services and platforms for points of sale and e-commerce, not forgetting integration services, consulting and infrastructure for retailers. Fujitsu is even the 8th largest technology services company in the world.
————————————————————
Yann Gourvennec a créé visionarymarketing.com en 1996. Il est conférencier et auteur de 6 livres. En 2014, il est passé d'intrapreneur à entrepreneur en créant son agence de marketing numérique.
- AGI (General Artificial Intelligence), Myth or Reality? - 08/10/2024
- What if your smartphone were the future of GenAI? - 27/09/2024
- Brand Talk Engages Only a Third of Consumers - 27/09/2024
