{"id":82361,"date":"2024-12-17T09:49:13","date_gmt":"2024-12-17T08:49:13","guid":{"rendered":"https:\/\/visionarymarketing.com\/?p=82361"},"modified":"2024-12-18T13:11:48","modified_gmt":"2024-12-18T12:11:48","slug":"ai-in-retail","status":"publish","type":"post","link":"https:\/\/visionarymarketing.com\/en\/2024\/12\/17\/ai-in-retail\/","title":{"rendered":"AI in retail: shrinking queuing times today, headcount tomorrow"},"content":{"rendered":"

AI is redefining retail for good, bringing in the kind of automation and professionalism once implemented in the manufacturing industry. In this case, it\u2019s mostly revolving around data-driven marketing decisions and in-store retail media capabilities. As shown by Axians<\/a>, a VINCI group company, AI isn\u2019t a mere toy for undergraduate students who are failing their tests and need better inspiration. It\u2019s a robust, state of the art high-tech engine for growth and better in-store management. Yet, as often with technology, there are two sides of the same coin. The other one is more ominous, though, depicting a future of retail where layoffs will continue to rise, mostly for those retailers who missed that boat of AI-driven customisation. Here is the account of our discussion with Hugo Rocha Gon\u00e7alves<\/a>, Axians’ head of Smart RetAIl, at Tech for Retail 2024<\/a>.\u00a0<\/i><\/p>\n

AI in retail: shrinking queuing times today, headcount tomorrow<\/h2>\n
\"AI
Zooming in on AI in retail with Axians’s Hugo Gon\u00e7alves at Tech for Retail 2024<\/figcaption><\/figure>\n

You\u2019re in charge of the smart retail solution at Axians. What is it?<\/h3>\n

Hugo Gon\u00e7alves<\/b>. We developed the Smart retAIl concept to address the main challenges that the retail industry is facing today. There is a strong need to better understand in-store consumer behaviour, profile and shopping habits. We provide this knowledge to improve store efficiency, and to enable data-driven decision-making.<\/p>\n

Can you describe the process of Smart RetAIl?<\/h3>\n

H.G<\/b>. We are using AI and computer vision to accomplish this.\u00a0<\/span><\/p>\n

    \n
  1. The first step is to understand how the stores are organised<\/b>, what the shop floor looks like, and also how we can capture this data anonymously \u2014\u00a0for obvious GDPR compliance reasons \u2014 to fuel a data-driven decision process.\u00a0<\/span><\/li>\n
  2. After capturing this anonymised data through computer vision, there are a couple of things we need to understand. Such as footfall<\/strong>, who are the buyers, when they are buying, and their paths through the store<\/b>. We need to map, with the help of AI, the hot and cold zones within the store. Within these zones, we can understand if people are proper shoppers or if they are merely passers-by, and how much time they spend doing their purchases.<\/li>\n<\/ol>\n

    In a sense, this is some sort of heat map within the store<\/h3>\n

    H.G<\/b>. This is precisely what it is. And with this heat map, we can also understand what products people are looking at, how much time they spend. With AI we are taking this to a new level. This new level includes product tasting and testing. Two good examples are chocolate tasting, where we need to understand through computer vision when a customer is tasting something, which is very important in chocolate stores, and perfume stores. With this technology we can detect if the customer is testing the perfume and then understand if he or she will buy it or not afterwards.<\/p>\n

    \"AI
    AI in retail : Axians had set up a heat map showing how their system was monitoring footfall in front of their Tech for Retail booth<\/figcaption><\/figure>\n

    This means you are automating the work of market researchers who used to observe in-store consumer behaviour<\/h3>\n

    H.G<\/b>. Indeed. It used to be very tedious work to have someone watching hours and hours of video, trying to understand customer behaviour, customisation, and buying habits. Now we have AI that can process 24 hours of video, covering all the opening hours of a given store. We can process all this data and obtain valuable insights as well as data enriched by AI and computer vision.<\/p>\n

    So you are capturing a flow of images through in-store cameras, how is it working?<\/h3>\n

    H.G<\/b>. This entire process demonstrates the beauty of machine learning and AI. No need to resort to supplementary intrusive devices in the stores. We are using existing in-store CCTV cameras. We subsequently apply AI image processing, frame by frame, on the existing footage. The data is recognised and categorised by the AI automatically. The resulting data provides a lot of KPIs like passerby\/buyer qualification, hot and cold zones identification, as I said already. We\u2019re also interfacing with other information systems such as CRM, ERP, or point of sale systems. Doing so we are able to match our data with the sales data.<\/p>\n

    How do you adjust your setup for sales optimisation vs shoplifting prevention?<\/h3>\n

    H.G<\/b>. Indeed, the technology is also helping us in that direction. All the innovation and sophistication lie in the AI processing the image. With the evolution we\u2019ve experienced in computer vision, we no longer need specific hardware to do this. We simply need AI to help us with good machine learning and AI models to process it.<\/p>\n

    What kind of AI are we talking about here, certainly it\u2019s not ChatGPT!<\/h3>\n

    H.G<\/b>. This system has demanded a great deal of knowledge and experience. We have a large group of data scientists at Axians. It\u2019s also important to mention that this solution originated from an AI program launched by VINCI Group called the Leonard program (editor\u2019s note: named after Leonardo da Vinci). This program focuses on solving real challenges we face as citizens in our daily interactions. It\u2019s aimed at using AI to solve real challenges. One of these challenges involves using human expertise and knowledge in conjunction with AI. Her me we are talking of a different kind of AI (coupled with computer vision), not generative AI.\u00a0<\/span><\/p>\n

    Hence it\u2019s either machine learning or deep learning. What does the training process involve?<\/h3>\n

    H.G<\/b>. Typically, we have a learning curve for these types of systems. We train the model using what we call manual labelling<\/b>. Manual labelling helps the model understand what a person is. There are already modules that assist us. We don\u2019t need to start from scratch. We have existing models, open models that identify a human in a shop and their interactions. On top of this, we use not only our retail clients for assistance (they help us with the training of the model), but also to understand and label the data correctly. It\u2019s important to note that ours is not an unsupervised process. Here we are talking of supervised AI image processing. Supervised learning ensures the correct labelling of data and effective leverage of AI capabilities.<\/p>\n

    What\u2019s sort of work was involved prior to launch?\u00a0<\/span><\/h3>\n

    H.G<\/b>. Beforehand a lot of preparatory work was required. We have extensive experience developing AI solutions, especially in computer vision, data processing, AI processing, and data quality. This represents at least two or three years of intensive work, collaborating, testing and trying to understand how to move forward. Whenever the packaged solution doesn\u2019t suffice, we propose POCs to our clients. Such POCs help us reduce overhead related to testing. For example, we are currently testing queue times AI management. From experience, we\u2019ve found that normally when customers are buying something, they won\u2019t wait more than 10 minutes. If the wait exceeds 10 minutes, they\u2019ll leave the queue and give up on their purchase. We\u2019re addressing issues such as these by providing data driven insights.<\/p>\n

    Can you share a real-life business case with our reader?<\/h3>\n

    H.G<\/b>. We have launched a POC in Italy. We\u2019re assisting a large retail client over there. This retailer had realised it was losing sales and that their conversion rate was decreasing because their staff wasn\u2019t supporting their customers, even though that was part of their onboarding training.\u00a0<\/span><\/p>\n

    The end gain was significant. They\u2019ve reduced queue time by 50% and increased sales in some stores by 12 to 15% due to this implementation.<\/p><\/blockquote>\n

    It was sufficient to break even and they are now challenging us with new use cases, including some very complex AI problems.<\/p>\n

    How long does it take to break even with that kind of solution?<\/h3>\n

    H.G<\/b>. It depends greatly on the size of the stores. It\u2019s not a one-size-fits-all solution, but we can say that recovering the cost of the investment in the platform typically takes between 6 to 12 months.\u00a0<\/span><\/p>\n

    Any examples from Portugal?<\/h4>\n

    H.G<\/b>. Regarding queuing times, we have another example in Portugal involving high-tech retail solutions. The main issue was the identification of the most profitable areas within the store. When selling technology hardware like smartphones, etc., hot zones are of the utmost importance. They are areas\u00a0 <\/span>where consumers spend extra time, allowing retailers to sell media space to vendors. This what is known as in-store retail media. In this particular case in Portugal, we achieved great results with a retailer who started to monetise the hot zones in its stores. This wouldn\u2019t have been possible with our platform. Now they know which areas provide more return on investment and can charge more for product placement in these zones. We\u2019re still in the early stages with this client, a major retailer in Portugal. Already, the return in euros is between four and five figures per store.<\/p>\n

    Can your solution help struggling retailers in the current economic environment?<\/h3>\n

    H.G<\/b>. Absolutely. We\u2019re living in a data-driven world. Decisions should all be made based on data. This platform provides extensive in-store data and enables many well-informed data-driven decisions. In the near future, retailers failing\u00a0 <\/span>to consider data-driven marketing and AI will have to layoff staff and make other last minute haphazard decisions. Our solution helps uncover KPIs and metrics that were previously hidden. Through data-driven approaches, we\u2019re confident we can help reduce redundancies and facilitate better data-informed decisions.<\/p>\n

    What will retail look like in five or ten years from now? With all these AI solutions, will it still be a labour-intensive business?<\/h3>\n

    H.G<\/b>. It won\u2019t be. There will be a major reconfiguration of stores. Luxury stores will continue to have staff assisting us with purchases. For everyday retail purchases, there will be a significant reduction in staff.\u00a0<\/span><\/p>\n

    In the future, retail will no longer be a labour-intensive industry<\/p><\/blockquote>\n

    The future of retail will also be about extensive customisation. We\u2019re already experiencing this level of customisation in streaming services that trace our personal and behavioural data very well. This means that each consumer will have its own bespoke catalogue tailored to his or her needs. To stay in business, retailers must possess in-depth knowledge of their customers. Moving forward, beyond this extensive level of customisation, a personalised care experience for each customer.\u00a0<\/span><\/p>\n