Demand forecasting, computer-vision checkouts, algorithm-driven pricing, agents that advise customers and handle their returns: retail has become one of AI's first large-scale proving grounds. McKinsey puts the potential at €240 to €320 billion for European retail over the next five years, 91% of retailers already use or assess AI, and the 2025 holiday season showed the scale of the shift: $262 billion of online sales influenced by AI and agents. Here is what already works, what it returns, and the framework to respect, seen from Belgium.
Adoption is massive. NVIDIA's third annual State of AI in Retail and CPG survey, published on 7 January 2026, shows that 91% of respondents are actively using or assessing AI, that 90% will increase their AI budgets in 2026, and that the effects are already on the books: 89% credit AI with increasing annual revenue and 95% with decreasing costs. Agentic AI is making its entrance: 47% of companies are using or assessing it, including 20% with agents already active and 21% planning deployment within the year.
On the customer side, the shift was visible during the 2025 holidays. Salesforce, which aggregates the activity of more than 1.5 billion shoppers, measured $1.29 trillion in global online sales between 1 November and 31 December 2025, of which $262 billion was influenced by AI and agents, or 20% of sales. Visitors referred by AI search engines (ChatGPT, Perplexity) convert nine times better than those coming from social media, and brands operating their own agents grew 6.2% versus 3.9% for the rest.
For Europe, McKinsey and EuroCommerce published on 10 June 2026 the report Rewiring retail in Europe: The AI imperative: an end-to-end AI transformation could unlock €240 to €320 billion in economic value over five years for the sector, with EBITDA improvement of 4 to 6 percentage points in grocery, 6 to 8 in hardline and 8 to 10 in fashion and beauty (softline). The same report names the real obstacle: fewer than 15% of prioritised use cases actually scale, and only 15% of AI investment goes to the commercial domain, where the potential is largest.
Belgium is no bystander. Colruyt is the country's first retailer to offer a self-scanning smart trolley: the Smart Cart, developed in-house around computer vision, launched for customers in September 2025 in Halle and extended to Kessel-Lo (March 2026) and Waterloo. It already accounts for 10% of shopping trips in Halle, and 8 out of 10 weekly users are repeat users. In parallel, the chain is rolling out its Easy Check-out system to all 273 Colruyt stores. The signal is clear: store AI is no longer a trade-show gadget, it is in Belgian aisles.
Behind the numbers, the uses that create value are concrete and often invisible to the customer. They share one trait: a high volume of repetitive decisions (ordering, pricing, restocking, answering) where every point of precision gained multiplies across thousands of SKUs and receipts.
The fastest gains sit where the data already exists: sales history, inventory, receipts, loyalty programmes. Demand forecasting reduces both stockouts (lost revenue) and unsold goods (destroyed margin, food waste); assisted customer service absorbs peaks without degrading response times; product content generation saves weeks on catalogue enrichment. NVIDIA notes that 54% of companies cite employee productivity as the first observed benefit, ahead of operational efficiency (52%) and customer service (41%).
The condition is organisational. McKinsey observes that the retailers who capture value are not those who multiply pilots but those who industrialise a few high-impact cases: more than a quarter of executives prioritise more than 50 use cases, while fewer than 15% have actually been scaled. The first barrier cited (24%) is change capacity: training, communication, process redesign. In other words, technology is rarely the limiting factor; execution is.
Start with a measurable case on existing data: forecasting on one or two product families, an internal assistant for teams, product sheet generation. Measure before and after (stockout rate, shrinkage, processing time, conversion), frame customer data, then extend store by store. Sensitive projects (personalised pricing, smart video) come second, with a validated legal framework.
Sales per SKU and per store, inventory, receipts, loyalty, e-commerce: inventory what exists, its quality and its freshness. This is the foundation of any retail project.
A limited perimeter (one product family, one channel), before/after indicators and a defined duration. Demand forecasting and customer service are the safest entry points.
GDPR for loyalty and personalisation, chatbot transparency, documented pricing policy, impact assessment for video or profiling. The AI Act and GDPR are handled from the pilot, not after.
Extend what proved its value, store by store, train teams on the floor and at headquarters, and track indicators over time: a forecasting model is monitored like inventory.
Retail touches the daily life of millions of consumers, and the framework applies today. The practices prohibited by the EU AI Act have applied since 2 February 2025: emotion recognition in the workplace (staff monitoring), biometric categorisation based on sensitive data, manipulative techniques exploiting vulnerabilities. Chatbots and sales agents must clearly tell the customer they are interacting with an AI. Most common retail uses (forecasting, pricing, recommendation) remain classified as limited or minimal risk: the bulk of compliance work therefore sits on the GDPR side.
The GDPR frames personalisation: legal basis for loyalty card profiling, minimisation, clear information, and particular vigilance on in-store video (purpose, proportionality, retention). For prices, discount announcements remain subject to the rules of the Belgian Code of Economic Law derived from the Omnibus Directive (reference price), whether pricing is algorithmic or not. Trust is earned through transparency: say what is automated, why, and keep a path to a human.
With a measurable case built on data already available: demand forecasting on one product family, product sheet generation, or a customer service assistant for recurring questions (delivery, returns, availability). Measure before and after (stockouts, shrinkage, response time, conversion), frame it, then extend. The big projects (dynamic pricing, fine-grained personalisation) come next, once data and governance follow.
Yes to propose, no to decide alone without a frame. Pricing algorithms optimise prices, promotions and markdowns, often coupled with electronic shelf labels. The frame remains the law: display of the reference price for discount announcements, non-discrimination, clear consumer information, and GDPR as soon as a personalised price relies on a profile. A pricing policy validated by a human and logged is the good practice.
Three concrete points: prohibited practices (including emotion recognition in the workplace and sensitive biometric categorisation) have applied since 2 February 2025; chatbots and agents must clearly tell the customer they are interacting with an AI; and smart in-store video must stay within the GDPR framework (purpose, proportionality, information). Most common retail uses (forecasting, pricing, recommendation) remain limited risk.
McKinsey puts the EBITDA improvement at 4 to 6 percentage points in grocery, 6 to 8 in hardline and 8 to 10 in fashion and beauty (softline) for an end-to-end transformation. Quick gains come from forecasting (fewer stockouts and less waste), assisted customer service and team productivity. The condition: moving from pilot to scale, which fewer than 15% of prioritised use cases achieve today.
Molderez Consult helps Belgian retailers, e-commerce companies and trade SMEs map their data, launch measured pilots (forecasting, customer service, product content), frame GDPR and the AI Act, and industrialise what pays off, from the first dashboard to the augmented store.
Discuss my project