Exclusive interview: How Sensei is giving stores a brain

Sensei is an AI company founded in Portugal in 2017 by a team of scientists and engineers. Our vision is Conscious Spaces: physical spaces that can see, understand and act.

For grocery, that means giving the store a brain. Purpose-built capture on the shelf, in the aisle and at checkout feeds the Sensei Retail World Model, a live representation of every product, shelf, shopper and transaction. Store operations, supply chain and loss prevention teams act on it through workflows and tasks, so the data turns into decisions rather than another dashboard.

The clearest example is Continente Bom Dia in Leiria with Sonae, which you covered in your report. It is around 1,200 m² with 11,000+ SKUs and is the largest intelligent store in the world. It handles more than 3,000 baskets a day, and over 70% of transactions go through the autonomous checkout option. Behind that experience, the store runs on a live model of itself, giving the team live visibility of availability, shrink, waste and shopper behaviour across the whole store.

In grocery, where margins are thin and losses are quiet, the value reaches well beyond the checkout: shelf performance, shrink, waste, staff productivity, and shelf-level data that retailers and brands can work from together. Checkout is simply where most retailers start.

Your report puts it well: the hyper-connected store is the starting point, and the sentient store is the direction of travel. Our shift follows exactly that path.

Because the checkout was only ever the visible part. When you walk into one of our stores, the faster exit is what you notice. Everything that makes it possible stays invisible: knowing which product is on which shelf, who picked it up, what is missing and what staff are doing. Once you have built that to run a checkout, you already hold the most valuable asset in the store.

Three things are driving the shift. First, retailers told us they did not want another point tool for each problem. Second, AI has changed expectations: every business wants to become AI native, and in retail most of the business happens outside documents and networks, in stores with real customers and real replenishment. Language models cannot see that. A live model of the store can. Third, the inefficiency we uncovered was far bigger than we expected. Millions are left on the table every day.

In practice, a retailer invests once in a foundation and reuses it. The same model supports availability, shrink, staff productivity, checkout and, over time, retail media. The retailer owns that model, built on their own stores and their own data.

The model has to capture the full reality of the store, continuously: every product and shelf, every shopper journey, every staff action and every transaction, and above all the relationships between them. It is the store as a manager would see it if they could be everywhere at once.

We built our own capture for this, because following every individual item in real time demands it. Then we connect it with what the retailer already holds: planograms, product master data, stock and sales systems.

The hardest part is that every store is a living environment. Products move, promotions appear overnight, layouts change and every team develops its own way of working. The technology has to adapt to retail rather than forcing retail to adapt to it.

Visualising is the easy part. Acting on it takes three things. The data has to be live, so a gap on a shelf triggers a task before the shelf is empty. It has to be structured, so supply chain, commercial and loss prevention can all query the same truth through APIs and workflows. And it has to reach the people and systems that do the work: a store associate's task list, a replenishment order, an AI agent running a report.

Decisions that today run on samples, audits and sales data that arrive after the fact. A live model changes both the timing and the precision, and in grocery it pays off across these areas:

• Shelf performance: store teams get a replenishment task before a shelf empties, instead of discovering the gap on a walk round or through lost sales. Planogram compliance is measured continuously across the estate, and layout changes are judged on real shopper behaviour, not assumptions.

• Shrink: loss prevention sees losses as they happen, at item level, rather than after the stock count.

• Waste reduction: teams see what fresh and short-life stock is actually on the shelf and how fast it moves, so orders, rotation and markdowns follow real demand instead of estimates.

• Checkout and staffing: queues disappear and managers deploy people where the store actually needs them, hour by hour.

• In-store picking: click-and-collect pickers know exactly where each product is and whether it is on the shelf.

• Assortment and supply chain: buyers see what shoppers pick up, compare and put back, which sales data alone can never show, and forecasts run on real consumption signals from the shelf instead of delayed sales reports.

• CPG data: brands get shelf-level insight on availability, compliance and shopper interaction, shared on the retailer's terms, so both sides plan from the same picture.

• Retail media: brands can see how shoppers interact with a display or promotion at the moment of purchase, with online-grade measurement.

The bigger change is at headquarters. For the first time, the leadership team can see the whole store network as one measurable system, and run it with confidence.

The most important return so far is proof that this works at full supermarket scale, with real customers, every single day.

Continente Bom Dia in Leiria has been open since December 2024. It works like any regular supermarket: shoppers walk in, pick what they need, and their basket builds in real time, appearing on screen for a quick confirmation at checkout. It covers the full grocery mission, from fresh fruit and vegetables, an assisted fish counter and packaged meats to bakery and pastry, a coffee station and ready meals.

The return shows up in three places:

• Shoppers: the majority of transactions go through the Sensei checkout, and the store is used by shoppers of every age. That tells you the experience works for everyone, well beyond early adopters.

• Store teams: time moves away from the tills and towards the shop floor, with the system telling the team where attention is needed.

• The retailer: a live data foundation for the whole store, which Continente keeps building on.

Our approach is always the same: measure against the retailer's own baseline first, then prove the uplift before scaling.

Today retailers and suppliers often argue from different data. The supplier has panel data and syndicated samples. The retailer has sales figures. Neither sees what happens at the shelf in between.

A live model gives both sides one shared view of reality, owned by the retailer and shared on the retailer's terms. That changes the conversation in three ways.

• Availability becomes a joint metric. If a product is out of stock or in the wrong place, both sides see it the same day, and can fix the cause together rather than debate whose number is right.

• Promotions and launches are measured properly. Suppliers can see whether the display went up, how shoppers interacted with it and what it converted, instead of waiting weeks for sales data.

• The store becomes a media channel. Every other channel builds awareness. The store is where the shopper is ready to buy, standing in front of the brand and all its direct competitors. That is the most valuable moment in the funnel, and today it is barely measured. With a live model, retailers can start to offer suppliers in-store media with the precision of online.

The retailer keeps ownership of the data. That is what makes this a partnership rather than a data sale, and it is how suppliers become the real-time commercial partners your report calls for.

Your maturity journey captures it: from hyper-connected, to predictive, to sentient. Today's systems tell you what happened. The next generation will act on what is happening and anticipate what comes next. Learning speed becomes the advantage.

I expect three shifts. First, the model will stop describing one store and start running a network: every store connected, with lessons from one applied across the estate automatically. Second, AI agents will work directly on top of the model. A regional manager will ask why availability dropped in ten stores this week and get the answer, the cause and the tasks already assigned. Store teams will spend their time on customers rather than checking shelves. Third, in-store retail media will grow into a business as important as selling products in the store, because for the first time it can be measured at the moment of purchase.

The retailers who move first will pull ahead fast. Converting a physical estate takes time, so the gap between those who start now and those who wait will be very hard to close. My advice to any grocery CEO is simple: make your stores AI native as quickly as you can. We call that destination Conscious Spaces: stores that can see, understand and act.