Shopping list: the bottleneck no one had seen
On average, an Italian shopper takes forty minutes to fill an online basket. Fifty-eight per cent start again from a previous order, not because they trust their usual choices, but because rebuilding the list from scratch is unbearable. Since the start of the pandemic, the number of digital consumers has tripled, accompanied by a surge in online purchases of products that we previously bought only in shops: detergents, frozen foods and fresh produce.
Yet the experience of building a shopping basket has remained unchanged for fifteen years: a text search across an extensive catalogue. We have transformed everything—payment, delivery and last-mile logistics—but have left the most important moment untouched: when the customer decides what they want. This is where loyalty is won or lost. If the app does not understand a request for ‘the usual coffee’, the customer goes back to the printed leaflet.
Three layers of chaos: data, behaviour, intent
Behind an apparently simple tool such as a list lie three overlapping layers of complexity.
The first is data. A product is never just a product: it is an EAN code, a product description, three photographs taken from different angles, nutritional information, a list of allergens, availability that varies by store, and a price that changes with the week's promotion. All this information is fragmented across the ERP, PIM, e-commerce system and printed catalogue, and is rarely consistent.
The second layer is behaviour. The same user adds products to the list from five different contexts over the course of the same day: a recipe found on a blog, a Google ad, a voice command in the car, the list shared with their grandmother, and an item dictated to their smartwatch while running.
The third, and most subtle, layer is intent. When I write ‘biscuits’, I might mean Gran Cereale because that is what we usually buy, Equilibrio because it is on offer, or Gocciole because that is what my daughter likes. No traditional interface can resolve the ambiguity. It is a language problem: the supermarket speaks in EAN codes, while the family talks about ‘the usual coffee’. A translator is needed.
The true cost of misalignment
The impact of not having resolved this issue can be measured on several levels. The most immediate is the abandonment rate: whenever the matching between what the user enters and what the system suggests fails, part of the basket is lost.
Then there is loyalty: customers who do not feel understood look elsewhere, and today elsewhere means global marketplaces that have turned grocery shopping into a matter of logistics rather than relationships.
The most overlooked area, however, is advertising. The shopping list represents the point of highest purchase intent in retail: users are literally stating what they intend to buy. But if we cannot interpret it, we cannot monetise it without becoming intrusive. And if we are intrusive, we further undermine the experience.
The real cost is not a figure on the balance sheet. It is the structural misalignment between how people think and how systems respond. That misalignment results in eroded margins.
Elista: the right infrastructure for a brain that did not yet exist
Elista was an initial attempt to address the problem systematically, by rethinking the shopping list as a strategic asset rather than creating yet another app.
The idea was to treat it not as a static checklist, but as a living object, available at every user touchpoint: their phone, smartwatch, CarPlay, lock-screen notifications and even QR codes on advertising billboards.
What we built with Elista comprised two innovations that were new to the Italian grocery sector in 2021. The first was enhanced family sharing: a household list where Dad adds coffee, Grandma adds ice cream and the daughter adds biscuits, with everyone able to see in real time who added each item. This was more than a niche feature: the household is the grocery sector's actual purchasing unit, and no one had yet treated it as such.
The second was suggested and sponsored lists: native advertising formats built around everyday situations, such as ‘Single? No problem’, ‘Make-your-own pizza’ and ‘Gnocchi with speck and fondue’. This model allowed brands to become part of the shopping list without being imposed on users, enriching the experience rather than interrupting it and bringing an increasing amount of editorial content into the shopping list.
The project had one significant limitation: generative AI was not yet ready. The matching between ‘biscuits’ and the right product was rule-based, and still struggled with synonyms, brand variants and degrees of preference. We had built the right infrastructure for a brain that did not yet exist.
Three acts. We are only halfway through the second.
To understand where we are heading, I use a narrative structure: three acts, three progressive transformations.
Act One: the shopping list becomes a playlist
Just as Spotify moved from being a repository of songs to curating contexts such as ‘relaxed mood’, ‘morning workout’ and ‘romantic dinner’, the shopping list moves from being a list of items to a collection curated around real-life situations: ‘Dinner with vegan friends’, ‘Sunday picnic’, ‘A week on a diet’. The product becomes less central, giving precedence to the human need it serves.
Act Two: the shopping basket that knows our habits
What seemed out of reach in 2021—understanding ‘crunchy biscuits like Grandma’s, but without palm oil’ and returning the right product on the right shelf—can now be achieved with just a few lines of code using generative AI.
The assistant remembers your preferences, allergies, purchase history and budget. It knows you. And when you say “biscuits”, it knows you mean the wholemeal ones you always buy, but it also knows that you bought them last week and that this time you might want to try a similar alternative that is on offer.
Act three: the assistant becomes a purchasing agent
This is the step the market will take over the next 24 months. The assistant will no longer make suggestions: it will fill the basket for you. You will simply approve it. Consumers will be happy to relinquish this control, provided that trust has been built during the previous stages.
Those who have not earned that trust will be excluded from the purchase.
Agora: building the shopping assistant of the future today
Agora stems directly from this understanding: demonstrating that ‘Act Three’ is no longer just a vision; it can already be built. What required a team of thirty people and a data lake, in 2021 can now be built in just a few months, with the right tools.
The result is an app that does exactly what we are discussing. Under the bonnet are architectural choices that I believe will define the future of these solutions. The first is to avoid being tied to a single AI model: the platform must be able to communicate with different providers and select the one best suited to each task. Categorisation is assigned to a cost-effective model, complex matching to the most powerful one, and suggestions to the fastest. No single model is the best at everything.
This is followed by a three-stage matching pipeline that linguistically deconstructs the query, compares it with a semantically enriched catalogue and returns a genuine confidence score, rather than the misleading ‘98% match’ we were seeing everywhere. Finally, the list is exported transparently to e-commerce platforms: the user creates it once, and the platform converts it into shopping baskets for the different retailers.
The real lesson from Agora is not a technical one. Technology is no longer the barrier: the challenge lies in clean data and the organisational commitment to treating the list as a strategic product rather than an app feature.
Whoever controls the list controls the customer
The shopping list is the new battleground for customer loyalty. Whoever controls the list controls the basket. Whoever controls the basket controls the customer. Whoever controls the customer controls the margin.
Retailers that do not turn their shopping list into an intelligent assistant over the next twenty-four months will relinquish that control to someone else. It may be Amazon with Alexa, OpenAI with a ChatGPT that orders directly from marketplaces, or a new player we do not yet know. But someone will take that place. At that point, the traditional retailer will revert to what it was in the 1990s: a provider of shelf space rather than a point of contact for the customer.
The technology is available. The data is available. Only two things are needed: strategic clarity and the courage to invest in an invisible asset. To a CFO, the shopping list might seem like a minor detail. It is not. It is the nervous system of retail for the next ten years.
