AI in retail and ecommerce: 5 ways to sell more

In short: artificial intelligence lifts retail sales in five places: personalised recommendations, demand forecasting, pricing and promotions, customer service, and joining up store, website and B2B channel. On top of these come generative AI for product content and, in physical stores, shelf, checkout and layout monitoring. It only works when product, order and customer data are clean and connected. Start from one measurable use case, not from buying a platform.

What is AI in retail, and how widely is it used?

AI in retail means models that read sales, catalogue and customer behaviour data and turn it into a forecast, a recommendation or a piece of text: which product to show, how much to reorder, what price to charge, how to answer a question. It is not a single piece of software. It is a layer that sits on top of your ecommerce platform, ERP, CRM and point of sale.

According to Eurostat, 19.95% of EU enterprises with at least 10 employees used at least one AI technology in 2025. In retail trade, 48.18% of the companies using AI apply it to marketing or sales. Italy shows how fast things are moving: Istat reports that AI use doubled in a year, from 8.2% of companies in 2024 to 16.4% in 2025, with marketing and sales as the most common area of use (33.1% of adopters).

The main brake is not technology. Among EU companies that considered AI but did not adopt it, 70.89% cite a lack of relevant expertise. Retailers that get their data and processes in order therefore gain an edge over competitors who have access to exactly the same tools.

The AI levers that drive retail sales

The table sums up the five main levers plus two supporting areas (catalogue content and physical stores): where AI affects sales, what data it needs and which indicators tell you whether it is working.

Lever	What it does	Data required	KPIs to track
Personalisation	Suggests products, content and offers based on behaviour	Order history, browsing data, catalogue with attributes	Average order value, conversion rate, revenue from recommendations
Demand forecasting	Estimates sales by product, store and period	Sales history, promotions, seasonality, stock levels	Stockouts, average inventory, forecast error
Pricing and promotions	Proposes prices and discounts based on demand, margin and competition	Price history, costs, competitor prices	Gross margin, sell-through rate, share of sales on promotion
Customer service	Answers questions on products, orders and returns, and hands complex cases to people	Product pages, terms of sale, order status	Cases resolved without an agent, response time, conversion after chat
Omnichannel	Aligns customer, stock and prices across store, site, app and B2B portal	Single customer record, real-time stock	Click-and-collect orders, cross-channel sales, returns
Catalogue content	Extracts attributes, writes and translates product pages, improves on-site search	Manufacturer descriptions, technical data from the ERP	Complete product pages, product page conversion, returns for items not as described
Physical store	Flags empty shelves, analyses shopper flows, speeds up checkout	Video and sensors, sales by department, stock levels	Empty shelves, sales per square metre, inventory shrink

How does AI personalisation increase average order value?

A recommendation engine compares what a shopper views and buys with the behaviour of similar shoppers and with product attributes. The output is the "frequently bought together" block, emails with genuinely relevant products, and on-site search that understands "non-slip outdoor tiles" even when no product page contains that exact phrase.

Ecommerce example: when a shopper adds a basin to the cart, an online bathroom store proposes the matching trap, waste and mixer tap. B2B example: an electrical wholesaler shows an installer the spare parts they ordered in recent months, plus compatible products, at the prices of their own price list.

The bottleneck is usually the catalogue, not the algorithm. If product pages lack consistent attributes (sizes, materials, compatibility), the model recommends almost at random. Fixing product data also helps you show up in AI-powered search engines, which reward complete, structured information.

Mind privacy: using purchase history for personalised emails and offers is profiling for direct marketing. Under Article 21 of the GDPR, customers can object at any time, and that right must be brought to their attention explicitly and separately, at the latest at the first communication. In B2B the rule applies to your contacts' data, since they are individuals.

Demand forecasting: fewer stockouts, less dead stock

Forecasting models estimate demand by product, store and week using sales history, seasonality, planned promotions and the events calendar. The advantage over a spreadsheet is the ability to handle thousands of SKUs at once and refresh the estimates every day.

Sales grow for a simple reason: the product is there when the customer looks for it. At the same time, cash tied up in slow-moving items is released. In B2B the same logic applies to spare parts and consumables: predicting repeat orders from regular customers lets you propose them before they call a competitor.

The minimum requirement is a clean sales history, ideally covering at least two full seasons, with promotions and stockouts flagged. Without that, the model mistakes a week with no stock for a week with no demand.

Pricing and promotions: what can you safely automate?

AI-assisted pricing proposes prices based on demand, stock, margin and competitor prices. It earns its keep on large catalogues, where nobody can check thousands of prices by hand. The practical rule: the algorithm proposes within margin limits set by the business, and a person approves the exceptions.

If you sell to consumers in the EU, two rules from the Omnibus Directive matter. First, discounts: any price reduction announcement must show as the prior price the lowest price applied in the previous 30 days (in Italy, article 17-bis of the Consumer Code, as clarified by the Ministry of Enterprises and Made in Italy for both online and in-store channels). A system that changes prices often has to keep track of that value. Second, personalisation: if a price has been personalised through automated decision-making, the customer must be told. Italy transposed both rules with Legislative Decree 26/2023.

In B2B, AI pricing works mostly on price lists and customer-specific discounts: it flags orders sold below margin, suggests a discount in line with the customer's history and helps sales reps put quotes together faster.

Customer service: when does an AI assistant actually sell more?

A generative AI assistant answers questions on availability, sizes, compatibility, shipping and returns, reads order status and hands complex cases to a person. It sells more when it removes the doubts that block a purchase, not when it simply repeats the FAQ page.

Three conditions apply. Answers must come only from verified company sources (product pages, terms of sale, order status). Handover to a human must be clear and fast. And you need transparency: under the EU AI Act transparency rules, applicable since August 2026, people using a chatbot must be made aware that they are interacting with a machine. In B2B the same assistant can live in the customer portal, covering availability, technical documents, reorders and delivery status without going through a sales rep every time.

Generative AI: product pages, search and content

On the catalogue, generative AI is often the fastest project to launch. It reads long manufacturer descriptions, extracts the attributes that matter (sizes, materials, compatibility) and writes copy focused on what shoppers weigh up before buying. In its overview of AI in retail, Oracle also lists on-site search that understands natural-language queries, visual search (the shopper shows a photo and finds similar items) and tools that help shoppers pick the right size, which cuts returns.

Ecommerce example: a homeware store generates consistent titles, descriptions and attributes for thousands of SKUs from supplier sheets, and a team member spot-checks them before publishing. B2B example: an exporting distributor turns technical data sheets into sales copy and translates it for each target market; for choosing the tools, see the guide to AI translation software.

Two rules prevent the costliest mistakes. Technical data (sizes, certifications, compatibility) must come from the manufacturer or the ERP and never be "filled in" by the model. And a human review is needed before publishing, at least for the categories with the most returns or complaints. How to roll out generative AI without losing control is covered in the article on generative AI as a business strategy.

Omnichannel: store, website and B2B channel on the same data

Customers do not think in channels. They search online, check availability at the nearest store and then order by phone through a sales rep. AI helps connect these touchpoints: it recognises the same customer across channels, shows real store stock and suggests what a store assistant or sales rep should offer based on history.

Without integration, though, each channel has its own version of the truth: different prices, stale stock levels, incomplete history. That is why the first job is connecting ecommerce, ERP and CRM through automated flows, as covered in the guide to business process automation. For a B2B example, see the Euroclone omnichannel customer experience case.

How is AI used in physical stores?

In the store, AI works with cameras, shelf sensors and devices carried by staff. The most common uses, as described in the same Oracle overview, are:

Shelves and replenishment: counting items on shelves and flagging gaps, including with robots, and spreading restocking through the day around peak footfall.
Layout and merchandising: analysing shopper movements reveals products nobody approaches, so they can be replaced or moved; choices are compared across stores with A/B tests.
Checkout: video and sensor systems that let shoppers pick up products and walk out without queuing, or that recognise items with unreadable barcodes.
Inventory shrink: detecting when a cheaper item is scanned than the one taken, or when a cashier deliberately undercharges a customer.
Assisted staff: repetitive tasks move to machines and store associates go back to helping customers, with a tablet showing availability and complementary products to suggest.
New store locations: simulations that combine competitors, population density, rents and demographic data to decide where to open.

B2B example: in a showroom or trade counter, the salesperson sees on a tablet the customer company's order history, its price list and warehouse availability, and suggests the accessories that usually get left off the order.

Before installing "smart" cameras, a few checks are needed; in Italy the data protection authority (Garante) summarises them in its video surveillance FAQ, in line with EDPB guidance. Shoppers must be informed before entering the filmed area, for example with a simplified sign that points to the full privacy notice. Systems that analyse images to detect unusual behaviour automatically may require a prior data protection impact assessment, which is always mandatory for large-scale systematic monitoring of publicly accessible areas. If the cameras also film employees, Italian labour law safeguards apply (article 4 of Law 300/1970, the Workers' Statute). For footfall analysis, use aggregated data that does not identify individuals.

AI agents that shop: what changes for ecommerce?

The next step is agentic commerce: AI assistants that search, compare and complete the purchase on the customer's behalf, inside a conversation. In January 2026 Google introduced the Universal Commerce Protocol (UCP), an open-source standard developed with Shopify, Etsy, Wayfair, Target and Walmart. It connects interfaces such as AI Mode in Google Search and the Gemini app to merchants' systems, from product discovery through checkout to order management. The merchant remains the merchant of record and keeps its own business logic.

For online sellers there are three practical consequences:

A machine-readable catalogue. An agent compares data, not promotional photos: complete, consistent attributes become a sales requirement.
Real-time prices and availability. UCP supports real-time inventory checks and dynamic pricing within the conversation, so stock and price lists must be exposed via API and kept in sync with the ERP.
Clear, structured terms. Shipping, returns and warranties must be written so an agent can relay them to the customer without ambiguity.

There is no need to chase every announcement, but well-ordered product data pays off anyway: for SEO, for recommendations and for agents. For how agents are changing business software, see the article on AI agents and SaaS.

Real case: Iperceramica, AI across stores and ecommerce

The case. Iperceramica, a retail chain based in Fiorano Modenese specialising in floor and wall tiles and bathroom furniture, has grown to more than 100 stores in Italy and other European countries. In September 2026 a collaboration was announced with Fattoretto AI, the business unit of Fattoretto Agency dedicated to training, process automation and custom AI tools. The project focuses on selected activities linked to the company's digital presence and to making better use of its information assets, with humans keeping control over decisions and the quality of results (source: Focus E-commerce).

On its website, Iperceramica runs a generative AI assistant built by Responsa and available around the clock: it guides shoppers through product choice, purchase and after-sales, shows order status and points to the nearest store. The vendor's case study does not publish numerical results.

The useful lesson for any retailer: the AI customers see, the chat assistant, only works if there is work on data, content and processes behind it. With a technical catalogue like ceramics (formats, finishes, installation), the quality of product information matters more than the interface.

Which AI project should you start with?

Your first project should fix a problem that already costs money and can be measured within a few weeks.

Situation	Recommended first project	Why
Many repetitive customer service requests	AI assistant on FAQs, shipping and order status	Visible effect on response times and team workload within weeks
Frequent stockouts or heavy inventory	Demand forecasting on best-selling products	Protects sales and frees up cash
Large catalogue with incomplete product pages	AI-assisted enrichment of product pages with human review	Foundation for on-site search, recommendations and SEO
B2B customers with recurring orders	Reorder suggestions in the portal or by email	Grows revenue from existing customers
Margins under pressure	Assisted pricing with margin rules	Control over thousands of prices without manual work

B2B ecommerce has its own logic, with customer-specific price lists, sales reps and recurring orders: I cover it in the article on the future of B2B ecommerce.

How do you choose AI tools and vendors for retail?

Before signing, check these points on every proposal:

Integration: does it connect via API to the ecommerce platform, ERP and CRM you already use? Ask for a trial on your own data, not just a demo.
Control: can you restrict the assistant's sources, set price and margin thresholds and see why the system made a given suggestion?
Data and privacy: where the data is processed, whether the vendor uses it to train its models, and which contract appoints it as a processor, as Article 28 of the GDPR requires.
Real costs: flat fee, usage-based or a percentage of sales. Estimate the bill on your actual order and conversation volumes.
Exit: if you switch vendor, can you export data, configurations and conversation history?
Ongoing management: who updates rules, content and integrations when the catalogue and price lists change?

The usual order of preference: first the AI features of the platform you already have, then a specialised subscription service, and custom development only last, when the process is truly yours and gives you an edge competitors cannot buy.

How to apply it: a 90-day plan

Pick one goal and one KPI. For example, cutting stockouts on your best sellers or raising average order value. Measure the current situation for at least four weeks; the guide to KPIs for measuring AI ROI explains how to set the baseline.
Map your data. Where do orders, catalogue, customers and stock live? Are they connected? Who updates them, and how often?
Fix the foundations. Consistent product attributes, a customer database without duplicates, stock synchronised between ERP and ecommerce.
Try what you already have. Your ecommerce platform or ERP often includes recommendations, smart search or forecasting. Build custom only where it is really needed, connecting systems with automation tools (n8n, Make) and AI models via API.
Set the rules. Price and margin limits, approved sources for the assistant, handover to a human, compliance on discounts and transparency.
Measure and decide. At 30, 60 and 90 days compare the KPI with the baseline. If it improves, extend to new products or channels; if not, fix data and rules before switching tools.

To see how ecommerce, ERP, CRM and automation fit into a single system, take a look at the solutions section.

Mistakes to avoid

Buying the platform before the use case. Without a specific problem and a KPI, nobody can tell whether the investment worked.
Letting the assistant make things up. A chatbot that gets a size or a delivery date wrong costs more than it saves.
Automating prices without limits. You need margin thresholds and a record of the 30-day lowest price for discount announcements.
Measuring conversations instead of sales. What counts is conversion, order value and margin, not the number of chats opened.
Treating B2B like B2C. In B2B, custom price lists, reorders and the relationship with the sales rep carry the weight: AI should support them, not bypass them.

Frequently asked questions (FAQ)

What is artificial intelligence in retail?

It is the use of models that analyse sales, catalogue and customer behaviour data to forecast demand, recommend products, propose prices and answer customers. It is not a single tool but a layer on top of ecommerce, ERP, CRM and point of sale, and it only performs well when those systems share clean data.

Will AI replace store staff?

It mainly replaces repetitive tasks, such as counting items on shelves, some checkout operations and standard answers. People can move to advice and assisted selling, with AI tools suggesting availability and complementary products. In B2B the sales rep stays central: AI prepares data and suggestions, while the customer relationship remains theirs.

How much does it cost to add AI to an ecommerce store?

It depends on where you start. Many features are already built into ecommerce platforms or available on subscription, so the biggest cost is often cleaning up product data and integrating with the ERP. Start with a small project, measure it against a baseline and scale it only if the numbers justify it.

Do you need a lot of data to use AI in retail?

You need clean data more than big data. Demand forecasting benefits from at least two seasons of history with promotions and stockouts flagged. Recommendations need product pages with consistent attributes. A customer assistant mainly needs up-to-date product pages, terms of sale and order status.

Is AI-driven dynamic pricing legal in the EU?

Yes, as long as it follows consumer rules. Discount announcements must show the lowest price applied in the previous 30 days as the prior price. If a price is personalised through automated decision-making, the customer must be informed. Both obligations come from the Omnibus Directive, transposed in Italy by Legislative Decree 26/2023.

Does an AI chatbot have to disclose that it is AI?

Yes. The EU AI Act requires that people interacting with a chatbot are informed they are dealing with a machine, unless this is obvious from the context. Beyond the legal duty, it protects trust: make it equally clear how to reach a human agent.

Does retail AI work for B2B companies too?

Yes, with different applications. In B2B, AI suggests reorders to regular customers, checks discounts and margins on custom price lists, answers questions on availability and technical documents in the customer portal, and gives sales reps guidance on what to offer each account.

Sources

If you want to find out which AI project can grow sales in your ecommerce store or B2B channel, book a free assessment.