Artificial intelligence in retail goes beyond chatbots or text generation.
When it can use point-of-sale data — sales, inventory, catalog, customers or locations —, it can above all make that data easier to query, detect situations worth watching and prepare certain actions.
Here are 10 real use cases of AI in a physical store.
1. Analyze sales
“What is my revenue for the week, and which products explain the change?”
Sales receipts contain a lot of information: products sold, quantities, amount, date and time of the transaction, or payment method depending on the data recorded.
An AI able to query this data can cross-reference several metrics to look for what contributes to a change.
For example, a drop in revenue may come from:
- fewer receipts;
- a lower average basket;
- a decline in a product family;
- key items selling less;
- an unusual comparison period.
The point is therefore not just to get the revenue figure, but to look for the factors in the data that help interpret its change.
2. Spot dead stock
“Which products have had no sales in 60 days?”
A catalog can contain items that stay in inventory for a long time without selling.
AI can help identify them based on measurable criteria:
- last sale date;
- quantity in stock;
- historical sales rate;
- stock value when the cost needed for the calculation is available.
An item with no sales in 60 days is not automatically a problem. The relevant duration depends on the product and its sales cycle.
The goal is therefore mainly to bring out the items that deserve a closer look, then decide whether to keep the stock, reduce orders, transfer it or run a promotion.
To go deeper: Dead stock and tied-up inventory.
3. Anticipate stockout risks
“Which products are at risk of a stockout before the next restocking?”
Knowing the current stock is not enough.
You also need to take into account how fast the product sells and, when the information is available, the lead time needed to replenish it.
An item with 20 units can be comfortable if it sells once a week and critical if it sells ten times a day.
AI can help cross-reference:
available stock + sales rate + replenishment lead time
in order to identify the items for which stock could become insufficient.
This is an estimate, not a certain stockout date: future sales may differ from the history.
See also: Anticipating a stockout.
4. Prepare restocking
“Which items should I probably reorder this week?”
Once the risks are identified, the next step is to prepare replenishment.
Several data points can come into play:
- available stock;
- recent sales;
- orders already in progress;
- safety stock;
- supplier lead time;
- packaging;
- minimum order;
- seasonality.
AI can help gather this information and prepare a restocking proposal.
But detecting that a product is dropping fast is not enough to determine how much to order.
The right quantity depends on the supply context and the stock target.
A proposal can then be checked before the supplier order is created or sent.
5. Understand changes in the average basket
“Why did my average basket drop this month?”
The average basket is revenue divided by the number of receipts.
Its change can be compared with other data:
- number of items per receipt;
- categories sold;
- product prices;
- promotions;
- products frequently bought together.
AI can help explore these relationships.
For example, it can show that a high-value category is less present in receipts or that certain product pairings come up frequently.
This provides leads to examine for merchandising, cross-selling or the commercial offer.
However, a correlation in receipts does not prove that a suggested pairing will increase the average basket.
6. Segment and analyze customers
“Which customers bought most often over the last three months?”
The notion of a “best customer” is ambiguous.
Depending on the goal, the store may look for:
- the most frequent customers;
- those who generated the most revenue;
- those whose last purchase is recent;
- the most regular customers;
- those who are starting to stop coming back.
AI can make it easier to cross these criteria when sales are linked to identified customers.
The RFM method — recency, frequency, monetary value — is, for example, a classic way to segment customers, see Customer segmentation.
The point is not to produce a universal podium of “best customers”, but to bring out the profiles that match a specific goal.
To go further: Identify your best customers.
7. Prepare a promotion
“Which overstocked products could be part of a promotion?”
Data can help choose which items to review before a promotion.
For example:
- high stock;
- low turnover;
- no recent sales;
- season coming to an end;
- dead stock.
AI can bring out the products that match these criteria and help prepare a promotion.
But a dead-stock product should not automatically be marked down.
You also need to consider margin, commercial positioning, the season, future inventory and any applicable constraints.
AI can therefore prepare the decision and the action, without replacing commercial validation.
8. Check consistency between store and online store
“Are there gaps between the stock available in store and the stock shown online?”
In an omnichannel setup, the availability shown online must stay consistent with the stock actually available.
When systems are synchronized, stock movements can be passed on between the different channels.
AI can be used as a control layer to look for certain inconsistencies:
- unusual quantity;
- item available on one channel but not the other;
- negative stock;
- synchronization lag visible in the data.
It does not replace the synchronization mechanism itself.
Synchronization is the job of the management system; AI can help check or query its result.
9. Compare several points of sale
“What major gaps show up between my stores this week?”
For a network, comparing revenue alone quickly produces an uninformative ranking.
Locations can differ in size, foot traffic, assortment or location.
The analysis can therefore cover several metrics:
- revenue;
- number of receipts;
- average basket;
- sales by category;
- stockouts;
- inventory turnover;
- change compared with a comparable period.
The point is less to determine which store is “the best” than to spot the gaps that deserve an explanation.
Strong sales of an item in one store and excess stock in another can, for example, lead you to look at whether a transfer makes sense.
For businesses with several addresses: see Multi-store management.
10. Automate certain recurring checks
Some analyses come back every day or every week. For example:
“Every Monday, summarize last week's sales and flag the main anomalies.”
Or:
“Every morning, show the items close to their stock threshold.”
When an analysis is repetitive and relies on available data, it can become a routine. See Automating your store's tasks for how routines work in detail.
The point is not to produce more reports.
It is to regularly surface the information that may require a decision, without manually redoing the same search.
The value comes mostly from cross-referencing data
Taken separately, each of these use cases can already be done with reports, filters or dashboards.
The contribution of an AI connected to store data shows up mostly when a question requires crossing several dimensions. For example:
“Which products have slowed down sharply while their stock remains high?”
combines sales and inventory.
“Which regular customers haven't bought anything in two months?”
combines customer history, frequency and recency.
“Which stores risk a stockout on an item that is overstocked elsewhere?”
combines sales, inventory and locations.
The point is therefore not simply to “add AI” to every store function. It is to be able to query several pieces of data linked to the same decision.
At Gillia, this logic links the POS, sales, inventory, customers and the online store in a single environment. To understand the general principle of this agent, see AI agent for retail: definition, use cases and examples and the details of in-store inventory management.
From analysis to action
There is also an important difference between three levels of use.
- Understand“What is my average basket this week?” The AI returns or calculates information from the available data.
- Analyze“Why did it drop?” It looks for the variations or factors in the data that may help explain it.
- Prepare or carry out an action“Prepare restocking for the at-risk items.” The AI uses the result of the analysis to prepare an operation in the tools it is connected to.
The closer you get to action, the more important control over data, rules and validation becomes.
An incorrect analysis is a nuisance. An incorrect supplier order or a badly configured promotion can have a direct impact on the business.
To dig into this difference, see AI assistant or agentic AI agent. Gillia use cases also detail specific cases such as anticipating stockouts, promotions on dead stock or comparing several points of sale.
What data do you need to use AI in retail?
What is possible depends directly on the data you can actually access.
To analyze sales, you need the transactions. To anticipate stockouts, you need at least the stock and the sales, and ideally replenishment lead times. To analyze customers, transactions must be linkable to identified customers. To compare several stores, the data must be available at a common scale.
An AI does not create missing data.
The quality of the result therefore depends as much on the quality and availability of the data as on the model used.
How does Gillia apply these use cases?
Gillia brings together the store data and tools it is connected to so that merchants can query them from a single interface.
A request can start with:
“Which products are at risk of a stockout?”
Then be refined:
“Which ones matter most in terms of revenue?”
And then lead to:
“Prepare restocking for these items.”
In the same way, a customer analysis can start by identifying a segment before preparing a loyalty action.
The goal is to enable the sequence:
understand → analyze → act
without presenting every AI result as a certain decision.
Operations with a real impact on the business can remain subject to validation before execution.
Retail AI is only useful if it improves a decision
The value of an AI tool is not measured by the number of features that use the words “artificial intelligence”.
In a store, the relevant question is simpler: does this use case let you understand a situation faster, spot something that might have been missed or prepare an action more efficiently?
If the answer is no, automation or a classic dashboard may sometimes be enough.
If the answer is yes, AI becomes particularly interesting when it makes it possible to move quickly from a business question to the relevant data, then possibly to a controlled action.
To go further, see these real examples of Gillia use cases.