
AI for Shoe Stores: Sizes, Exchanges and Stock by Location
How a shoe store can use AI to check sizes and stock by location, organize exchanges, and alert staff when a requested size is not on the shelf.
Want to apply this in your company?
Book a call with Juan →An AI agent for a shoe store checks size and stock at each location, tells the customer on WhatsApp what is actually available, organizes size exchanges, and alerts the team when the requested model is not in the range. The result depends less on whether the conversation sounds natural and more on three things: a reliable stock source, explicit exchange rules, and a predictable path for the case where the product does not exist.
Key takeaways
- Sizes and stock must come from the store's system, per location, and never from an answer generated by the model.
- Size exchanges should follow a written policy, with a clear split between what the agent can offer on its own and what needs a staff member.
- When the item does not exist in the requested size, the agent should offer real alternatives and hand the case over with full context.
- Every inquiry, exchange and alert should create a record, so the team can review decisions and fix the rule.
- The pilot should start with one category and one location, with a metric agreed before scaling.
Why sizes and stock are the starting point
In most shoe stores, the customer question is simple and repetitive: do we have this model in this size, at this location, or at another branch? When the answer depends on someone checking the shelf or calling the warehouse, the service stalls, and the customer often gives up or travels to the store without any guarantee.
An AI agent can solve this step if it is connected to the right source. That means reading stock from the management system or point-of-sale, filtering by model, color, size and location, and answering with what the system shows at that moment. If the system has no integration, the alternative is a spreadsheet updated on a set schedule, but in that case the agent must say the information may not reflect the latest sale.
The most expensive mistake is an agent confirming a size that does not exist. That creates frustration, complaints and extra work for the staff. So the first decision is not which language model to use, but which data the agent can query and how old that data is when it arrives.
How to organize stock checks by location
Before connecting any tool, map the current flow. It usually follows an order that can be described like this:
- List the most common availability questions, such as model, color, size and nearest location.
- Identify where that information lives today: system, spreadsheet, marketplace or a salesperson's memory.
- Define the freshness rule: what the agent can state with certainty and what must be confirmed with the store.
- Separate the availability check from the reservation. Checking is a low-risk action. Reserving or promising a set-aside item requires explicit permission and a record.
- Write a standard reply for each case: available at this location, available at another location, not available, and available to order, if the store works with special orders.
This mapping can feel like paperwork, but it is what stops the agent from inventing an answer when the data is incomplete. The practical rule is simple: if the tool cannot confirm, the agent says it cannot confirm and hands the question to the team.
How to handle exchanges without promising what does not exist
Size exchanges are often the point of greatest friction because they involve three parts: the customer, the product they already bought, and the stock of the desired size. The exchange policy is a commercial decision that varies by store, but it must be written down before the agent starts answering. That includes the time window, the condition the product must be in, whether a receipt is needed, and whether the exchange can happen at another location.
A good division of responsibility separates what the agent resolves alone from what goes to a person. In general, the agent can look up the policy, explain deadlines and check whether the desired size is in stock. Approving exceptions, deciding on an exchange outside the window, assessing a defect or handling a complaint should stay with a staff member.
The exchange path also needs to be recorded: who asked, which product and size left the store, which size was requested, which location handled it, and what decision was made. That history lets the store correct the policy and see where the failures are, without relying on the memory of whoever was at the counter.
From a data protection standpoint, the agent should collect only what the exchange requires, such as name, order or receipt number and contact details. Payment data should not appear in the conversation. When the service involves personal data of Brazilian customers, it is worth aligning the operation with the LGPD from the start. The ANPD guidance on personal data processing is an official starting point for that.
When the size does not exist: the alert to the team
The out-of-stock case matters most to the operation, because this is where a sale can be saved or lost. The customer's question is clear, but the answer should follow a predictable sequence. A decision guide helps the team understand what happens in each situation:
- If the size exists at another location, the agent gives the location, address and hours, and asks whether the customer wants the store to hold the item, when that function exists.
- If the model exists in another color or variant in the same size, the agent suggests that alternative with a photo and price taken from the official catalog.
- If a special order is possible, the agent explains the lead time the store has provided and logs the interest request for human confirmation.
- If nothing is available, the agent closes the conversation honestly, asks permission to notify the customer when the size arrives, and hands the case to the team with a complete record.
The alert to the team is a central part of the flow. It should arrive with the model, color, requested size, locations checked, time, channel and what the customer has already been told. That way the salesperson does not repeat questions and can pick up the conversation with context. To log demand without excess, the store can use a service queue or its own CRM, as long as the agent does not store more information than needed.
One detail changes the outcome: arrival alerts for products are outbound messages, and sending them depends on the customer's permission. Before sending any notice, the store must collect that consent and follow the platform's rules. On this point, it is worth reading about the 24 hour window on WhatsApp API, which explains when the agent can reply freely and when it needs an approved template.
Limits, data and oversight in customer service
A shoe store agent should run with limited permissions. It can check stock, read the exchange policy and log service requests. It should not change prices, grant discounts, cancel a sale, touch payments or promise supplier lead times without validation. Keeping reading, suggesting and executing separate is what prevents a conversation error from becoming a financial error.
It is also important to treat the agent as a possible target for manipulation. A customer may try to make the system ignore the policy with phrases asking it to disregard earlier instructions. Security guidance for applications built on language models, such as the OWASP guidance, recommends limiting what the model can access and validating the actions it proposes, rather than relying only on the prompt text.
Finally, oversight has to be planned. The agent should know when to transfer to a human: a complaint, a defect, a request outside policy, an upset customer or a question the knowledge base does not cover. That handoff should include a summary, not an instruction for the staff member to ask everything again.
How to start with a pilot
The safest path is to start small. Choose one category, such as running shoes or dress shoes, and one location with enough volume to test. Before any setup, record how service works today: how long it takes to confirm a size, how many exchanges need approval and how many customers leave without an answer. The pilot metric should be agreed with the store before starting, not chosen afterward to justify the result.
The method we use in projects like this has four steps: map the current process and its baseline, limit data and actions, pilot a real flow with records and an agreed metric, then hand the operation to the team and decide whether to expand, adjust or stop. This sequence does not guarantee results, but it reduces the chance of automating a process nobody fully understands.
Illustrative scenario: a store with three locations gets many questions about the same sandal model in a less common size. In the pilot, the agent checks stock at all three locations and tells the customer where it is available. When there is no size, it logs the interest request with context. The team reviews the records every week and finds that the problem is not the conversation but the lack of restocking of that size at one specific location. The decision becomes a purchasing decision, not a conversation problem.
If the store is still deciding which channel to use for this kind of communication, compare options before choosing. The article on SMS, RCS or WhatsApp helps think through that choice from both the customer's and the operation's side.
Frequently asked questions
Can an AI agent confirm on its own that a size is available? Yes, as long as it checks an up-to-date stock source and can say when the information cannot be confirmed. If the system is not integrated, the reply should say the check depends on the team, rather than assuming availability.
Can the agent approve an exchange outside the deadline? It should not. Approving exceptions involves commercial judgment and, in some cases, defect and consumer rights questions. The agent's role is to check the policy, explain what is covered and transfer the case to a staff member with the full history.
Which data should a shoe store agent collect? Only what the lookup or exchange requires, such as name, order or receipt number, contact details and the desired model. Payment data should not be requested or stored in the conversation. The store should also define how long records are kept and who can access them.
Should we start with all stores and all products? Usually not. A pilot with one category and one location quickly shows where the data, policy and handoff gaps are. Once the flow is validated, the store decides whether to expand, adjust the rule or stop the project.