AI for Wineries: Wine Club Agent With Age Check Before Any Offer

AI for Wineries: Wine Club Agent With Age Check Before Any Offer

How an AI agent can handle wine club subscriptions and require age verification before any alcohol offer, with limits and human review built in.

•

Want to apply this in your company?

Book a call with Juan →

An AI agent can run the subscription side of a wine club: take inquiries from prospects, collect details, present membership plans, record changes, and answer questions about delivery and billing. To do that safely, it must follow one rule that cannot be negotiated: no alcohol offer happens until age verification is completed and recorded. In practice, this means designing the flow so that verification is the first gate, not a notice buried in the footer.

Key takeaways

  • The agent treats age verification as a mandatory step, before the catalog, pricing, promotions, or any subscription.
  • It works within narrow permissions: it checks subscription status, creates orders that fall within the plan, and routes exceptions to a person.
  • Every relevant decision is logged, with timestamp, channel, verification result, and the reason for any block.
  • Age and contact data are handled with the discipline LGPD requires, with minimal collection and a defined retention period.
  • The safest path starts with a pilot of a single flow, with a success metric agreed before launch.

What the agent does and what it does not do

Before getting into technology, it helps to separate the agent's work from the work of people. The agent is good at repetitive, well-defined tasks: answering frequently asked questions about the club, explaining the difference between plans, confirming that a delivery address is inside the service area, recording preferences the customer states, and reminding the team of shipping dates. It can also organize a subscription's history so staff do not have to search three different systems for the same information.

What it should not do is decide alone on anything with legal or commercial consequences outside the standard. Approving an age exception, releasing a shipment to an address outside the delivery route, granting a discount that is not in the price list, negotiating a cancellation while a complaint is open, or interpreting a delivery restriction are decisions for the responsible people. The agent should recognize these cases, tell the customer what will happen next, and pass the full context to the team.

A simple design rule helps here: if the correct answer depends on a written rule, the agent applies the rule; if it depends on judgment, the agent routes the case to a person. This avoids the most common mistake in customer service projects, which is letting the model improvise on sensitive topics simply because it can write a convincing answer.

Why age verification comes before any offer

In Brazil, the Child and Adolescent Statute (ECA) prohibits selling alcoholic beverages to minors, and selling online or by message does not change that obligation. For a winery or wine club, this means the flow must prevent a minor from receiving an offer, seeing subscription prices, or completing an order. It is also worth asking a lawyer about alcohol advertising rules, which include self-regulation guidelines, and about any local delivery regulations, since these rules vary by case.

In operation, the check usually follows a gate logic. The customer arrives through WhatsApp or the website and receives a message explaining that conversations about beverages require confirmation of legal age. Only after that confirmation does the agent unlock the catalog. If the confirmation fails or the customer declines to provide it, the agent closes the commercial part of the conversation with a neutral reply, without explaining internal rules that could serve as a guide to getting around the system.

It is important not to treat the check as a simple form field the customer fills in with whatever date they like. The design depends on a verification method the company can actually stand behind. That choice should be made with the legal team, because it involves balancing rigor, friction for the customer, and data protection.

Step by step: designing the subscription flow

A practical design can follow these steps:

  • Map the current flow: how a prospect arrives, who responds, where the subscription is created, and where billing and shipping take place.
  • Define the age gate: which question or verification comes first, which answer unlocks the catalog, and which answer ends the conversation.
  • List the actions the agent is allowed to take: look up information, answer standard questions, create an order within the plan, record preferences.
  • List the actions it is forbidden to take: change prices, approve exceptions, reverse an age block, edit sensitive registration data.
  • Define the triggers for handing off to a person: a complaint, a request outside the rules, a legal question, a customer who appears to be a minor, or a request to delete data.
  • Log every step: time, channel, stage, result, and the reason for any block or handoff.
  • Test the flow in a simulated environment with edge cases before any real customer uses it.
  • Agree on a simple success metric, for example the share of handoffs that the team reviews and confirms were correct, and set a date for the review.

This list is a starting point. Each company will add its own specifics, such as clubs with in-store pickup, shipping through a carrier, or in-person events.

Illustrative scenario: an order that almost goes through

Illustrative scenario: someone writes on WhatsApp asking to sign up the club as a gift and gives a birth date that indicates they are below the permitted age. The agent does not show the catalog, does not mention prices, and replies that purchasing beverages depends on the account holder confirming legal age, and it explains the next step. It records the attempt in the log, without keeping more data than necessary, and creates a note for the team containing only the reason for the block. If the person asks to speak with someone, the agent transfers the conversation with a short summary, so they do not have to repeat their story.

The lesson of this scenario is that the agent does not need to be clever to be correct. It needs to follow the right order of steps and know when to stop. When the rule is written and tested, the behavior becomes predictable, and that is what leadership should require.

Personal data, LGPD, and decision records

Date of birth, identity documents, address, and purchase history are all personal data. LGPD requires that collection have a clear purpose, that only what is necessary be used, and that the company know how long it keeps each piece of information. For the club, this suggests storing only the result of the verification, not an image of the document or the full birth date, when the chosen method allows it. The company also needs to define who can access the records and how a customer can request a review or correction.

When the agent makes a decision that affects a customer, such as blocking access to the catalog, there should be a path for review by a person. The topic of automated decisions and the right to review is covered in another article on this blog, which is worth reading with the company's lawyer. The agent's records should also be readable for audits: what was asked, what was answered, which rule was applied, and who reviewed the case.

If the operation uses models hosted outside Brazil, the question of international data transfer also enters the discussion. That point requires case-by-case analysis and should not be settled by assumption.

When the agent should hand the conversation to a person

A wine club deals with trust, gifts, holiday occasions, and sometimes delicate situations, such as a customer who wants to cancel after a family loss. These cases call for human sensitivity. Likewise, any question about age that the standard method cannot resolve, any request for an exception, and any suspicion of fraud should go to a person with the authority to decide.

The handoff has to be done well. The agent should tell the customer that another person will continue the conversation, state the expected timeframe if there is one, and send the team a summary of what has already been said. Transferring without context is what makes the customer feel they are starting from zero.

How to start with a pilot

The safest way to deploy this kind of agent is to start small. Our method, which we describe as map, limit, pilot, and hand off, applies well here: first we survey the current process and the baseline, then we limit data and actions, next we pilot a real flow with logs and an agreed metric, and finally we hand the operation over and decide whether to expand, adjust, or stop. For the club, the pilot can cover only the handling of new prospects, without changing billing or shipping.

Before opening the flow to the public, it is worth testing it with edge cases: an invalid birth date, a customer who keeps pushing, a discount request, a message in another language, an attempt to bypass the check. These tests show whether the rules hold when someone tries to break them. To go deeper on testing, the blog has a article on mocks and sandboxes for agents.

On WhatsApp, outbound messages initiated by the company also require care. A club that sends delivery reminders or news needs to treat contact permission and message templates with the same seriousness. The articles on outbound WhatsApp messages and on what to automate in customer support help separate what is worth automating from what should stay with a person.

At the end of the pilot, the decision should be concrete. If the flow meets the age rule, respects its limits, and produces useful handoffs, it can be expanded. If the agent gets the basics right but fails on edge cases, adjust before growing. If the agent does not bring clarity to the operation, it is better to stop and return to the manual process, with no embarrassment.

Frequently asked questions

Can an AI agent sell wine subscriptions over WhatsApp? Yes, as long as the flow includes an age verification step before any alcohol offer and the company defines what the agent can and cannot do. The rule has to be built into the design, not only written in the introduction text. Validate the verification method with legal counsel before going live.

What should the agent do if a customer gives an age that does not allow the purchase? It should end the commercial part of the conversation with a neutral reply, without explaining internal details that would help someone get around the rule, and log the block. If the customer has questions or asks for a review, the agent should route the case to a person with a summary of the conversation. It should not ask the customer to provide a different date.

Which age data should the company store? The practical rule is to keep the minimum needed to prove the check was done and to support a possible audit. In many cases, the verification result is enough, without copies of documents or the full birth date. The final decision should follow LGPD and the guidance of a legal professional.

Where should a winery start if it has never used AI in customer service? The best starting point is to map the current process, list the actions the agent can take alone and the ones that require a person, and choose a single flow for the pilot. With logs and an agreed metric from the start, leadership can decide based on facts whether to expand, adjust, or stop.

Sources