Superintelligence for Decor Stores: Custom Orders Without Lost Details

Superintelligence, in the strict sense, is not something a decor store buys or runs day to day. What makes a difference is a narrow AI agent with a closed scope: it logs the custom order, checks the production lead time and reports the deposit status, under explicit rules, with a handoff to a person when the decision is not its to make. This article shows how to organize that flow before any automation.

Key takeaways

• Superintelligence describes a theoretical horizon. For a store, what matters is an agent with well-defined tasks, data and permissions.
• A custom order needs a complete record: measurements, finish, material, promised lead time, payment terms and change history.
• The agent looks up and records information, but lead times, discounts, refunds and scope changes stay with the team.
• Deposit and production status must come from a single source, or customers will get different answers on different channels.
• A pilot with one product family, logs and an agreed metric before launch teaches more than automating the whole store.

Why custom orders are different from shelf sales

With shelf sales, the product information is almost fixed: it is in stock or it is not, it costs what the label says, and it leaves the store the same day. A custom order carries a story. The customer chooses a color, asks for a different measurement, wants to see a fabric sample, changes the finish after paying the deposit, and asks three times when it will be ready.

Without a structured record, that information is scattered across WhatsApp threads, notebook notes, spreadsheets and the memory of whoever took the call. This is where orders get lost, deadlines are promised without checking, and customers go unanswered when they most need an answer. Before talking about an agent, it is worth understanding where the store already loses information. A similar case in custom furniture, covering measurements, steps and deadlines, is discussed in AI for Custom Furniture Makers: From Quote to Installation.

What each order record needs to contain

An agent can only follow what is recorded. Before configuring any flow, the store should define the minimum fields for each order:

• Customer, preferred contact channel, and the name of the person who approves the order when it is someone else.
• Item and variant, such as size, color, fabric, material or finish.
• Confirmed measurements, with a reference photo of the space when the item depends on fit.
• Promised lead time, the cut-off date for changes, and the pickup or delivery date.
• Agreed price, recorded in the store system and never reconstructed from memory by the agent.
• Planned deposit, confirmed deposit, balance and the payment method used.
• Change history, with date, customer request and who approved it.

The most sensitive point is the lead time. It should be the date given by the workshop, upholsterer or supplier, plus an agreed margin. If the store does not have that information in hand, the agent must not invent one. It should say that it will confirm with the team.

Status flow: from quote to delivery

A simple status flow keeps the agent from improvising. One possible sequence, adaptable to each store, looks like this:

• Quote open: the customer asked, measurements still being checked.
• Awaiting deposit: quote sent and accepted, payment pending.
• Deposit confirmed: payment recorded and the order released for production.
• In production: the supplier or team confirmed the start, with a reference lead time.
• Ready for pickup or delivery: notice sent with the agreed window.
• Balance confirmed and delivered: order closed, with any incident recorded.

The central rule is that deposit status comes from the payment record, not from the conversation. If the customer says they have already paid, the agent can say the store will check, but it only confirms the status once the payment appears in the system. This protects the store from improper confirmations and the customer from a false expectation.

Illustrative scenario: made-to-measure curtains

Illustrative scenario, no real data: a customer asks for curtains for her living room and sends photos of the window with a measurement written by hand. The agent collects the measurements, marks the order as quote open, and asks the team to verify the measurement before sending a price. After the check, it sends the quote with the deposit terms and records the acceptance.

A few days later, the customer asks whether production has started. The agent checks the status, sees the order is in production, and answers with the reference lead time on record, without promising a date beyond it. If the customer asks to change the fabric, the agent does not confirm the change. It records the change request, marks it as pending store approval, tells the customer the team will reply, and hands the conversation over with a summary: order, current status, requested change and what has already been paid.

The value of this scenario lies in the record and the handoff, not in a clever reply. Someone who picks up the customer's conversation with the full history in hand can resolve in minutes what used to take an afternoon of back and forth.

Limits the agent should not cross

Customer service automation works best when deterministic work sits in fixed flows and judgment stays with the team. Anthropic's guidance on effective agents recommends starting with the simplest solution that solves the problem and adding complexity only when it is needed. For a store, that means:

• Do not promise a lead time that production has not confirmed.
• Do not negotiate discounts, refunds or cancellations.
• Do not confirm measurements that depend on an in-person check.
• Hand off to a person whenever there is a complaint, a legal question or a request outside the record.
• Restrict each action to a specific permission. Whoever can change an order is not necessarily whoever can read its history, and this separation is a basic principle of OWASP's authorization guidance.

On WhatsApp, the agent must respect the customer service window and the rules for outbound messages. Status notices sent outside the window require an approved template and prior customer permission. The details are in WhatsApp 24 Hour Customer Service Window: How an AI Agent Should Act.

Customer data and LGPD

A decor store collects less sensitive data than a clinic, but it still handles names, phone numbers, delivery addresses and photos of the inside of customers' homes. The practical recommendation is to collect only what the order requires, keep photos only as long as needed, and limit access to the team handling the order.

If the AI tool processes data outside Brazil, the analysis must cover the international transfer rules under LGPD. The topic is covered in International Data Transfer Under LGPD When Using AI Providers. The Brazilian National Data Protection Authority (ANPD) publishes guidance and is the official reference for questions on this subject.

How to pilot with our method

The path we use for projects like this has four steps: map, limit, pilot and hand over. Mapping means understanding how an order starts, who touches each step and where information is lost today. Limiting means defining fields, statuses and allowed actions before any configuration. Piloting means choosing one product type and an agreed period, with logs of every conversation and a success metric agreed in advance, such as the reduction in orders without an up-to-date status. Handing over means giving the operation to the team with documentation, then deciding, with data, whether the flow should expand, be adjusted or stop.

For ongoing operations, the agent also needs an owner. When the store has no internal technical team, the managed operation can be handled by Be Human (be-human.ai), under the same permission and recording rules.

Frequently asked questions

Do I need superintelligence to organize custom orders?

No. The store needs a well-structured record, clear statuses and an agent with limited permissions. Those three things solve most of the information loss, and none of them depends on a system with general capabilities.

Can the agent confirm the delivery date on its own?

It should not. The date must come from the workshop, upholsterer or supplier and be recorded in the store system. When the information is not available, the agent should say the team will confirm and log the question for a reply.

How does the agent know whether the deposit was paid?

It should check the payment record, not the conversation with the customer. If the customer says they have already paid, the agent tells them the store will verify and only changes the status after the payment is confirmed in the record.

Should I start on WhatsApp or on the store website?

It depends on where customers already place orders. If most orders arrive by message, it makes sense to start on WhatsApp, respecting the platform's rules. The channel matters less than having a single order record, which should exist regardless of where the conversation starts.

Sources

• Anthropic: Building effective agents
• OWASP: Authorization Cheat Sheet
• ANPD: Autoridade Nacional de Proteção de Dados