Superintelligence for Clothing Boutiques: Size Notes, Holds and Returns
Keep size notes, holds and return records in structured form so a future agent can draft accurate personal outreach that staff approve first.

Superintelligence for a clothing boutique means an agent that can write useful, personal outreach because the shop's memory is organized and reliable. That memory consists of three kinds of record: client size and fit notes, item holds with clear start and end dates, and return records with consistent reasons. An agent is only as accurate as these records, so the boutique should structure them and decide what the agent may read, draft and send before any automation starts.
Key takeaways
- Size notes are only useful when they name the garment, the cut and a plain fit comment, not a vague size label.
- Every item hold needs an end date, an owner and a status, or an agent will remind clients about stock that is no longer held.
- Return reasons should come from a fixed list so patterns can be read later without guessing what a free-text note meant.
- Start with read access and drafts that staff approve, and keep sending, refunds and hold changes with people until accuracy is proven.
- Choose one role and one observable success criterion, such as correct hold reminders, before adding fit notes or returns to the scope.
Why a boutique's memory lives with the staff
In most boutiques, the most useful information about a client lives with a few people. The senior stylist knows that one client wants a longer sleeve. The associate remembers that a dress was set aside for a wedding and that the client asked to hear before it was released. When that knowledge stays in someone's head, a notebook or a message thread, no future system can use it, and outreach falls back to generic newsletters that most clients have already learned to ignore.
Personal outreach works when the shop can answer three questions quickly: what this client wears and how it fits, what is currently held for them and until when, and what has come back and why. Those answers are the boutique's memory. Turning them into records is not a technology project first. It is an operations project that decides what to write down, where to write it, and in what form. Anthropic's guidance on context engineering makes a similar point for agents in general: give the agent the smallest set of relevant information for the task, not everything the business has ever recorded.
Which three records to keep, and how
Keep the list short. Each record below should answer a question that staff ask every day, and each should be linked to one client profile rather than floating on its own.
- Size and fit notes: the client, the brand and cut, the size bought, and one or two plain fit comments such as runs short in the body or prefers the hem above the ankle. Avoid vague wording like medium-ish.
- Item holds: the client, the item identifier, the date the hold started, the date it ends, who placed it, and whether a deposit was taken. A hold without an end date cannot be reminded correctly.
- Return and exchange records: the item, a reason from a fixed list, the condition on return, what the client chose (refund, exchange or store credit), and the outcome.
Fixed categories beat free text for anything a machine has to read. A list of six to eight reasons with an other option covers most boutiques, and it should be reviewed twice a year so it reflects what customers actually say. Free text remains useful for nuance, but the category field is what makes later reading reliable. If you want a parallel from another retail setting, AI for Ski Rental Shops: Fitting Records and Seasonal Holds shows how fitting records and holds can be structured in a similar way.
A step-by-step plan to clean the records
You can do this in a shared spreadsheet or in a client management tool you already use. The tool matters less than the discipline of the steps below.
- Step 1: Write down the five questions your staff answer from memory each week, such as which client is waiting for a size to be restocked.
- Step 2: Map each question to one record and its required fields. Remove any field that nobody will use.
- Step 3: Choose one place for each record. A client's size note should not be split between a notebook, a chat app and a spreadsheet.
- Step 4: Add an end date to every hold and run a daily check that either releases the hold or asks a named person to extend it.
- Step 5: Create the return reason list and agree on who selects the category at the counter, so the same reason is not recorded three different ways.
- Step 6: Agree on consent. Record only what the client has accepted, and give staff a simple way to remove a note when a client asks.
- Step 7: Test the records. Give a colleague who did not write the notes three real client questions and see whether they can answer from the records alone.
What an agent may read, draft and send
Access should follow the job, not the software. The OWASP authorization guidance describes the basic principle: each action should be checked against what the requester is allowed to do, and access should be as narrow as the task allows. For a boutique, the rules can be written in four lines.
- Reading client size notes and open holds for a follow-up role: allowed, because the role needs them to do its work.
- Drafting a personal message: allowed, with every draft waiting in a queue for a named staff member.
- Sending a message: only after approval, or through a narrow template that makes no claims about stock, price or availability beyond what the record says.
- Changing a hold, issuing a refund or editing a size note: staff only. The agent may suggest the change, but it does not make it.
Treat text inside records as data, not instructions. A comment written by a client, such as a request to ignore the hold rule, is still just text and should never change what the agent is allowed to do. Keep the agent's instructions separate from anything a client wrote, and keep the action list short. For a broader view of risk, the NIST AI Risk Management Framework offers a structure for mapping, measuring and managing risk that a small business can apply without a large team.
Illustrative scenario: a hold reminder that goes out correctly
This scenario is illustrative and does not describe a specific shop. A boutique holds a coat for a client for five days, with the start and end dates recorded and a deposit noted. On the fourth day, the follow-up role finds that the hold ends the next day. It drafts a short reminder that names the coat, states the end date from the record and mentions the deposit, then places the draft in the store manager's queue.
The manager confirms that the dates are correct, changes one sentence so it matches how the shop usually speaks to that client, and approves it. The agent does not release the hold or change the deposit. If the hold had no end date, the agent would have had nothing reliable to write, and the correct behavior would be to flag the gap for staff rather than guess.
How to measure the setup and keep it current
Choose one success criterion before anything is connected. Examples include the share of drafted reminders that staff approve without edits, or the number of holds that expired without a recorded decision. Count them weekly. If the numbers do not move, the cause is usually a missing field or an unclear rule, not the wording of the message.
Record the outcome of every agent action, including the draft, the approver and the final message sent, so a later review can explain what happened. When a client asks why they received a message, the shop should be able to answer from the log.
Be Human designs and operates this kind of setup as a managed service. A first engagement defines one useful role, the memory and tools it needs, the approval boundaries and an observable success criterion. Managed operations can then cover monitoring, recurring evaluations, integration maintenance, permission review and controlled improvements, so the records and access rules keep matching how the boutique actually works. The service is remote and available in English and Portuguese.
Where to start
Start with one record and one role rather than the whole shop. The hold list is often the best first choice because its end dates make errors easy to see and easy to check. Once the hold records are clean and the approval queue works, fit notes and returns can follow the same pattern.
If you want a structured view of which record and task should come first in your shop, the free Be Human AI diagnostic will help you find the first task worth handing to an agent.
Frequently asked questions
Do I need special software to keep size notes for an agent? No. A well structured spreadsheet or a client management tool can hold the records, as long as each note is linked to one client and uses consistent fields. Software becomes more important when several people update the same record and an agent must read it reliably. The structure of the records matters more than the product you choose.
Can an agent message clients on its own? It should not, at least at the start. A safe first setup lets the agent read records and draft messages that a named staff member approves. Automatic sending can be considered later for narrow templates that make no claims about stock, price or availability, once the drafts have been reviewed over time and the permission rules are written down.
What happens to a client's notes if they ask us to delete them? The shop should be able to remove the note and any drafts built from it, and the request should be logged. Decide how staff handle these requests before an agent is connected. Privacy obligations depend on where your clients live, so confirm them with your advisers rather than relying on a general answer.
Where should a boutique start if its records are messy? Start with the hold list, because holds have clear start and end dates that make errors easy to see. Clean the fields for one week of holds, test a colleague on them, and only then expand to fit notes and returns. Trying to fix every record at once usually stalls the project.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


