Superintelligence for Glass Shops: Checking Measures Before the Cut

Superintelligence for Glass Shops: Checking Measures Before the Cut

See how an AI agent checks mirror and shower enclosure measurements, flags jobs that need an on-site visit, and organizes orders before cutting glass.

•

Want to apply this in your company?

Book a call with Juan →

An AI agent for a glass shop receives a request for a custom mirror or shower enclosure, checks whether the measurements the customer provided agree with each other and with the product type, and separates jobs that can move to a quote or cut from those that need an on-site check. It does not replace the final measurement at the job site. Its role is to reduce cuts made on incomplete or contradictory data and to record why each order was released or held.

Key takeaways

  • The agent checks units, required fields and measurement consistency before any cutting decision.
  • Measurements sent by the customer are for triage and quoting, not for a final cut, unless your shop's internal rules say otherwise.
  • Uneven walls, uneven floors, out-of-square openings, holes, cutouts and unfinished construction go on the on-site check list.
  • Every order gets a status with a recorded reason, and the final release for cutting stays with your glass shop team.
  • Start with one product type and a flow that logs every check, before expanding to anything else.

Why the measurement a customer sends rarely settles it

Customers usually measure with a tape measure, alone, at a single point. On a bathroom wall, the width at the top is rarely the same as the width at the bottom. Skirting boards, tile or cladding of different thickness, door frames and thresholds all change how the glass will fit. For a shower enclosure, the usable dimension depends on the type of opening, the direction the door swings, the clearance the profile needs and the point where the glass will be fixed. For a mirror, what often matters is the clear opening, the shape of any cutout and the type of fixing, which may be masonry, a panel, or a surface with a finish that cannot take a drill hole.

The agent cannot see any of this. It reads the text and photos it receives. That is why it must be configured to distrust anything incomplete and to treat a loose measurement as a hypothesis, not as confirmed data.

What the agent checks before any cut

Triage usually follows a fixed list defined by the glass shop itself. A typical checklist includes:

  • An explicit unit of measure, in millimeters or centimeters, not a bare number in a message.
  • Width and height taken at more than one point, when the product rules require it.
  • Consistency between product type and dimensions: the relationship between width, opening type and swing direction must make sense for the chosen model.
  • Glass type and thickness that match the stated use, such as a shower enclosure or a wall mirror.
  • Holes, cutouts, handles, hinges or special finishes, with a drawing or photo attached.
  • Photos of the site when the rules require them, with the elements that affect fixing clearly visible.
  • Address and appointment date, if a visit is needed.

When an item fails, the agent does not fill the gap with a guess. It asks, notes the open item and keeps the order pending.

When the decision must go to an on-site visit

The most useful part of the flow is the rule that separates what the agent can resolve from what must be checked on site. A simple decision guide might look like this:

  • Diagonals measured by the customer do not match, which suggests an opening that is not square.
  • An uneven floor or a threshold that changes the usable height of the enclosure.
  • A room still under construction, with plaster, screed or cladding still being installed.
  • A fixing wall of unknown, fragile material or one that cannot take a hole.
  • Holes, cutouts or corners that depend on a technical drawing not yet agreed with the customer.
  • Measurements that differ between the message, the photo and the form.
  • Products that are very large or heavy, or installations that require more than one person.

A practical rule helps the team trust the flow: if the agent cannot explain a measurement with a photo, a simple calculation and one sentence, the order goes to a site visit.

A measurement the agent cannot explain with a photo, a calculation and one sentence should not become a cut.

Illustrative scenario: a customer sends the width of a shower enclosure and a photo of the bathroom through WhatsApp. The photo shows a visible drop in the floor near the door, and the skirting board occupies the area where the profile would be fixed. The agent records both observations, does not close the measurement and books a technical visit before any cut. The conversation then continues with a summary for the staff member, who schedules the measurement with the information already organized.

How to build the WhatsApp flow, step by step

A lean flow usually has six stages:

  • Collection: the agent asks for product type, room and measurements with explicit units, and requests photos when the rules require them.
  • Validation: it checks fields, consistency and shop rules, without inventing a value for a missing measurement.
  • Classification: each order receives one of three statuses: cleared for quote, cleared for cutting after a simple check by a staff member, or on-site visit required.
  • Handoff: when the customer asks for an exception, disputes a measurement or sends an unclear photo, the conversation moves to a person on the team with a summary of what was collected.
  • Logging: every decision is recorded with date, reason and the person responsible for confirming it.
  • Review: the technical lead confirms the measurements before the cutting order is issued.

Keep in mind that WhatsApp conversations follow their own customer service window rules. If following up with a customer depends on a message sent outside that window, the message must follow the API's template rules, as we explain in WhatsApp 24 Hour Customer Service Window: How an AI Agent Should Act.

Customer data, photos and LGPD

Photos of a home, the address and the name of the person living there are personal data. Brazil's data protection law, the LGPD, requires that collection be limited to what the purpose needs, that the data subject knows how their data will be used, and that access is controlled. In practice, a mirror quote does not need a tax ID number, and a site photo should show only what serves the measurement. Brazil's data protection authority (ANPD) publishes the official guidance on this topic and is the reference to check.

If the AI tool processes data outside Brazil, you also need to verify the rules on international data transfer. That point is covered in International Data Transfer Under LGPD When Using AI Providers.

How Juan Carlo would apply this to this operation

In our approach, the work starts by mapping the current process: how the request arrives, who measures, at which stage measurement errors appear and who corrects them. Then we limit the agent to triage and preliminary quoting, with no permission to release cuts and restricted access to the shop's system. The pilot covers one product type, for example bathroom mirrors, with a log of every check and an agreed metric defined before it starts, such as the number of orders sent back for review. At the end, the team decides whether to expand, adjust or stop the flow. The same logic of measurement and approval appears in Superintelligence for Woodworking Shops: Measure, Design, Approve.

Frequently asked questions

Can the agent confirm a measurement and release a cut on its own? In our approach, no. The agent triages, organizes and records, but releasing a cut stays with a responsible person, because the final measurement depends on site conditions the agent cannot verify. That decision must also be logged for audit and review.

Can a mirror based on a customer's measurement go straight to production? It depends on your shop's internal rules. When the measurement came only from the customer, the order should normally pass a simple check by a staff member or technician before cutting. When a technician measured on site and recorded it, the flow is more direct.

What data should the agent ask for on WhatsApp? It needs the product type, the room, the measurements with units, the glass type, photos when the rules require them, and a visit date if one is needed. Data that does not serve the measurement, such as personal documents, should not be requested for an initial quote.

How do you know whether a pilot is worth starting? It is worth starting when the team already sees orders coming back because of poorly informed measurements, and when someone can define the check rules. Pick one product, agree on the metric before starting and log every decision the agent makes. If the results do not justify expansion, adjusting or stopping the pilot are also valid outcomes.

Sources