AI for Scrap Yards: Check Weigh Tickets, Invoices and Suppliers

AI for Scrap Yards: Check Weigh Tickets, Invoices and Suppliers

How to record weigh-ins, supplier details and purchase documents at a scrap yard so AI and staff can check invoices and records before closing the day.

•

Want to apply this in your company?

Book a call with Juan →

In a scrap yard, the useful AI is not the one that decides what to pay for material. It is the one that records every weighing, matches the supplier against the registry, and checks that the invoice agrees with what came through the gate before the cash count closes. Your team stays responsible for the scale, the payment and tax validation. The agent only flags discrepancies and records the reason.

Key takeaways

  • Every purchase needs one record with weighing, supplier, material and document, linked by the same internal number.
  • AI checks consistency (required fields, weights outside the normal range, missing documents), but it does not replace the scale or your accountant.
  • The supplier registry should ask only for the data it needs, with a clear purpose and restricted access, because of Brazil's data protection law (LGPD).
  • No purchase closes with an open issue unless a person approves it and the approval is recorded with name and time.
  • Start with one yard, one material type and an agreed metric before expanding.

What every purchase record needs

Before talking about AI, look at what the operation already produces. In many scrap yards the information is scattered: the scale ticket, a notebook, the seller's name in a WhatsApp chat, a photo of the material, and an invoice issued later, sometimes at the end of the day. When those pieces do not match, checking depends on the memory of whoever was at the gate.

A minimum purchase record usually brings together:

  • An internal purchase number, unique and generated at the moment of weighing.
  • Date, time, shift and the identity of the person who registered the purchase.
  • Gross, tare and net weight from the scale, with the identification of the equipment.
  • Material type, according to the classification your team already uses.
  • Supplier identification and the agreed payment method.
  • The document presented by the seller and the number of the invoice or receipt issued.
  • Photos of the material and the vehicle, when your house policy requires them.

The central point is the internal number. Without it, the invoice, the ticket and the photo become loose files, and the AI has no reliable way to cross-check them.

How to set up the supplier registry without collecting too much data

Individual suppliers bring personal data such as a tax ID number and an address. Company suppliers bring a company registration number and public business details. In both cases, the registry should keep only what the purpose requires: issuing the correct document and keeping the purchase history. For guidance on handling personal data, the reference is Brazil's National Data Protection Authority (ANPD), at gov.br/anpd.

A simple plan to start:

  • Define the supplier types you deal with: individual, company, cooperative or intermediary.
  • For each type, list the minimum fields your accounting and the issued document require.
  • Keep the registry separate from chat history. The agent reads the registry but should not copy full documents into a conversation.
  • Record who can see each field, and keep tax IDs and documents out of any message sent to support.
  • Review with your accountant what must be stored and for how long.

If the AI model is hosted outside Brazil, the transfer of personal data needs to be assessed before it goes live. This topic is covered in international data transfer under LGPD when using AI providers.

Weighing: what the AI checks and what stays with the scale

The AI does not weigh anything. It receives the record that the scale and the operator produced and checks things like these: whether net weight equals gross weight minus tare, whether the weight falls outside the historical range for that material and vehicle, whether the same vehicle identification appears in two purchases on the same day, and whether any required field is empty. Calibration and verification of the equipment remain the responsibility of the yard, following the metrology rules that apply to it.

A practical rule for alerts:

  • Record and continue: fields completed, weight within the expected range, document present.
  • Alert the operator: inconsistent net weight, a new supplier, or document data that differs from the registry.
  • Block closing and call the responsible person: missing tare, the same document used in two purchases, or a supplier on your internal restriction list.

Every alert should be logged with the date, the rule that fired, the person who handled it and the decision taken. That history is what lets you review later whether the rules are well calibrated.

Purchase documents: the checklist before closing

The closing of each purchase can follow a short list. It does not replace the tax review by your accountant, but it keeps open issues from slipping through at the end of the day:

  • Invoice or receipt with number, date and supplier identification matching the registry.
  • Weight and material on the document consistent with the scale, within the tolerance your yard has defined.
  • Value calculated by the price table rule, with no free-text entry at closing.
  • Payment method agreed and recorded under the same internal number.
  • Document file attached to the purchase, named so it can be found easily.
  • Approval from whoever can release exceptions, with name and time.

The AI can read a photo of an invoice and compare its fields with the record, but a person should review the reading when the image is blurred, cropped or handwritten. One extra alert is better than a purchase closed with wrong data.

Checking works when the open issue is visible to the person who decides, not hidden in an observations field.

Illustrative scenario: a busy afternoon at the yard

Illustrative scenario: a scrap yard with two gates and one scale operator. A seller who has worked with the yard before arrives, but his registry entry still has an old address. The weighing record is complete and the net weight matches the tare, but the photo of the document came out unreadable. The agent marks the purchase as pending, shows the operator the two missing items, and does not release the closing. At the end of the day, the responsible person checks the paper document, updates the registry and releases the purchase. The AI did not set a value or approve anything. It stopped an open issue from passing unnoticed and kept a record of the decision.

How to apply the method without replacing everything at once

In how we work with clients, the path starts by mapping how a weighing turns into an invoice today and who checks each step. Then we limit the agent: it reads registry and records, flags open issues, and does not change weight, value or payment method. The rules can be tested with simulated purchases before any real use, as described in mock tools for testing AI agents in a sandbox.

The pilot runs in one yard and one material type, with a log of every alert and an agreed metric from the start, for example the time between weighing and closing with all fields checked. At the end, the decision is to expand, adjust the rules or stop. The same model applies to purchases that involve quotes, as discussed in an AI agent for procurement with human approval.

Frequently asked questions

Do I need to replace the yard's management system to use AI?

Not necessarily. Many operations start by exporting scale and purchase records into a spreadsheet or a simple database, using the internal number as the key. The agent works on that data and records alerts, without taking over the role of your tax or inventory system.

Do I need to keep a photo of the document and of the seller?

Keep what the purpose requires and what your accountant asks for. For photos of the material and the vehicle, define in your house policy when they are mandatory. Before collecting any data, confirm the requirements with your accountant and assess the legal basis for processing, using the public guidance from the ANPD.

Can AI read handwritten receipts?

In some cases, yes, but the reading should be treated as a suggestion. When the image is poor or the handwriting is ambiguous, the system should ask for human review and should not fill the field on its own.

Can AI decide how much the yard pays for a purchase?

The value should come from a price table rule set by the yard and reviewed by the responsible person. The AI can calculate, compare and flag deviations, but exceptions need a recorded human approval, with reason and time, so the decision can be audited later.

Sources