Skip to content
Giallo

AI in warehousing: agents that check delivery notes and draft reorders and inventories.

This page follows a typical day in a distribution warehouse or a small business with its own stock, from the first unloading in the morning to the inventory close. For each moment we show the agent we would build, the programs it connects to, the model we choose and who has to approve before any data changes in the management software.

AI in warehousing: white shelving with neatly arranged boxes. A thin robotic arm pulls out a single yellow parcel, shown head-on as a square

How do you use AI in a warehouse?

AI in warehousing pays off most on the documents and questions that move around the shelves: supplier delivery notes to check against orders, availability requests, delivery confirmations, reorder proposals, inventory discrepancies. With Giallo Studio, GPT reads delivery notes and confirmations, Gemini answers on stock levels and Claude reasons about reorders; goods receipts and adjustments are signed off by a person.

A working day, agent by agent.

The reorder agent's Friday: the shelf empties with the week's consumption, items that fall below minimum stock enter the draft order with the reason, the unusual consumption of springs stays with the buyer, and incoming goods appear only after the buyer signs.
  1. When the warehouse opens, as the first trucks and the supplier delivery notes arrive.

    Goods receiving agent

    Compares the supplier's delivery document with the purchase order and prepares the goods receipt as a draft.

    1. Reads the delivery note photographed at receiving or received by email and extracts lines, codes, quantities and batches.
    2. Matches supplier codes to internal items using the cross-reference table already in the management software.
    3. Compares each line with the open purchase order and flags overages, shortages and mismatched units of measure.
    4. Prepares the draft receipt transaction and the checklist for the warehouse worker, with the lines to recount at the top.

    Connects to: Management software or ERP · Receiving handheld · Purchasing email inbox · Custom connector

    GPT
    It extracts order lines from delivery notes with ever-changing layouts into a rigid format, which the connector checks before proposing the goods receipt.
    Codex
    It writes the connector to the management software and the script that maintains the supplier code table; it doesn't read supplier documents.

    Who approves. The warehouse worker confirms the goods receipt after the physical check: the agent never posts a final transaction based on the delivery note alone.

  2. All morning long, every time a colleague asks whether an item is on the shelf.

    Stock availability chat agent

    Answers colleagues on stock, location, batches and expected deliveries by reading the management software in read-only mode.

    1. Recognizes the item even from rough descriptions, internal abbreviations or half-written codes.
    2. Reads available stock, quantities already committed, location and batches with their expiry dates.
    3. If the item is out of stock, it gives the next delivery expected on the supplier order, saying it's an unconfirmed date when that's the case.
    4. When the question is ambiguous, it proposes the two or three closest items instead of picking one at random.

    Connects to: Company chat · Management software or ERP · Item master data

    Gemini Flash
    Many questions a day, short answers: if the company works in Google Workspace, it integrates with the chat colleagues already use and keeps costs low.
    Claude Haiku
    It's the alternative we test on the same questions when the company uses a different chat, choosing whichever gets the item wrong least often.

    Who approves. The warehouse manager decides which data each department can see; purchase prices and margins never appear in the answers.

  3. In the early afternoon, when order confirmations, delay notices and updated price lists come in.

    Supplier confirmations agent

    Aligns the delivery dates confirmed by suppliers with open orders and flags what changes for sales.

    1. Sorts confirmations, delay notices and new price lists among the day's emails.
    2. Compares confirmed dates and quantities with those requested and updates the expected date on the order as a draft.
    3. If a delay affects items already promised to a customer, it alerts the sales rep concerned with the list of affected orders.
    4. For price lists, it highlights the items with price changes that need approval before the update.

    Connects to: Purchasing email inbox · Management software or ERP · Supplier price lists

    GPT
    It reads price list tables and PDF confirmations from different suppliers and returns orderly lines that can be compared with orders.
    Claude Sonnet
    It writes the alert to the sales rep clearly and without alarmism, distinguishing what's confirmed from what's only announced.

    Who approves. The new dates stay in draft until the purchasing office confirms them; updated prices always go through the buyer.

  4. Friday morning, before the purchasing office closes out the week.

    Reorder agent

    Proposes a draft purchase order to suppliers by applying the stock rules set by purchasing.

    1. Reads stock, commitments, recent consumption and orders already in transit for every stocked item.
    2. Applies the rules written with the buyer: minimum stock, reorder quantity, lead times stated by the supplier.
    3. Calculates quantities with a formula fixed in the code, not with the model, so every number is reproducible.
    4. Writes the reason for the proposal next to each line and flags items outside the rules, such as those with sudden spikes in consumption.

    Connects to: Management software or ERP · Stock rules sheet · Google Sheets or Excel

    Claude Sonnet
    It explains in a few readable lines why an item is in the proposal and recognizes when unusual consumption calls for a question rather than an order.
    Gemini
    Useful when the stock rules live in a Google Sheet shared with purchasing: it reads and updates the sheet without manual steps.

    Who approves. The buyer sends the order to the supplier after reviewing the draft; the agent never commits the company to a supplier.

  5. On inventory evenings, at month-end or year-end, when the counts come in.

    Inventory discrepancy agent

    Groups discrepancies between the count and book stock by likely cause and prepares draft adjustments.

    1. Collects counts from handhelds or count sheets and compares them with book stock on the same date.
    2. Searches recent movements for the typical causes: returns not posted, transfers between locations, mixed-up units of measure, similar items swapped.
    3. Groups discrepancies by cause and sorts the lines by value, so the manager starts with the ones that matter most.
    4. Prepares draft adjustments, each with an explanatory note, ready for review.

    Connects to: Counting handhelds · Management software or ERP · Count sheets

    Claude Opus
    It works a few times a year on long lists of movements and has to reason through several hypotheses for each discrepancy: here the most capable model is worth it.

    Who approves. Every adjustment is signed off by the warehouse manager or the accounts department; the agent explains discrepancies, it doesn't close them.

Where time gets lost today.

These are the points we hear about most often. Each one is a possible first agent.

The supplier's delivery note doesn't match the order

The warehouse worker counts the packages, then someone in the office checks the delivery note against the purchase order line by line. Supplier codes that differ from yours, mixed-up units of measure and partial deliveries only surface once the goods receipt has already been posted.

Before the goods receipt, the agent translates the supplier's codes into yours, compares quantities and prices with the open order and gives the warehouse worker a short list of discrepancies to check on the pallet.

Everyone asks the warehouse whether an item is in stock

Sales reps, the sales counter and customer support call or write to ask about availability, batch and the date of the next delivery. Whoever is in the aisle puts down the handheld to answer, or the question stays pending until evening.

Questions go to an internal chat where an agent reads stock, committed quantities and expected deliveries and answers with the figure from the management software, noting when it was last updated.

Minimum stock levels live in one person's head

Whoever handles purchasing knows by heart which items run out before the holidays or after a promotion. When they're away, reordering slips, and the stockout comes to light through a customer complaint.

Reorder rules are written down once, the agent applies them every week and proposes a draft order, explaining item by item why it suggests it.

Inventory ends with a list of unexplained discrepancies

After the count, hundreds of lines remain where physical and book stock don't match. They get adjusted in bulk to close on time, and the causes, like a return that was never posted, come back the next year.

An agent groups discrepancies by likely cause and links them to suspicious movements, so the manager adjusts knowing what they're correcting.

Which AI model we use, and for which task.

We don't tie you to a single provider: we work with Claude by Anthropic, GPT and Codex by OpenAI, and Gemini by Google, and we choose the version during testing, on your own cases.

GPT

  • Goods receiving agent. It extracts order lines from delivery notes with ever-changing layouts into a rigid format, which the connector checks before proposing the goods receipt.
  • Supplier confirmations agent. It reads price list tables and PDF confirmations from different suppliers and returns orderly lines that can be compared with orders.

Codex

  • Goods receiving agent. It writes the connector to the management software and the script that maintains the supplier code table; it doesn't read supplier documents.

Gemini Flash

  • Stock availability chat agent. Many questions a day, short answers: if the company works in Google Workspace, it integrates with the chat colleagues already use and keeps costs low.

Claude Haiku

  • Stock availability chat agent. It's the alternative we test on the same questions when the company uses a different chat, choosing whichever gets the item wrong least often.

Claude Sonnet

  • Supplier confirmations agent. It writes the alert to the sales rep clearly and without alarmism, distinguishing what's confirmed from what's only announced.
  • Reorder agent. It explains in a few readable lines why an item is in the proposal and recognizes when unusual consumption calls for a question rather than an order.

Gemini

  • Reorder agent. Useful when the stock rules live in a Google Sheet shared with purchasing: it reads and updates the sheet without manual steps.

Claude Opus

  • Inventory discrepancy agent. It works a few times a year on long lists of movements and has to reason through several hypotheses for each discrepancy: here the most capable model is worth it.

How we choose between models: the best AI model for agents.

A delivery note, a reorder, an inventory: what weighs on you most?

Tell us how delivery notes, availability questions and reorders come in today, and which management software or spreadsheet you record them in: we'll start from the point where goods most often sit waiting for a check.

Prefer the full form? Go to the contact page.

No spam or sales pitches. Response within 1 business day.

We use this data to contact you and, if we work together, to manage the project. We store it in Giallo's internal management system, on servers in the European Union, where only our team can see it; by email we only receive a notice, without your data. If you do not become a client, we delete it 3 months after our last contact. Details and your rights: privacy policy.

Do you prefer to write or call us?

How we work with you.

  1. A walk between receiving and the purchasing office

    We follow a delivery note from the truck to the goods receipt, and an availability question from the sales rep to the handheld, to see where the work stalls.

  2. Item master data before AI

    We check codes, units of measure and supplier cross-references: an agent is only as reliable as the master data it reads.

  3. Stock rules written with the buyer

    We turn the buyer's experience into readable rules, by item or by product family, that the agent applies and that can be changed.

  4. Testing on a closed inventory

    We run the agents on past counts and delivery notes whose outcome you know and compare their proposals with the adjustments actually made.

Frequently asked questions, straight answers.

Yes, for a small warehouse with few users you can start from an Excel or Google Sheets file: the agent reads and updates the file following precise rules. When several people move goods at the same time, it's better to switch to warehouse management software, even custom-built, with Excel only for analysis.

No. AI for warehouse management connects to what you already have, ERP, management software or shared spreadsheets, through an API or exports if the software offers them. A WMS becomes useful when you have many locations and picking routes, regardless of the agents.

No, it prepares a draft with its reasoning. Quantities come from fixed rules and formulas agreed with the buyer, and sending the order to the supplier remains a choice for the person who signs off purchases.

It's not the starting point we recommend. A reliable forecast requires clean history and well-organized master data; first we put receiving, reorders and inventories in order, then we assess together whether a forecasting model is worth the investment.

The goods receipt doesn't go through. The delivery note it reads becomes a draft compared with the order and checked by the warehouse worker at the pallet, so a reading error shows up as a discrepancy to check, not as wrong stock.

Yes, and it's one of the jobs where the agent helps most. It uses the cross-reference table in the management software and, when it finds a new code, proposes the most likely item and asks for confirmation before adding it to the table.

AI agents for warehouse management

Which moment of the warehouse day should we start from?

Tell us how delivery notes, availability questions and reorders come in today, and which management software or spreadsheet you record them in: we'll start from the point where goods most often sit waiting for a check.