AI agent for inventory management
An agent that reads your stock, sales and open orders every day, warns before an item runs out, and prepares the reorder as a draft purchase order for your buyer to approve.
- Stock and sales read for 2 warehouses and 6 branchesDone
- 11 items will run out before the next deliveryDone
- Reorder list drafted for 3 suppliersAwaiting approval
- Count mismatch on one item in branch 4Asks a person
An example. Your agent is built around your own work.
How it works
-
Read stock and sales
Each morning the agent pulls on-hand stock by location, open purchase orders, open sales orders and the last days of sales from your ERP, shop system or spreadsheets.
-
Work out what is running short
It compares how fast each item sells with the stock left and the supplier's lead time. Items that will run out before a new delivery could arrive go on the list, with the reason next to each one.
-
Draft the reorder
For each item it proposes a quantity and a date, respecting the supplier's minimum order and your budget limit, and creates a draft purchase order. Nothing is sent to a supplier at this point.
-
Your buyer decides
The buyer opens one daily list, approves, changes or rejects each line, and only then does the order go out. Items whose system count looks wrong are flagged for a recount instead of being ordered.
New to AI agents? How they work, in plain words
Real examples, setup steps and tool connections for this agent
An example conversation
An example of how a conversation runs. The grey lines show what the agent does in your systems meanwhile.
- CustomerGood morning. Anything I need to order today?
- Read stock for 2 warehouses (1,240 items), 38 open purchase orders and the last 60 days of sales from the ERP.
- AgentYes, 9 items will run out before their next delivery. The most urgent is the 12 mm hex bolt: 3 days of stock left, and the supplier takes 10 days. I have drafted orders for all 9.
- CustomerWhy 4,000 bolts? We usually order 2,000.
- Checked the last 8 weeks of sales and the open sales orders for this item.
- AgentSales of this bolt doubled in the last three weeks, and two open customer orders need 1,500 between them. 2,000 would cover about 12 days. If you think the rise is temporary, tell me a number and I will change the draft.
- CustomerMake it 3,000.
- Updated draft PO 7731 to 3,000 units. Total is under the buyer approval limit.
- AgentDone. Draft 7731 is ready for your approval. One more thing: the system shows 640 of item 4471 on the shelf, but there were no sales or receipts for 5 weeks and the last count was 2 months ago. I did not draft an order for it. Can someone count it?
How it is built
- Where the data comes from
- The agent reads stock, orders and sales from your ERP, inventory app or shop platform through its API, or from a daily export or spreadsheet when there is no API. It reads at least once a day, so every figure it quotes has a time stamp.
- Numbers are calculated, not guessed
- Demand per item and location is calculated from your sales history with ordinary forecasting methods, starting with a simple baseline and using something more advanced only where the baseline is wrong. The language model is not asked to guess quantities. It explains the proposal in plain words and answers the buyer's questions about it.
- Your rules set the limits
- Supplier lead times, minimum order quantities, a maximum value per order and a stock ceiling are rules we set up with you. A proposal that breaks a rule is blocked or flagged, never sent.
- A short list of actions
- The agent can read stock, calculate a reorder, create a draft purchase order, send a stock alert and ask for a recount. It cannot release an order above the approval limit, change a supplier's prices or edit stock levels.
- Stock that does not match the shelf
- System counts drift from what is on the shelf. The agent compares counts with recent movement, flags items with unexplained differences, and treats days with zero stock as missing sales, not as zero demand.
- Approval and logging
- The buyer approves purchase orders in a daily list. Every proposal is logged with the figures used, and each week the agent reports stockouts, excess stock and how far its forecasts were off, so the rules can be tuned.
- Cloud or your own servers
- It can run in the cloud, or on your own servers, for example on the warehouse or head office network. Then the agent, its database, its logs and the language model itself all sit with you, using an open model such as Llama, Qwen or Mistral, so stock, supplier and sales data stay inside. Reading your ERP and spreadsheets needs no internet connection. Anything that uses an outside service, such as sending an email to a supplier, still passes through that service. Local open models are smaller than the largest cloud models. For explaining a reorder and answering from your own figures they are usually enough, and we test on your own cases first. Access is set only for the systems it needs.
What you can connect to it
- ERP (SAP, Odoo, Microsoft Dynamics 365 Business Central, NetSuite)
- Reads stock by location, open purchase and sales orders, and item data, and creates draft purchase orders through the system's API.
- Shopify, WooCommerce or another online store
- Reads inventory levels and orders, so online sales count in the forecast and the shop stays in step with the warehouse.
- POS or retail system (Lightspeed and others)
- Reads daily sales per shop and per item, which is the main input for demand.
- Zoho Inventory or similar inventory app
- Reads items, stock and purchase orders, and creates draft orders through its API.
- QuickBooks or Xero
- Where your accounting software holds items and bills, it reads item cost and supplier bills, and drafts bills for goods received only if you ask for that.
- Google Sheets or Excel
- If stock lives in spreadsheets, it reads and updates them, and uses a sheet as the buyer's daily review list.
- Supplier price lists and emails
- Reads price lists and delivery confirmations sent by suppliers as PDF or Excel, and updates lead times from what was actually received.
- Email, Teams or Slack
- Sends the buyer the daily list and urgent stock alerts, and takes questions and changes in reply.
- Barcode scanners and count sheets
- Takes cycle count results from exports or sheets and compares them with system stock.
Connects to your tools. You stay in control.
- Works with what you have
- ERP (SAP, Odoo, Dynamics 365 Business Central, NetSuite), online store (Shopify, WooCommerce), POS, inventory apps, QuickBooks, Xero, spreadsheets, supplier emails and price lists.
- A person approves what matters
- Payments, orders, discounts or anything you choose wait for approval. Every action is logged.
- Can run on your own servers
- The agent, its data, its logs and even the language model (open models such as Llama, Qwen or Mistral) can run on your servers, so documents and customer data stay inside your network. How that works
What it does not do
- It does not send a purchase order to a supplier without approval, unless you decide that small routine reorders under a value you set may go out on their own.
- It does not change stock levels by itself. A difference between system and shelf is flagged for a person to count and correct.
- It does not negotiate prices or change suppliers.
- A forecast is an estimate. It cannot see a customer's one-off bulk order or a supply problem that nobody has recorded, so tell it, or the system, about those.
- It controls the stock records and orders, not the physical warehouse. Picking, putting away and moving goods stay with your people and your warehouse system.
Questions
Does it place orders with suppliers?
It drafts them. A person approves before anything is sent, unless you decide some routine reorders can go out on their own.
We keep stock in spreadsheets. Can it still work?
Yes. It can read and update spreadsheets, and many first projects start there. We look at how your sheets are laid out in the first session, and an ERP can come later.
How does it know when to reorder?
It compares how fast an item sells with the stock left and the supplier's lead time, and proposes an order when the stock will run out before a new delivery could arrive. The rules for lead time, minimum order and budget are yours, and each proposal shows the figures behind it.
Is this the same as AI inventory management software?
It overlaps, but a custom agent is built around your own systems, suppliers and approval rules instead of a fixed product screen. If you are comparing options, test any of them on three months of your own history and compare the forecast error with your current method.
Can it warn me about shortages before customers notice?
Yes. It compares stock with open orders and supplier lead times every day, and alerts the buyer with a suggested action: expedite, substitute an item, or move stock between locations. Many teams start with a daily report and add alerts once they trust the suggestions.
What if the stock count in the system is wrong?
It flags items whose count looks unreliable, for example no movement for weeks or a count that disagrees with sales, and asks for a recount instead of ordering against a wrong number. It cannot fix the shelf itself, but it can tell you where to look first.
Does it help with warehouse management?
It works on the data your warehouse system already holds: stock by bin, open orders and receipts. It can suggest replenishment, flag count mismatches and draft tasks, while your warehouse system keeps running the floor.
Do we have to change our software?
No. It connects to what you use now through an API, an export or a spreadsheet. If a system has none of these, we say so in the first session.
How much does an agent like this cost?
It depends on the task and the systems it connects to. After a first call we give you a clear estimate of cost and time.
Want this agent built around your work?
Tell us in a few lines what the job looks like today and we will set up an online intro call.
Request sent
We will contact you to set up an intro call.