SynthesisLogic

AI agent for order processing

An agent that reads orders arriving by email, PDF, portal or online store, checks them against your prices and stock, enters the clean ones in your system and sends the odd ones to your order desk.

Book an intro call How it works
A morning with this agent
  1. 27 orders received from email, messages and the online storeDone
  2. 24 orders entered and sent to the warehouseDone
  3. One order asks for more than the stock on handAsks a person
  4. 2 orders over a customer's credit limit wait for approvalAwaiting approval

An example. Your agent is built around your own work.

How it works

  1. Collect the orders

    Emails with PDF or Excel orders, portal orders, EDI files and online store orders are gathered into one queue. The original file is always kept next to the order.

  2. Read and match

    The agent reads the customer reference, ship-to address, requested date and each line's item code, description and quantity. It matches the customer and items to your records, including old product codes the customer still uses.

  3. Check

    It checks the price against your price list or contract, the stock, the customer's credit limit, the minimum order quantity, and whether the same order has already come in.

  4. Enter or hand over

    A clean order becomes a sales order in your system and the customer gets an acknowledgement. Anything that fails a check goes to the order desk with the source document beside the proposed order.

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.

  1. Customer(email from a customer, with a PDF) Please find our PO 88213 attached. Delivery needed by Thursday.
  2. Read the PDF: 14 lines, ship-to Jebel Ali warehouse. Matched the customer and 13 of 14 items to the ERP.
  3. Checked prices against the contract list: all 13 match. Checked stock: 12 lines fully available, 1 line short by 20 units.
  4. AgentHello, we have received PO 88213. 12 lines are confirmed for delivery on Thursday. One line is short by 20 units; we will confirm the date for those shortly. You will receive the full acknowledgement within minutes.
  5. Created sales order 55102 with 12 lines. Held 1 line with a stock shortage and 1 line with an unknown item code for the order desk.
  6. Agent(to the order desk, on Teams) Order 55102 is ready except two lines. Line 9: item code HV-200-B does not exist; the description matches product 200-BLK, please confirm. Line 14: 20 units short, the next receipt is Friday. Your decision on splitting the shipment?
  7. Customer(order desk) HV-200-B is 200-BLK, add it to the alias list. Ship the rest Thursday and the short line Friday.
  8. Saved the alias. Released the order, set the second shipment for Friday, sent the final acknowledgement to the customer.
  9. AgentDone. Order 55102 ships Thursday, with the remaining line on Friday. The customer has been told about the split.

How it is built

How orders come in
The agent watches the order inbox through the mail system's API, picks up files from your supplier or customer portal and EDI folder, and reads online store orders through the platform's API. Where a customer portal has no API, it can use a browser with a dedicated, limited account.
Reading the document
Scanned and PDF orders are first read with OCR, then the language model pulls out the fields into a fixed form: customer reference, ship-to, date, and each line with item, quantity and unit. Anything it is not sure of is marked unsure and not filled in.
Matching to your master data
Customers and items are matched to your ERP records. Customers often use their own product codes, so a mapping table of aliases is kept, which your staff can maintain. A code the agent cannot resolve goes to a person and, once answered, is saved.
Your rules decide
Price against contract, credit limit, minimum quantity, stock and duplicate detection are rules, not model opinions. The agent never accepts a price override from the order itself.
A fixed list of actions
The agent can create a sales order, hold a line, send an acknowledgement, add a note and send a case to the order desk. It cannot change prices or terms, approve credit or cancel shipped orders.
Review, logging and tuning
Every order keeps its source channel, the extracted fields, the checks run and the outcome. The order desk can see why each order was held, and the weekly report shows how many orders went through without anyone touching them.
Cloud or your own servers
It can run in the cloud, or on your own servers, for example on the office or factory network. Then the agent, its database, its logs and the language model itself stay with you, using an open model such as Llama, Qwen or Mistral, so orders, prices and customer data stay inside. Order files already on your network can be handled without an internet connection, while the mail system, portals and EDI links still pass messages through those services. Local open models are smaller than the largest cloud models. For reading orders and extracting fields they are usually enough, and we test on your own orders first. The agent connects with permissions limited to order entry.

What you can connect to it

Microsoft Dynamics 365 Business Central
Looks up customers and items and creates sales orders through its API.
SAP S/4HANA or Business One
Reads business partners, prices and stock, and creates sales orders through OData services.
Odoo
Reads customers, products and stock, and creates sales orders through the external API.
NetSuite
Creates sales orders and checks customer and item records through its web services.
Shopify, WooCommerce or Magento
Reads new online orders and writes back status and tracking.
Salesforce or HubSpot
Matches the order to the account and deal, and records the order against it.
Outlook or Gmail
Reads the order inbox, saves the attached file with the order, and sends acknowledgements from your address.
Customer portals and EDI
Collects orders from portals and EDI files, and sends back confirmations in the format the customer expects.
Warehouse system or carrier tools
Passes the released order on for picking, and later brings back tracking numbers.
Excel or Google Sheets
If you have no ERP, it keeps the order log and price list in a spreadsheet.
Book an intro call

Connects to your tools. You stay in control.

Works with what you have
Email (Outlook, Gmail), PDF and Excel orders, EDI, customer portals, online store (Shopify, WooCommerce), ERP (Dynamics 365 Business Central, SAP, Odoo, NetSuite), CRM, warehouse system, spreadsheets.
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 change prices, discounts or terms. It uses your price lists and contracts, and anything outside them goes to your sales team.
  • It does not approve credit. An order over a customer's limit goes to the person who decides.
  • It does not guess an unclear item or quantity. It marks the line and asks.
  • It does not cancel or change an order that has already shipped.
  • A badly scanned or handwritten order may be sent to a person instead of being entered.

Questions

Can it read purchase orders sent as PDF?

Yes. It reads PDF, scanned and Excel orders, and sends anything unclear to a person instead of guessing.

Does it change prices or terms?

No. It uses your price lists and terms; exceptions go to your sales team.

How do you automate order entry?

The agent reads the order from the email, PDF or EDI file, runs the same checks a clerk would, and writes it to your ERP. A person handles only the orders that fail a check. It is best to begin with the channel that brings the most orders.

What if a customer sends the same order twice?

It compares customer, lines and dates against recent orders and holds a likely duplicate for a person. It also sends the acknowledgement quickly, because customers often resend when they hear nothing.

What about customers who use their own product codes?

The agent keeps an alias table that links their codes to yours. A new code it cannot resolve goes to your order desk once, and the answer is saved for next time.

Can it also handle fulfilment, routing and tracking?

Partly. After an order is accepted it can pass it to the warehouse system, apply your routing rules for which warehouse or supplier ships it, and bring carrier tracking events back to your system and the customer. It starts with one warehouse and one carrier; your warehouse system still runs the picking.

Do we need to change our ERP?

No. It works with the ERP you have, through its API. If your system has no API, it can work from exports or a browser with a limited account.

What happens when it is wrong?

A wrong line is caught by the checks or by the order desk before shipping, because each order shows the source document next to what was entered. Every decision is logged, and when we see a repeated mistake, we fix the rule or the alias table behind it.

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.

Type of organisation