SynthesisLogic

AI agent for your ERP system

An agent that works inside your ERP the way a careful clerk does: it enters orders and bills as drafts, checks them against your master data, chases missing information and answers questions about orders and stock.

Book an intro call How it works
A morning with this agent
  1. Read yesterday's transactions from the ERPDone
  2. 5 sales orders created from customer emailsAwaiting approval
  3. 2 items show negative stock in warehouse BAsks a person
  4. Exceptions report sent to operationsDone

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

How it works

  1. Read what arrives

    Orders, bills and requests come by email, PDF, Excel file, portal or EDI. The agent extracts the fields into the shape your ERP expects and keeps the source page for each value.

  2. Look up in the ERP

    It finds the customer, product, tax code and price list in your ERP. If there is no match, the record goes to a review queue and no new master data is created.

  3. Create a draft

    It applies your rules for discounts, credit limits and shipping, then creates the record as a draft, not a confirmed one, with a reference to the source document.

  4. A person confirms

    A sales or finance user confirms the draft. The agent also answers questions about order status and stock, and prepares the reports your team asks for.

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. CustomerMorning. The Northgate Dental order came in as a PDF at 8:10. Can you enter it?
  2. Read the PDF order from Northgate Dental: 14 lines. Matched the customer by tax ID. Matched 13 products by SKU in the ERP.
  3. Applied the customer's discount tier and the free-shipping rule for orders over the limit. Checked credit limit: within it.
  4. AgentI created sales order draft SO-20418 with 13 lines. One line, "scaler tip, size B", has no matching product, so I left it off and did not create a new item.
  5. CustomerThat's SKU ST-02B, they renamed it. Add it.
  6. Searched the product list for ST-02B. Found it under the name "Scaler tip B". Added the line to the draft and kept a note of the old name.
  7. AgentAdded. The draft total is 4,310.00 and delivery is set to Thursday. It stays a draft until you confirm it.
  8. CustomerConfirmed. And what is the stock on SKU ST-02B now?
  9. Read stock for ST-02B across both warehouses from the ERP.
  10. AgentThere are 120 in the main warehouse and 15 in the branch, so the order can ship from the main warehouse.

How it is built

Start with one workflow
We choose one object and one job first, such as sales orders from email, vendor bills or product master data, and test on real records including the messy ones, before anything is switched on. Time the current process for a couple of weeks first so you have your own baseline.
How the agent reaches the ERP
Through the ERP's API where one exists, such as Odoo's external API, SAP's OData services, NetSuite's REST services or Business Central's API. Where there is none, it uses exports and imports, or the web screens in a browser with a limited account.
A dedicated user with limited rights
The agent signs in as its own integration user, with rights only to the objects in scope and without administrator access. It cannot change your ERP setup, and everything it does is logged under that user.
Model reads, code decides
A language model reads emails and documents and fills the fields. Discounts, credit limits, tax codes and shipping methods are applied by rules in code, and each rule applied is logged.
Drafts, never final postings
Records the agent creates stay as drafts or unconfirmed documents. Only named users can confirm. A stored external reference stops the same document from being entered twice if a step is retried.
No guessing at master data
Customers are matched on tax ID and products on SKU before any record is created. Records with no match go to review, so you do not end up with the same customer or product twice.
Monitoring and hosting
We track the reject rate, how many manual edits each record needs, and the time from receipt to confirmation, and review them with you each month. The whole agent can run on your own servers: the agent, its database, its logs and the language model itself, using open models such as Llama, Qwen or Mistral, so documents never leave your network. Open models you host yourself are smaller than the largest cloud models. For reading orders, matching records and answering from your ERP they are usually enough, and we test on your own cases before choosing. Order emails and portals still pass through the services that carry them.

What you can connect to it

Odoo
Searches and creates partners, products, sales orders and vendor bills through its external API.
SAP S/4HANA
Reads and creates sales and purchasing documents through its OData and REST APIs.
NetSuite
Reads and creates records and runs saved searches through its REST web services.
Microsoft Dynamics 365 Business Central
Reads customers and items and creates sales documents through its API.
ERPNext
Reads and creates records of any document type through its REST API.
Oracle Fusion Cloud ERP
Creates suppliers, invoices and purchase orders in an unapproved state through its REST services.
QuickBooks Online
Handles invoices, bills and customers where accounting is kept outside the ERP.
Shopify
Reads orders and products through its Admin API when the ERP sits behind an online shop.
Gmail or Outlook
Receives order and bill emails and sends confirmations or questions.
Excel files and EDI
Reads order files that customers send in their own format.
Book an intro call

Connects to your tools. You stay in control.

Works with what you have
Odoo, SAP, NetSuite, Microsoft Dynamics 365 Business Central, ERPNext, Oracle Fusion Cloud ERP, QuickBooks Online, Shopify, Gmail, Outlook, Excel files, EDI.
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 confirm or post to the ledger on its own. Records stay as drafts until a named user confirms.
  • It does not create new customers, products or suppliers when there is no match. It sends them to review.
  • It does not change ERP settings, users or rights.
  • It does not guess a field it cannot read. It asks, or leaves the record for a person.
  • It does not use an administrator login.

Questions

Which ERPs can it work with?

Any ERP that has an API, data exports or web screens. Odoo, SAP, NetSuite, Business Central, ERPNext and Oracle all offer APIs. We check what yours offers in the first session.

Will it change our ERP setup?

No. It works as a user with limited rights inside your existing ERP, and every action is logged. It cannot change settings or other users.

Can it enter orders from email or PDF into the ERP?

Yes. It reads the order from the email, PDF or Excel file, looks up the customer and products, applies your discount and shipping rules and creates a draft sales order. A person confirms it.

What if it creates a wrong record?

It creates drafts, so a wrong one is corrected before it counts, and every field can be traced back to the source page. Records that do not match your data go to a review queue. Repeated mistakes lead us to fix the rule behind them.

Is it an AI ERP, or does it work with our current one?

It works with your current one. Some ERP vendors add their own AI features. Before you buy any of them, ask which objects the agent can write to, whether it writes drafts or final postings, and what the log records.

Can it answer questions about orders and stock?

Yes. Staff ask by email or message, and it reads the answer live from the ERP and shows where it came from.

How do we know it is worth doing?

We time the current process before starting, then measure the reject rate and the manual edits per record in the first months. You compare against your own figures.

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