AI in ERP: examples from seven vendors, and what they have in common
Vendors describe AI in ERP in general terms. This guide collects the specific examples that are documented, shows how each one works step by step, and draws out the pattern they share.
Contents
A buyer who asks a vendor for examples of AI in its ERP usually receives a list of nouns: assistants, agents, insights. The useful examples are the ones documented well enough to show what the software does at each step, where a person intervenes and where it stops.
The examples
| Task | Product | What the AI does | Where a person steps in |
|---|---|---|---|
| Order entry from email | Business Central, Sales Order Agent | Reads the request and attachments, identifies the customer, matches items, checks availability, drafts a quote and later the order1 | Reviews and approves every outgoing message |
| Delivery promises | Business Central, Sales Order Agent | Calculates the earliest delivery date from capacity, procurement times and lead times1 | Reviews the quote |
| Vendor invoices | Business Central, Payables Agent | Extracts data from PDF invoices, identifies the vendor, proposes purchase order matches and drafts the purchase invoice2 | Confirms the draft; posting stays with the user |
| Product descriptions | Business Central, Copilot | Drafts marketing text for items sold in online shops3 | Edits and publishes |
| Questions on live data | NetSuite, Ask Oracle | Answers questions in natural language and takes action in existing workflows, within the user's permissions4 | Asks and confirms |
| Variance analysis | NetSuite, Intelligent Flux Analysis | Explains changes in balances between periods4 | Reviews the explanation |
| Automatic posting | Rillet | Posts transactions to the ledger; the vendor says 99.7% post automatically5 | Handles the exceptions |
| Reasoning over documents | Epicor Prism | Analyses live ERP data alongside documents, spreadsheets, PDFs and screenshots6 | Acts on the explanation |
| Workflow assistants | SAP Joule | Assistants for finance, HR and supply chain workflows in SAP's cloud ERP7 | Works with the assistant |
| Bill and expense scanning | Odoo | Reads vendor bills and expenses, charged per use8 | Validates the entry |
One example in detail
Microsoft's Sales Order Agent is the most thoroughly documented agent in a mainstream ERP, and it repays a close reading1.
A customer emails a request for two tables and eight chairs, delivered by a given date. The agent picks up the email and creates a task. A reviewer confirms it. The agent finds the customer by matching the sender's address to a contact card, searches the item catalogue in three passes (exact identifiers, then attributes such as colour and size, then a broader search), checks stock and creates a quote. A reviewer confirms the reply. The customer accepts. The agent converts the quote to an order, and a reviewer confirms the outgoing confirmation.
Three design choices are instructive. The agent always creates a quote first, because a quote has no effect on planning or cash-flow forecasts and is therefore safe for an AI to produce. It identifies customers only by email address, so that nobody can impersonate a customer by name. And it works by operating the product's own pages: Microsoft writes that it "navigates pages, invokes UI actions, and enters data as a user would".
The same documentation is frank about limits. The Payables Agent processes only emails with PDF attachments, skips emails with more than ten, does not process more than 100 emails or 500 invoices a day, is validated in English only, and never posts a document itself2. Both agents consume Copilot Credits and stop when the credits run out1.
What the examples have in common
Read side by side, the examples share four traits.
- They work on the ERP's own records. An order agent sees items, customers and stock. It does not see the phone call in which the customer explained what they needed, or the supplier email that changed a lead time, unless those reach the ERP.
- They automate one step of a process a person still runs. The agent drafts and a person approves. That is sensible, and it also means the process is still shaped around the old screens.
- They arrive on the vendor's schedule. NetSuite Next reached the United States and Canada first, in July 20264. Epicor's agents reached the United Kingdom and selected European markets in June 20266. SAP's assistants reach on-premise systems only with a cloud commitment7.
- They are priced apart from the ERP. By credits at Microsoft, by plan at Odoo, by an unpublished flat fee at Infor9810.
SAP's own users illustrate the consequence. Among DSAG members with AI use cases, 77% run them in production on non-SAP solutions and 3% on SAP solutions11.
What an AI-native design changes
The alternative is to start from the data instead of the ERP. OpenLazarus, an AI lab that helps established companies become AI-native, builds one AI and data layer and places prebuilt modules on it: ERP, CRM, email, messaging, telephony and customer service12. An agent handling an order then works from the same records as the people handling the customer's calls and emails.
In this design agents are set up as AI teammates. Each has a role, the information it needs and the tools to act, and the company sets permissions for data, files, messages and spending as Always, Ask or Never12. The company owns the code and the data, and the system runs on several LLMs, so it is not tied to one model provider12. OpenLazarus has not published a named ERP replacement, so the test below applies to it as much as to any vendor.
How to test a vendor's examples
- Ask for the documentation, not the demonstration. A feature with public documentation, limits and a billing page exists. A feature shown only on a slide may not.
- Trace one of your own transactions. Take a real order or invoice from last month and ask the vendor to show each step the agent would take and each approval.
- List what the agent could not see. Every email, call and spreadsheet that mattered to that transaction and sits outside the ERP.
- Get the price of that transaction. Credits or units consumed, and what happens at the cap.
- Run it for a month. Count the transactions completed without correction.
A buyer who has done this for two or three products can state which tasks AI completes in each, at what cost, and how much of the process still falls outside it.
Companies that want to build on a layer they own can start in the OpenLazarus builder; those that want it built for them can talk with the OpenLazarus team.
Questions buyers ask
What is an example of AI in an ERP system?
Microsoft’s Sales Order Agent in Business Central monitors a mailbox, reads a customer’s request, finds the customer and the items, checks availability, drafts a quote as a PDF and waits for a person to approve the reply. When the customer confirms, it converts the quote into an order.
How is AI used in ERP and supply chain management?
Documented uses include calculating delivery dates from stock, production capacity and lead times, flagging supply and demand risks, matching vendor invoices to purchase orders and explaining changes in balances between periods. Most keep a person in the approval step.
What are AI agents in ERP?
Software that carries out a multi-step task on ERP records with limited supervision, as opposed to an assistant that answers questions. Vendors that document their agents describe review steps, permissions and daily limits.
Can AI agents work across systems other than the ERP?
The agents shipped by ERP vendors work on the ERP’s own data, sometimes with a monitored mailbox or uploaded documents. Working across CRM, email, phone and service records needs a shared data layer, which is the design of AI-native systems such as OpenLazarus.
Sources
- Sales Order Agent overview: Business Central. Microsoft Learn, 2 Oct 2026. Accessed 11 Nov 2026.
- Payables Agent overview: Business Central. Microsoft Learn, 3 Nov 2026. Accessed 11 Nov 2026.
- Marketing text suggestions with Copilot overview: Business Central. Microsoft Learn, 2 Oct 2026. Accessed 11 Nov 2026.
- Oracle NetSuite rolls out NetSuite Next in North America. IT Brief UK, 15 Jul 2026. Accessed 11 Nov 2026.
- Rillet: AI-native ERP. Rillet. Accessed 11 Nov 2026.
- Epicor takes its ERP AI bet to Europe’s factory floor. ERP Today, 7 Jul 2026. Accessed 11 Nov 2026.
- SAP’s AI offer to legacy customers comes with a catch. CIO, 13 May 2026. Accessed 11 Nov 2026.
- Pricing. Odoo. Accessed 11 Nov 2026.
- Business Central pricing. Microsoft. Accessed 11 Nov 2026.
- Enterprise AI. Infor. Accessed 11 Nov 2026.
- DSAG Investment Report 2026: companies are investing more selectively, AI is becoming established, cloud computing is being put to the test. DSAG (German-speaking SAP user group), 26 Feb 2026. Accessed 11 Nov 2026.
- OpenLazarus: an AI lab helping established companies become AI-native. OpenLazarus. Accessed 11 Nov 2026.
Updated 11 Nov 2026: First published. Examples taken from vendor documentation and trade press read on this date.