AI for ERP: how to use AI with the ERP you already have

You can use AI with the ERP your company already has, without switching systems. There are three paths: your ERP vendor’s own AI; an integration between the ERP and an AI assistant, through an API, a connector or MCP; or an AI that works with the ERP as it is and does the chasing for the company. In all three, three controls stay the same: clear permissions, clear rules about what the AI does on its own and a record of what it did.

This is a practical guide for people who run operations. I build Hey Will, which is one of those paths, so I’ll keep what applies to any tool separate from what’s specific to Will. And I won’t review any specific ERP’s AI: every vendor moves fast, and the best source is the vendor itself.

What can AI do with your ERP data?

Your ERP holds what the company sold, bought, has in stock and owes. With access to that data, an AI can help with five jobs:

• Answer questions in plain language, like "has order #1842 shipped?", "how many linen pillow covers do we have?" or "which orders have been open for more than a week?"
• Find stalled orders before the complaint comes in: what was sold and hasn’t shipped, since when and what’s holding each one up.
• Reply to customers with what’s in the ERP: where the order is, what was delayed and the new date.
• Cross-check the ERP against other sources: the order against the supplier’s promise in Gmail, or the purchase against the ClickUp card of whoever owns it.
• Reconcile: match what the ERP says against the bank statement through Open Finance (regulated by Brazil’s Central Bank), clear what matches and set aside what doesn’t.

The examples come from a fictional store, Lume Casa, which runs on Bling. Notice that the work is reading, cross-checking and doing: replying, chasing, updating a date, clearing a payment. It’s the running around that fills someone’s day on your team right now.

What are the ways to connect AI to an ERP?

In practice, there are three paths, and they can coexist.

1. Your ERP vendor’s own AI

When it exists, it’s the shortest path: the AI comes built into the system and knows its data. It’s worth asking your ERP vendor whether it has an AI assistant, what that assistant reads, what it can change, how much it costs and whether your data is used to train models.

Also ask whether that AI can see what’s outside the ERP, like the email with the supplier’s promise and the task card. If it can’t, that part of the running around stays with your team.

2. An integration: API, connector or MCP

Here, someone connects the ERP to an AI assistant such as ChatGPT, Claude or Gemini. The connection can go through the ERP’s API, a ready-made connector or MCP. The official Model Context Protocol site describes MCP as an open standard for connecting AI applications to external systems, a kind of USB-C port for AI. The official ChatGPT Business and Claude pages also talk about plugins and connectors built for a company’s own systems.

It’s a flexible path, and you choose the AI. The effort is in building and maintaining it: someone has to set up the connection, decide what the AI can read and do, and look after it when the ERP or the AI changes. It’s worth asking your ERP vendor whether it has an open API, ready-made connectors or an MCP server.

3. An AI that works with your ERP as it is

The third path is to leave what works alone: a right hand that works with the ERP, the bank, email and the tools the company already uses, cross-checks the sources, chases whatever stalled and tells you what it did. That’s what Will does.

Will works with Bling, Omie, Tiny, TOTVS, SAP and other ERPs, and answers questions about orders, products and inventory with the source for every answer. Search looks for an order, a customer or a supplier across the ERP, Gmail and ClickUp at once. And every stalled order becomes a task with a plan that Will sees through to the end: it chases the supplier, updates the date in the ERP and replies to the customer.

If your ERP isn’t on that list, that’s fine: if the system has a login, Will uses it too, like anyone on the team.

What are the risks of using AI with your ERP?

The ERP is where the company keeps everything that’s worth money. Before giving any AI access, agree on four controls:

• The right permissions. The AI accesses only what it needs for the job, and each person sees only what they’re allowed to see.
• Clear rules. The company defines what the AI does on its own and what stays with the team, such as pricing, discounts and cancellations.
• A record of everything. Who asked, what was done, in which system and when.
• A source for every answer. The AI shows where each number came from and says what it couldn’t see, instead of making things up.

What about the LGPD?

Your ERP holds customers’ personal data: name, address, CPF (individual taxpayer ID) and purchase history. Under the LGPD, Law No. 13,709/2018, any operation involving personal data, access and processing included, counts as data processing. Among other principles, the law requires processing to be limited to the minimum necessary and security measures to be in place against unauthorized access. It also requires a record of processing operations.

In practice: bring in whoever handles the LGPD at your company, ask the AI vendor where the data is stored and whether it’s used to train models, and keep a record of what the AI accesses and does. This isn’t legal advice. When in doubt, talk to whoever handles this at your company.

How to choose an AI for your ERP: a checklist

• What does it read in the ERP: orders, inventory, finances, customers? With what access?
• Does it only read, or does it also act: reply, chase and update the ERP?
• Is there a record of who asked and what was done, in which system and when?
• Does it show the source of every answer and say what it didn’t see?
• Can it see what’s outside the ERP, like email and tasks?
• Is company data used to train models? Where is it stored?
• Who looks after the connection when the ERP or the AI changes?
• Does it work for one person or for the whole team?

To start, pick one job your team does every day, like answering "Where’s my order?", and let the AI do it for two weeks. If it gets it right and shows the source, hand it the next job.

AI doesn’t need a new ERP. It needs clear permissions, clear rules and a record of what it did.

To see how Will handles sources and records, read Sources and receipts and Security and LGPD. If your question is whether to let AI write to customers and suppliers, Can AI send email on behalf of your company? explains how authorization works. And to see which AI tools connect to what, read the best AI tools for business.

Sources, checked on September 27, 2026: the official Model Context Protocol site (modelcontextprotocol.io); Law No. 13,709/2018, the LGPD, on the Brazilian government’s Planalto site (planalto.gov.br); OpenAI’s ChatGPT Business page (chatgpt.com/business); and Anthropic’s Claude overview (claude.com/product/overview).