Industrial companies have invested years in their ERP: configuration, customizations, item masters, pricing rules, and the habits of the people who use it every day. When AI enters the conversation, some vendors suggest that the path forward is a new platform. For most companies, that is the wrong trade. An ERP migration is one of the riskiest projects a business can take on, and it is not what stands between you and automation.
What actually needs to connect
The work AI can take on usually starts outside the ERP, in email, PDFs, spreadsheets, and shared drives, and ends inside it, as an order, a quote, an update, or a record. The integration layer needs to read reference data from the ERP (customers, items, prices, inventory, open orders) and write prepared transactions back into it.
Most ERPs used by distributors and manufacturers, including NetSuite, Epicor, Infor, Microsoft Dynamics, Acumatica, and SAP, offer APIs or integration tools that make this possible. Where an API is limited, there are usually import routines or a middleware layer that can be used safely. Each system is assessed on its own terms.
Writing with the same controls as a person
The principle that keeps this safe is simple: the workflow should write into the ERP with the same permissions, validations, and audit trail as the employee it supports. If a person cannot override a credit hold, neither can the workflow. If a transaction needs approval above a threshold, the workflow submits it for approval rather than posting it.
This keeps your existing controls meaningful and makes the automation auditable. Anyone looking at a record can see what was prepared by the workflow and who approved it.
When the ERP is the problem
Occasionally an assessment reveals that the system of record really is the constraint: data that is too inconsistent to match against, or a platform that cannot be integrated at all. When that happens, it is better to say so plainly and fix the data or plan the migration on its own merits, rather than hide the problem underneath an AI project.
See how this applies in practice: industrial AI solutions.
