How AI is changing the ERP conversation
ERP is shifting from a system of record to a system of recommendations. That only helps if the record is already trusted.
For years the ERP conversation was about recording: did the order, the receipt and the invoice land in the same place? That question is not finished. A new one now sits on top of it. Can the system suggest what to do next?
Recommendations need a record
Forecasting demand, flagging a margin that does not match the price list, drafting a collection email, proposing a reorder: these are real uses. Every one of them reads the transactions you already store. If those transactions are late, duplicated or coded differently in each branch, the suggestion will be confidently wrong.
AI does not average out a messy foundation. It scales it. A bad item master becomes a bad forecast at the speed of a demo.
Ask a harder question than "does it have AI?"
Ask what decision changes. Ask which data it reads, who is allowed to see that data, and who signs the action if the suggestion is wrong. Ask for the use case on your numbers, not on a sample company in the slide.
A copilot that writes a summary of a report you do not trust is a faster way to circulate a bad number. A copilot that highlights three orders with a broken promise, and leaves the call to a person, is a tool.
The sequence
Clean the record. Name the decision. Then let the model assist. Companies that reverse this order spend a year proving that AI "does not work here", when what did not work was the data they asked it to believe.