AI in ERP: what actually works on the shop floor today

Every ERP vendor now claims AI. The useful question is not whether a product has AI, but which tasks it takes off someone's plate. Here is what is working in real factories and warehouses today, and what to be careful with.

Demand forecasting

SKU-level demand forecasts are one of the most reliable wins, because the inputs already exist in your order history. A good forecast does not need to be perfect. It needs to be better than the spreadsheet someone updates once a month, and it needs to refresh on its own.

• Works best with at least a year of order history
• Should show the history it is based on, so planners can judge it
• Should feed straight into reorder suggestions, not a separate report

Reading documents

Turning vendor invoices, delivery notes and purchase orders into entries saves hours of typing every week. The AI reads the PDF or photo, matches it to the right vendor and purchase order, and a person checks and posts it.

Catching anomalies

Duplicate bills, unusual payments, a stock adjustment far outside the normal range, scrap that suddenly doubles on one machine. These are hard for people to spot in a busy month and easy for software to flag.

Asking questions in plain language

Being able to ask which orders will ship late this week, and get an answer with the records behind it, is genuinely useful. It works when the assistant sits on top of one connected database. It struggles when your data is scattered across tools.

If the AI cannot show you which records it used, you should not act on what it says.

Keep humans in the loop

Good AI drafts and suggests. People should still approve anything that moves money or stock, and every suggestion and approval should be logged. That is not a limitation, it is what makes teams comfortable using it every day.