Why does my AI get Odoo revenue wrong?

Because "revenue" is not a column in Odoo, and the field whose name sounds closest, amount_total on account.move, means something else. account.move is the shared table for customer invoices, vendor bills, and credit notes, and amount_total is the document total including tax, in draft as well as posted. An AI that sums amount_total across that table adds vendor bills and unposted drafts to sales, counts tax as income, and never subtracts credit notes. It is not the model being unintelligent; it is that the field names are built for documents, not analytics, and nothing tells the model which is which. Verified revenue is posted customer invoices (move_type = out_invoice, state = posted), net of credit notes (out_refund), measured before tax.

It is the document total, including tax, which is not the same as revenue. amount_total on a customer invoice includes the tax you collect and remit, so it is not money you earned. The untaxed amount is closer to revenue for a single invoice. And amount_total exists on vendor bills and credit notes too, because they all live in account.move, so summing it across the table mixes money coming in with money going out.

Across account.move, distinguished by move_type: out_invoice is a customer invoice, in_invoice is a vendor bill, out_refund is a customer credit note, in_refund is a vendor refund. Only posted entries (state = posted) are real; drafts are not. So revenue for a period is the posted out_invoice entries, net of posted out_refund credit notes, measured before tax. Get any one of those wrong, the edition, the move type, the state, or the tax, and the number is wrong with total confidence.

Because it reaches for the field that reads like an owner and picks the wrong one. On account.move, the salesperson is invoice_user_id. partner_id is the customer, not the seller, and create_uid is whoever keyed the record in, often an accountant or an integration. An AI asked "how did each rep do" will happily group by partner_id or create_uid and hand you a confident, wrong leaderboard. The right field exists; the model just has no way to know which one carries the meaning you asked for.

No, and this is the important part. A more capable model reads the same ambiguous field names and makes the same reasonable, wrong guess, only more fluently. Intelligence is not the missing piece; verified knowledge of your business is. That is what a context layer provides: a checked mapping from what you asked ("revenue", "the salesperson", "on hand") to the exact fields, move types, and states that carry that meaning in Odoo 19. The founder wrote about the distinction directly: intelligence isn’t knowledge (nanti.ai/notes/intelligence-isnt-knowledge). The product that supplies the knowledge is the Context Layer for Odoo (nanti.ai/context-layer).