AI built into Odoo — decisions, not chatbots

AI where it decides, not where it decorates.

Almost all the AI shown in demos is a chat panel bolted onto the side: impressive for ten minutes, and it changes exactly zero processes. The kind that works sits inside the flow, in the places where someone on your team spends hours on a mechanical task or makes a call by eye. That's where we put it: reading the documents that come in, anticipating demand before the warehouse feels it, adjusting prices by channel and moment, and answering customers with real ERP data instead of polite phrasing.

Documents that read themselves

Vendor invoices, delivery notes and receipts come in by email or as a photo and come out as drafts with amount, taxes and vendor already filled in. People review the exceptions, which is the only part that genuinely needs human judgment.

Demand forecasting

Models trained on your own history — sales, seasonality, each vendor's real lead times — that propose what to buy or manufacture, and when. The proposal still gets reviewed; what disappears is the guesswork.

Prices that react

Margins adjusted by channel, customer, volume and moment, with rules and models working on top of your commercial policy. You set the limits; the system decides inside them, a thousand times a day.

Agents with real data

Assistants wired into the ERP that answer about orders, stock or due dates by querying the database, not by inventing. An agent that doesn't know something says so, which is the only way you can put one in front of a customer.

Classify and route

Incoming email, issues and orders classified and assigned to the right team with the right priority. Work reaches whoever should handle it without passing through a shared inbox where everything ages at the same rate.

Inside your own house

You choose which data leaves and which doesn't, with models hosted wherever makes sense for your business. Useful AI doesn't demand that you hand your entire operation to a third party.

Questions we get asked

Is this a chatbot?

No, or at least not mainly. The biggest return comes from the quiet processes: reading invoices, proposing purchases, sorting email. Chat is just one of the ways to talk to the system, and the least decisive one.

Does it make up answers?

The agents we build query the real ERP data and answer from it; when there's no data, they say so. An assistant that fills gaps with imagination isn't useful: it's a risk in front of a customer.

Does our data end up training somebody else?

Only if you want it to. We define what information leaves the system, what gets processed inside and with which provider, and for sensitive cases we work with models hosted on our own infrastructure.

Where do we start?

With the process that eats the most mechanical hours, almost always vendor invoice entry. We measure the time saved over a real month and, with that number in hand, decide the next step.

AI where it decides, not where it decorates.