Some stories explain better than any pitch why it's worth integrating AI into the real system a business runs on, instead of bolting a demo on top of it. This is one of them. It's a real case, anonymized for confidentiality: an industrial distributor with more than 50,000 products in its catalog.
The problem: quoting well, across a huge catalog, took half an hour
At a distributor with that many SKUs, quoting isn't just looking up a price. Every quote requires checking several things at once:
- Who the customer is and which specific price list applies to them (not everyone pays the same).
- Whether the requested product is in stock, and in which warehouse.
- Whether a discount or a special condition agreed on earlier needs to apply.
- Whether, once closed, it needs to be logged in the ERP as an opportunity and, if it goes through, as an order.
Before, that whole process went through a person on the sales team. They'd receive the request (often over WhatsApp, since that's how most of their customers communicate), open the ERP, look up the customer, check the price list, check stock on another screen, calculate the final price, and reply. When the product wasn't a common one or the pricing had nuances, it took even longer.
On average, each quote took about 30 minutes of human work. Multiplied across the daily volume of requests, that was a huge chunk of the sales team's day spent on mechanical, repetitive tasks instead of selling.
Why "just adding a chatbot" wasn't enough
The easy temptation, and where most attempts like this fail, is plugging in a generic chatbot that answers from what it "knows" about a static catalog. It works in the demo. The moment a real customer with a special price list shows up, or a product whose price changed yesterday, the chatbot either makes something up or goes silent. And in distribution, a wrong price isn't a minor detail, it's money.
That's why the approach here was different from the start: the AI wasn't going to "know" the catalog from memory, it was going to query it and operate inside the ERP in real time, the same way a person on the team would. That means every answer goes through the same checks a human salesperson would apply, not a statistical guess at what's probably correct.
What we built
The result is an AI sales agent that handles WhatsApp and that, inside Odoo, follows the exact same path a person used to follow:
- Validates the customer against the ERP, not against an assumption.
- Applies the real price list that specific customer has.
- Checks stock for the requested product in real time.
- Calculates and replies with the correct quote.
- Creates the opportunity and, where applicable, the order, as a real record in the system, not in a separate spreadsheet someone has to copy over later.
That's the difference between an assistant that "talks about" your business and one that works inside it. If you want the technical detail on how a piece like this gets built (the state machine, the API tools, the testing before it goes to production), we walk through it step by step in how to connect AI to your ERP without losing your mind.
The result: from 30 minutes to minutes
The change shows up mainly in two places: time and availability.
| Before | Now | |
|---|---|---|
| Time per quote | About 30 minutes on average | A matter of minutes, often instant |
| Who answers | A person on the sales team | The AI, applying the same criteria and the same checks |
| Hours covered | Business hours only | Whenever the customer writes, including off hours |
| Where it gets recorded | Sometimes only in the memory of whoever answered, or logged late in the ERP | Directly in the ERP as a real opportunity or order |
| Sales team's focus | Split between quoting and selling | Focused on negotiating and closing, not typing |
The sales team hasn't lost work, it's changed. They still step in on quotes that need a human call (a special negotiation, a pricing exception), but they've stopped losing half an hour on the ones that don't. That's the part these projects rarely get credit for: it's not about "replacing people," it's about giving them back the time for the part of the job where they actually add value.
What we learned (and it applies beyond this case)
- Generative AI on its own isn't enough when real business rules sit behind it. Price lists, stock, and special conditions can't be guessed, they have to be checked against the real system.
- The channel matters as much as the engine. If your customers talk over WhatsApp, the AI needs to be where they already are, not force them into a new portal.
- Every answer has to leave a trail in the ERP. If the AI "knows" something but doesn't write it where the rest of your business lives, you've just created another silo, not solved one.
- Real time savings show up when the AI handles the whole path, not just the first reply. Quoting fast but then having to manually push the order into another system leaves the problem half solved.
Is your team losing time quoting or answering the same thing over and over?
This is just one of the projects where AI actually works inside a client's system, not next to it. You can see this full case, with more context, at AI sales agent in the ERP, and the rest of our real projects at success stories. If you want the general framework before diving into a project like this, what it really means to integrate AI is a good place to start.
And if this sounds like the same time drain your team deals with, tell us about your case. We'll tell you, no strings attached, whether it makes sense, and what we'd build and what we wouldn't.
