AI agents that answer with data from your ERP and keep a record
An agent on WhatsApp answers the questions your system already has the answer to, logs each conversation on the customer record and hands over to a person whatever it should not resolve on its own.
Agents connected to Odoo 17 to 20 or to your current system, with rules written by your team
Dmunoz Group AgentOrder assistant · online
Hi, has my order 1045 shipped?
Yes. It left today at 10:20 and arrives tomorrow before noon.
Could you give me a discount?
Discounts are approved by sales. I am passing you to a team member in this same chat.
Can you resend invoice F001-2381?
Here it is as a PDF. It is due on 15/11.
Inquiry logged on the customer
Your team answers the same five questions every day
They arrive on WhatsApp and by email. The answer is in the ERP, but someone has to look it up, copy it and send it, and the conversation is not recorded anywhere.
- Has my order shipped?
- Can you resend the invoice?
- Do you have this product in stock?
- How much do I owe to date?
- What are your Saturday hours?
What staying this way costs
Answers that depend on who replies
Two people give two different dates for the same order, and the customer remembers the worse one.
Conversations that never reach any system
What was promised on WhatsApp lives on one salesperson's phone. If they leave, it leaves with them.
Sales and collections busy with status questions
The team's hours go into looking up data that is already in the ERP instead of selling or collecting.
Diego writes the agent's limits before switching it on
An agent makes things up when it answers without data or without a rule. DMG-RT, Diego's own method, starts from an indicator measured before touching the technology; in this service, first-response time and the share of questions solved without a hand-over.
Diagnosis: we read your real conversations
Diego takes a sample of recent WhatsApp and email messages, groups them by type of question and marks which ones have the data in your ERP. He measures first-response time and how many questions are solved without passing them to someone else.
Deliverable: Inventory of questions, findings letter and baseline, in 15 business days.
Design: we write what it answers and what it hands over
With your team we pick a single case to start with. We define which data the agent answers from, in what tone, which topics always go to a person and to whom. The agent only sees the data that case requires.
Deliverable: Hand-over rules, approved texts and indicator sheet.
Implementation and control: a pilot connected to your ERP
The agent goes live with that case. Every two weeks Diego reviews the conversations with your team, fixes the rule or text that failed and records the change. The scope grows once the indicator holds.
Deliverable: Agent in operation and a panel of answered and handed-over conversations.
Handover: your team runs it
An agent owner is named with a deputy, along with a manual for reviewing conversations and changing texts, and reviews at 3, 6 and 12 months using the same baseline measurement.
Deliverable: Closing record, operating manual and credentials in your company's name.
- Channel: WhatsApp, email or website chat
- Agent: written rules, tone and limits
- ERP: orders, invoices, stock and log
The agent answers only with data and hands over everything else
The usual question is whether the AI will make things up. In this design, every answer comes from an ERP record or from a text your team approved. When there is no data or the topic calls for judgment, the conversation goes to a person and is logged.
| Question | Data used | What the agent does |
|---|---|---|
| Order status | Sales order and delivery | Answers Answers with the dispatch date and time |
| Invoice resend | Posted invoice | Answers Sends it to the customer's registered email |
| Product stock | Inventory by warehouse | Answers Answers available or not, without quantities if so decided |
| Outstanding balance | Customer statement | Answers Only to the customer's registered number or email |
| Discount or special price | No approval rule | Hands over Hands over to sales in the same chat |
| Complaint or return | Service ticket | Hands over Creates the ticket and assigns it to a person |
| Question with no data in the ERP | None | Hands over Does not answer: hands over and logs it |
Screenshots from a standard Odoo 20 demo instance, with fictitious data.
Bring the question you get most often and review with Diego whether an agent can answer it.
Book a 30-min diagnosisWhat you provide and what you receive
You provide
- A sample of real WhatsApp or email conversations.
- Read access to your ERP for the data of the chosen case.
- One person who approves the texts and the hand-over rules.
- The WhatsApp Business number or the channel where the agent will run.
You receive
- An agent in operation for one well-defined case.
- Every conversation logged on the customer or the order.
- Written hand-over rules and a panel of what was answered and handed over.
- Operating manual, credentials and configuration in your company's name.
When an agent is not a fit
If the answers are not in any system and depend on what one person remembers, the agent has nothing to read from. In that case it is better to first record orders, invoices and stock in a single ERP. If you receive few questions a day, WhatsApp Business quick replies may be enough. See ERP with Odoo
Diego designs the service process, then connects the agent
Diego Muñoz is a process and systems consultant. In his own support work he uses agents connected to Odoo that read the record, draft the reply and leave the note. That practice is the basis of what he proposes: one well-defined case, backed by data and by a responsible person.
He chooses the AI model and the messaging service for each case and explains in writing which service processes each message. Code, credentials and texts stay in your company's name.
- Odoo 17, 18, 19 and 20
- Agents connected to ERPs
- Enterprise architecture with TOGAF
Questions before putting an agent in front of your customers
Can the agent make up answers?
It is configured to answer only with data from your ERP or with texts your team approved. If it cannot find the data, it does not answer: it passes the conversation to a person and logs it.
Where does my customers' data stay?
In your ERP and on the infrastructure agreed with you. Before the agent goes live you receive in writing which service processes each message and which data the agent can see.
Do I need Odoo to have an agent?
No. The agent connects to any system with an API or an accessible database. With Odoo the connection is direct and the record history already exists.
How much does it cost to run?
There is a usage cost for the AI model, another for the messaging channel and one for support. They are estimated from the real message volume measured in the diagnosis.
How is the project paid?
A fixed fee per phase, paid when each milestone is delivered. At the end of each phase you decide whether to continue.
What happens if the agent gets it wrong?
The conversation is logged. At the fortnightly review the failing rule or text is corrected and the change is recorded with its date.
What happens if I decide to switch the agent off?
It is switched off in the channel settings and the conversations go back to your team. The history in the ERP, the approved texts and the credentials stay in your company's name.
Review your first agent case with Diego in 30 minutes
- You bring the question your company gets most often.
- Diego checks whether your system has the data and what should be handed over.
- You receive in writing the first case for an agent, or the reason not to build one yet.
