AI Readiness
AI for Customer Service in Odoo
Most support teams answer the same twenty questions over and over, and spend the rest of their time working out who should handle what. Both are countable, both are dull, and both are exactly what AI is good at.
Three workflows carry the value: sorting what comes in, drafting the replies that are nearly identical every time, and summarising a long thread so the next person does not have to read all of it. The retention argument sits on top, and we treat it as upside rather than as the case.
The three workflows we build
Scoped, quantified and approved before anything is built.
1. Ticket triage and routing
Reading each inbound ticket, deciding what it is about, how urgent it is and who should own it. Individually it takes a couple of minutes; across a few hundred tickets a month it is most of somebody’s week.
- Verbs: Classify, then Suggest
- Where a person confirms: Routing is suggested and an agent can override it. Auto-assignment only once you have watched it be right for long enough to trust it.
- What it needs from your data: Categories that mean something and a team structure Odoo knows about.
2. Response drafting from your knowledge base
Writing a near-identical reply to the same question for the four hundredth time. The answer is not hard; the typing is the cost.
- Verbs: Draft, from your own material
- Where a person confirms: The agent reads the draft, adjusts it and sends. Nothing goes to a customer unread.
- What it needs from your data: A knowledge base or a set of past replies good enough to draft from. If your answers only exist in people’s heads, capturing them is the first job — and it is worth doing regardless.
3. Escalation summaries and handover
A ticket gets escalated and the next person reads six weeks of back-and-forth to work out where things stand. Every handover pays this cost, and the customer pays it too by repeating themselves.
- Verbs: Summarise
- Where a person confirms: The summary is a starting point, not a substitute for the thread — the full history stays one click away.
- What it needs from your data: Ticket history actually recorded in Odoo rather than scattered across inboxes.
The chatbot question, answered up front
Most people hear "AI in customer service" and picture a bot that stands between the customer and a human being. That is not what this is, and we would argue it is usually a mistake.
Everything here sits behind your team, not in front of your customers. The AI drafts, sorts and summarises; your people read, adjust and send. The customer still deals with a person, and that person now has more time to actually be useful because they are not retyping the same paragraph or catching up on a thread.
If you want a customer-facing bot, that is a different conversation and a different risk profile, and it should be taken deliberately rather than arrived at by accident.
Where the value comes from
Two kinds, and we are careful about the weight we put on each.
Time, which is countable
Tickets per month, times minutes per ticket, times your loaded service cost. Triage, drafting and escalation catch-up are all measurable today and measurable again afterwards. This is what we ask you to decide on.
Customers kept, which is arguable
Faster, more consistent service does reduce churn — that is not in dispute. But attributing a specific customer's decision to stay to a specific improvement in response time is not something we can prove to you, so we show it, we explain how we would estimate it, and then we leave it out of the payback.
The same rule runs through the sales program: lead with what can be counted, show the rest as upside. If the case needs the upside to work, it is not a case yet.
Common questions
Will customers know they are getting an AI-drafted reply?
They are not, strictly speaking — they are getting a reply your agent read, adjusted and sent. The draft is a starting point, the same as a template is. What matters is that a person took responsibility for what went out, and one always does.
What if it drafts something wrong?
Your agent catches it, the same way they would catch a bad template. That is what the review step is for. Over time you will see which categories it handles well and which it does not, and the scope narrows to where it earns its place.
We do not have a knowledge base. Can we still do this?
Response drafting needs source material, but it does not have to be a formal knowledge base — a body of past replies is often enough to start from. And if the answers genuinely only live in people's heads, capturing them is worth doing whether or not you ever automate anything. It is the single most fragile thing in most support teams.
Does this mean fewer support staff?
That is your decision, not a consequence of the build. What it reliably produces is capacity: the same team handling more, with less of the day spent on the parts that require no judgement. Most businesses we talk to have a backlog of things support could be doing and has never had time for.
Which workflow should we start with?
Usually triage, because it is the highest volume and the lowest risk — getting a routing suggestion wrong costs somebody thirty seconds. Response drafting is the bigger prize and the better second step.
Start with triage
Highest volume, lowest risk, and the easiest place to see whether this works in your business. A short call is enough to size it.