Odoo 20 was released on 24 September at the Odoo Experience event in Brussels. As an Odoo Official Partner we went through the official release material line by line, and compared it against projects we are currently running.
This is not a feature list. It covers three things that affect day-to-day operations, and what to check before you upgrade.
1. Light User: split roles first, then decide who gets a light account
For small and mid-sized companies the blocker is rarely functionality. It is the cost of accounts. If sixty people on the shop floor need to scan and report time, giving everyone a full account usually kills the budget.
Odoo 20 adds Light User, billed monthly with a noticeably lower unit price than a full account. It is aimed at frontline operations.
The permissions behind a light account are limited, and the official documentation is brief on this point. In practice the split below works:
| Role | Suggested account | Why |
|---|---|---|
| Owner, finance, purchasing, sales | Full account | Reports, approvals, cross-module work |
| Shop-floor reporting, warehouse scanning | Light User | Document entry and status updates only |
| Retail checkout | Light User | Narrow, high-frequency actions |
| External collaborators | Portal user | Submit only, no access to internal data |
A common mistake is giving the shop-floor supervisor a light account. He then cannot see the capacity report, and two days later you open a full account anyway. The order should be: list the actions per role, then assign account types, then talk about quantities.
2. Offline mode: it solves network blind spots, not "work without internet"
Last year we worked on a precision parts project. The factory building was steel structure and the WiFi had three dead zones. The scanner lost connection there, workers wrote things on paper, and everything was entered after the shift. Anyone who has done shop-floor work knows what that does to data quality.
Odoo 20's offline mode targets exactly this: with no connection you can still create, edit and query documents, and everything syncs once you are back online. Documents are stored locally first; if there is a conflict with the server, the system asks for manual confirmation instead of overwriting. That is safer than automatic merging, but it means someone has to clear the conflict queue.
Two things to know up front. Master data changes made offline, such as a new item code, do not take effect immediately, so the order of work matters. And offline mode covers short interruptions, not days without a network — it is not a substitute for a local deployment.
3. Continuous production, weekly planning board and BOM comparison
Three changes on the manufacturing side:
Continuous production releases the next operation automatically when the previous one is completed, which cuts waiting between operations. The weekly planning board lays manufacturing orders out by week and lets you drag to change priority. BOMs can be compared side by side, which saves time in engineering change reviews.
The implementation caveat: continuous production depends on accurate routings and lead times. If the settings are off, automatic release multiplies the error and is harder to trace than manual scheduling. Run one representative routing end to end before you enable it.
On the inventory side, two smaller changes are worth noting: suggested safety stock based on the last 30 days of sales, and automatic valuation carry-over for inter-company transfers. The second is useful for groups with several legal entities, where this used to be a manual adjustment.
4. Three preconditions before AI Agents go live
This release lets you create automations in natural language. Describe something like "when a supplier sends a confirmation email, confirm the related manufacturing order", and the system explains what it will do; once you confirm, the flow runs on schedule.
Three preconditions are usually left out of the announcements:
First, AI capability is billed separately in credits and is not part of the licence, so it is easy to miss in the annual budget.
Second, dirty master data amplifies errors. Duplicate item codes or duplicate customer records simply make the wrong result appear faster.
Third, pilot one process with clear rules and low cost of failure, such as replenishment reminders, for two weeks before expanding.
Two built-in assistants are worth a look: the AI accounting assistant flags anomalies based on the ledger, and the AI website assistant builds pages and edits sections through conversation. For companies without a dedicated designer, the second one may see more real use.
5. Five checks before upgrading
| Check | Action | Pass criteria |
|---|---|---|
| Customisations | List every custom module, its maintenance status and the business it serves | Decide what to keep or drop before anything else |
| Master data | Check item code uniqueness, BOM depth, duplicate customers | Clean the data before upgrading, not after |
| Interfaces | Inventory external interfaces and third-party connections | Confirm API compatibility in the new version |
| Test environment | Run one non-critical process end to end | Only then schedule the upgrade window |
| Licence structure | Recalculate account types against the new role list | Know how much of the base a light account covers |
Cross-version upgrades are technically possible. The risk sits in custom modules and data compatibility, not in the version itself. Do not reverse the order: clean the data, assess the customisations, then talk about a schedule.
Frequently asked questions
Q: We are on 17 or 18. Can we upgrade directly to 20?
Technically yes. Assess custom modules and data compatibility first, and rehearse the full upgrade in a test environment before agreeing a date.
Q: Does AI cost extra?
Yes. AI capability is purchased separately in credits and is not included in the licence; usage is visible in the back office.
Q: Can a light account run approvals?
Not advisable. Light accounts are designed for frontline operations; approval flows and cross-module analysis still need a full account.
Q: Can offline mode lose data?
Documents created offline are stored locally and synced when a connection returns. Conflicts are flagged for manual confirmation rather than overwritten, so someone needs to work that queue or it accumulates.
Next step
If you are assessing whether to upgrade, or unsure where to start, begin with a management diagnosis. In thirty minutes we can tell you which problems a system can solve and which have to be sorted out in the process first.
Call 400-891-2518 or email info@mikorou.cn — we reply within one business day.
Related reading
- Odoo digital system: one integrated, modular foundation
- Manufacturing: how BOM and cost accounting work in practice
- Entering the German market: registration, tax and compliance
- Delivery standards: whether it truly works
- Where processes break, by industry
Disclaimer: technical information in this article is compiled from Odoo's official release material. Features and billing are subject to Odoo's latest documentation and do not constitute a delivery commitment or price quotation.
