Odoo demonstrates well. That is part of the problem: the demo makes it look like configuration, and the difference between a configured Odoo and an implemented one is roughly the difference between a furnished flat and somewhere you can live. Here is what the work really consists of.
Community or Enterprise
Community is the open-source edition and costs nothing in licence. Enterprise adds the studio customiser, full accounting automation, mobile applications and official support, and is charged per user. The decision is rarely ideological. It comes down to how many named users you have, whether you need the accounting automation, and whether you want a vendor obligation behind the system. We size both and show the five-year number, because the licence is usually the smaller half of the cost.
The data migration is the project
Chart of accounts, opening balances, customers, suppliers, items, price lists, open invoices, stock on hand. Most of it arrives as spreadsheets with duplicates, inconsistent units and items that three departments name differently. Cleaning that is unglamorous and it is where a go-live is won or lost, because a system whose opening numbers nobody trusts will be run in parallel with the old spreadsheets forever.
Customise late and sparingly
Every customisation is a small debt payable at the next version upgrade. We configure first, run the process as standard Odoo intends for a fortnight, and only then decide what genuinely has to change. A surprising share of requested customisations turn out to be habits from the previous system rather than requirements of the business.
The modules people underestimate
Inventory valuation method, because changing it after go-live is painful. Multi-company and inter-company rules, if there is any chance of a second entity. Analytic accounting, which is how you will get cost per project, per branch or per vehicle later. Access rights, which nobody thinks about until the wrong person sees margins. These four decisions are cheap on day one and expensive in month nine.
Training is not a week at the end
Role-based sessions on the client's own data, run before go-live and repeated after it, with a named internal owner per department who becomes the first line of questions. The alternative is a consultant answering the same question forty times and a team that quietly returns to Excel.
A realistic shape for the timeline
A single-company implementation covering finance, purchasing, inventory and sales is a matter of weeks rather than days, and manufacturing or multi-entity work extends it. Anyone promising a full ERP in a fortnight is either selling a demo or planning to leave before month-end closes for the first time.