I spent eleven years selling ERP licences before I stopped. The thing that eventually got me was not a moral objection — it was watching the same conversation happen every year at renewal, in which a client who had done nothing differently was asked for more money, and had no position from which to argue.
That conversation is the ten-year argument, and it is the one most platform business cases never model. A three-year comparison makes proprietary software look competitive, because the costs that compound have not compounded yet. Run the same comparison over ten years and the shape changes completely.
The four numbers that actually compound
In a typical mid-market platform decision, four costs grow after the initial purchase, and only one of them is on the first invoice.
Seat growth. You buy for the organisation you are, and pay for the one you become. A 200-seat deployment growing at 6% a year is a 358-seat deployment by year ten. Under per-seat licensing you have paid for 158 seats you did not budget for, at whatever the rate has become by then.
Annual uplift. Maintenance and subscription typically rise 4–8% a year, and the rise is rarely tied to anything you can audit. Compounded over a decade, a 6% uplift means the year-ten invoice is 1.79 times the year-one invoice for identical software.
Edition creep. The feature you need in year four is in the tier above. This is not an accident of product design; it is product design. Once your processes depend on it, the upgrade is not really a decision.
Exit cost. The least visible and the largest. When leaving means re-implementing from scratch because your data is in a format you cannot fully extract, the renewal is not a negotiation. Every year you stay increases the cost of the year you leave.
A worked comparison
Below is a comparison we have run, with variations, for a good number of clients. It assumes 200 users growing at 6% annually, a mid-market ERP at £180 per user per year with a 6% uplift, and an open-source deployment with the same implementation scope, managed hosting and Enhanced support. Implementation is deliberately held equal — the work of configuring a platform against your processes is roughly the same either way, and any vendor claiming otherwise is selling you a demo rather than a deployment.
| Cost line, ten-year total | Proprietary | Open source |
|---|---|---|
| Licence or subscription | £574,000 | £0 |
| Implementation | £85,000 | £85,000 |
| Support and maintenance | included | £180,000 |
| Hosting and infrastructure | included | £168,000 |
| Major upgrades (two) | £40,000 | £48,000 |
| Ten-year total | £699,000 | £481,000 |
| Cost to leave in year ten | Re-implementation | £0 — you keep it running |
Two things about that table are worth saying plainly, because they cut against us.
First, the open-source column is not cheap. £348,000 of support and hosting over a decade is real money, and anyone presenting open source as free is either inexperienced or hoping you are. The software costs nothing. Running it properly costs what running anything properly costs.
Second, at three years the gap is far narrower — around £40,000 on these assumptions, which is well inside the margin of error on an implementation estimate. If your planning horizon is genuinely three years, this argument does not decide anything. It is the compounding that decides it, and compounding needs time.
The number that is not on the table
The bottom row is the one that changes behaviour. Not because anyone plans to leave, but because of what it does to every conversation before that.
When leaving is genuinely possible, the renewal is a negotiation between two parties who both have options. When it is not, the renewal is a notification. We have clients who have never needed to invoke that position and have benefited from it every single year, simply by having it.
This is also why we structure things the way we do. Your configuration and any extensions we write are yours outright, in your repository, from the first commit. Hosting can run in your own cloud account. If you leave us, you keep a running system and the runbooks that explain it. That is not generosity — it is the only arrangement consistent with the argument I have just made. A supplier who tells you lock-in is bad while quietly building some is not worth listening to.
When this argument does not apply
Three cases, and we run into all of them.
If you need deep vertical functionality that only a specialist proprietary system has, the ten-year total cost is not the deciding factor, and it should not be. Buy the thing that does the job.
If your organisation has no appetite for an operating commitment and no budget for someone else to carry it, self-hosting will go badly regardless of the licence position. That is an operating decision, not a licensing one — my colleague Samuel has written about it separately.
And if you are small enough that per-seat pricing is genuinely cheap, it is genuinely cheap. Twenty users on a well-fitted SaaS product is not a problem in need of a solution. Come back when you are eighty.
For everyone in between — which is most mid-market organisations — the ten-year argument is the one worth having. It just requires modelling ten years, which is the part most business cases skip.