Why "we'll customise it later" is the most expensive sentence in ERP
We pulled the invoices from 40 businesses that moved to us from an open modular ERP. On average, 61% of what they had spent went on making the software match a process nobody had ever written down.
Every ERP sale contains the same promise, delivered in the same reassuring tone: the platform is fully modular, so we can customise anything you need later. It sounds like generosity. It is, in practice, an invoice with a delayed due date.
We wanted a number rather than a hunch. So in March we asked forty businesses that had migrated to us to send whatever they still had from their previous system - quotes, AMC renewals, change requests, partner invoices. Thirty-one of them found the paperwork. What follows is what those files said.
The money went where nobody planned it to
Licences were rarely the problem. Across the thirty-one businesses, licences accounted for a median of 24% of the three-year spend. Implementation and customisation took 61%. The remaining 15% went on hosting, training and, in four cases, an emergency consultant hired when the original partner stopped replying.
The pattern behind the numbers was consistent enough to be uncomfortable. A business would buy a flexible platform precisely because it was flexible, then discover that flexibility is a question, not an answer. Somebody has to decide what the approval chain should be. Somebody has to decide whether a delivery challan can be edited after dispatch. Those decisions had never been made explicitly - they lived in the head of whoever had been doing the job for eleven years.
We paid four lakh to build a purchase approval flow. Two years later I found out the storekeeper was still approving everything on WhatsApp and marking it in the system afterwards.
- Owner, auto components unit, Chakan
Customisation is a tax on the future, not just the present
The cost that hurt most was not the original bill. It was what customisation did to every subsequent decision. Once a module has been forked, upgrading it is a project. Once upgrading is a project, it gets deferred. Once it is deferred long enough, the business is running a version nobody supports, and the only person who understands the modifications has left.
Eleven of the thirty-one were on a version at least two major releases behind. Six could not tell us who had written their customisations. Two were paying an annual maintenance fee to a partner who had, on inspection, done no work for nineteen months.
What we decided to do instead
None of this makes flexible platforms bad. If you employ a product manager and two developers, an open ERP is genuinely the right choice, and you will build something better than anything we could sell you. That is not most businesses.
So we drew a hard line. Modules cannot be modified. Not by you, not by a partner, not by us for one customer. What you can do is choose which modules make up your pack, rename anything to match your vocabulary, decide who approves what, and write as many automations as you like. Everything above the module is yours. Everything inside it is ours, and it stays the same for every customer.
The trade-off is real and we will not pretend otherwise. Roughly one in nine businesses that talk to us needs something a pack genuinely cannot do. When that happens we build a custom pack - assembled from the same unmodified modules, with new actions and automations written around them - or we tell them we are the wrong fit and suggest someone else.
Three questions worth asking your vendor
If you are evaluating any business system right now, these three questions surfaced most of the trouble in our sample:
- What happens to my customisations when you release version 5? If the answer involves a "re-implementation exercise", budget for it now.
- Can I see the total three-year cost, including the partner's time? A vendor who will not put implementation in writing is telling you something.
- What does day 30 look like if I change nothing at all? If the honest answer is "not much", you are buying a construction kit, not a system.
We publish the anonymised dataset behind this piece on request - write to us and we will send the spreadsheet. It has flaws: it is self-selected, everyone in it left their previous vendor, and thirty-one is not a large number. Read it with that in mind.