Count how many times a single customer address gets typed into your company. Once in the CRM when the enquiry arrives. Again in the ERP when it becomes an order. Again in accounting when the invoice goes out. Possibly once more in the shipping tool.
Four times, the same address. And when that customer moves, you update it in one place — and from that moment three systems are wrong.
That is a data silo. Not a technical problem, but a place where your company loses time every single day without anyone noticing.
The test that finds a silo
There is one question that reveals it reliably: when a piece of information changes, how many places do you have to change it?
An address. A price. A contact person. A delivery status.
If the answer is "one", you have no silo. If it is "two or more", you have one — and every additional place is a failure point waiting for its moment.
The second test is less comfortable: which number wins when two systems disagree? If you cannot answer that immediately, you do not have a data problem. You have a decision problem.
Why it costs more than it looks
Double entry is the visible part, and the cheapest one. Two minutes per record is survivable.
What follows is not.
Someone calls a customer back and quotes a price the ERP changed weeks ago. A shipment goes to the old address. A payment reminder reaches someone who paid the day before yesterday. None of those cost two minutes. They cost a phone call, an apology, and a piece of trust.
And the most expensive part is the one nobody measures: decisions made on numbers nobody fully trusts. If every report starts with checking which system is right, you decide slower — or you stop deciding.
What an ERP-CRM integration is. And what it is not
An integration connects two systems so that a piece of information is entered once and then flows automatically to wherever else it is needed.
Typically that means:
- customer and contact records moving from CRM to ERP
- quotes and orders moving in both directions
- invoice and payment status flowing back into the CRM
- delivery and stock information reaching the people who sell
What an integration is not: a substitute for order. If your master data is messy, an integration distributes that mess faster and more reliably than any human could. This is the most common reason these projects disappoint — not the technology, but the sequence.
Four ways to connect two systems
There is no single correct answer. There are four, and they differ in effort, cost, and in what happens when something breaks.
1. Native connector. Both vendors already built the bridge. Fast to set up, inexpensive, little room to manoeuvre. If your process deviates from the standard, you either bend the process or you need a different route.
2. Integration platform. A service between the systems that maps fields onto each other. Flexible, no development required, but recurring cost per record or per connection — and a dependency on a third vendor.
3. Custom API work. Built directly against both systems' interfaces. Full control, no licence fees, highest upfront effort. Worth it when the workflow *is* your business rather than a side effect of it.
4. Export and import. A file and a schedule. Often dismissed, but it is the right answer when a daily refresh is enough and the budget is tight. More honest than a real-time integration nobody maintains.
The choice comes down to one question: how quickly does the information need to be on the other side? Seconds, hours, or a day — that decides more than any vendor presentation.
The sequence that determines the outcome
Most projects start with the technology. That is the mistake. This order holds:
- Decide the system of record, field by field. Who owns the customer address — CRM or ERP? It can only be one. That decision is not technical; it belongs to management.
- Clean the master data before anything flows. Duplicates, dead records, inconsistent spellings. Thankless work, and without it the rest is worthless.
- Connect one workflow. Not everything at once. The first workflow is the proof that the approach holds.
- Measure what changed. How many records pass, how many fail, and why. Without those numbers you do not know whether it works — you are hoping.
- Only then expand.
Step one is where most projects fail, and it costs no development hours at all. It costs a decision.
What regularly goes wrong
Mismatched data formats. An address is one field in the CRM and five in the ERP. Solvable — but someone has to decide what happens when the mapping is ambiguous.
No error handling. What happens when a transfer fails? Without an answer, the integration appears to run for months, and then someone notices nothing has arrived since eight weeks ago. A failure nobody sees is worse than a loud one.
Too much at once. Connecting seven systems simultaneously means debugging seven failure points simultaneously.
Legacy software with no interface. Sometimes the honest answer is that the system should be replaced, not connected.
When it pays off — and when it does not
It pays off when the same information moves from A to B by hand several times a day. The effort returns within a reasonable period, and the gain grows with the company.
It does not pay off when the task happens ten times a month. Then a clean checklist is cheaper than any interface — and an interface that rarely runs will fail without anyone noticing.
Do that calculation before the project, not after. An honest calculation kills some integrations and builds the remaining ones properly.
Where AI fits — and where it does not
Once two systems share the same data, things become possible that were not before: analysis across the whole workflow, automatic classification of incoming documents, prioritisation of open cases.
But the order stands: clean data flow first, intelligence on top of it. A model fed contradictory data produces contradictory results — only faster, and phrased more convincingly. Which workflows are worth automating first is covered in our article on using AI in your business.
The honest conclusion
An ERP-CRM integration is not a digitalisation project. It is a clean-up operation carried out with technical means.
The hard part is not the interface. It is deciding which system owns the truth — and being willing to fix the master data before anything flows. Skip those two steps and you have built a fast connection between two unreliable datasets.
We build custom interfaces and integrations at ALPENIQ Labs, and the automation on top at ALPENIQ AI.
Frequently asked questions
How do I know we have a data silo?
Ask how many places you need to update a changed customer address. More than one means a silo. The second test: when two systems show different numbers, do you know which one wins?
How long does an ERP-CRM integration take?
Almost entirely down to the state of your master data and the number of workflows, not the technology. A single well-scoped workflow is a matter of weeks; a project spanning every department takes correspondingly longer. Anyone quoting a duration without looking at your data is guessing.
Do we need new software for this?
Usually not. Most current ERP and CRM systems ship with interfaces. Replacement only becomes the honest answer when a system has no interface at all or is no longer maintained.
What is the difference between integration and automation?
Integration ensures both systems hold the same information. Automation ensures something then happens — an invoice is created, a notification goes out. One is the precondition for the other.
What happens when a transfer fails?
That belongs settled before the first record moves. A robust integration logs every failed attempt, retries in a controlled way, and escalates to a human when it cannot proceed. An integration that fails silently is more dangerous than none at all.