You close a sale, start processing it, and discover there's no inventory left. Yet just a short while ago the ecommerce platform said the product was available.
Now the investigation begins: check the ERP, look at the marketplace, ask the warehouse, and try to establish which system holds the right number.
The problem isn't necessarily that any of these systems is broken. Each one may be doing its job correctly while showing a different version of the same information.
Integrating systems means more than connecting them
An update can arrive late, never arrive, be processed twice, or fail without anyone noticing. This can happen with inventory, but also with prices, orders, customers, payments, reservations, or any data flowing between platforms.
The more systems involved, the more important it becomes to define how information moves and what happens when something goes wrong.
First, define the source of truth
If the ERP, ecommerce platform, and warehouse show different quantities, someone needs to determine which system represents the authoritative state. That decision should be made before an inconsistency appears.
A source of truth doesn't mean every system must store the same value. It means knowing which system is authoritative for each type of data and which others consume or derive it.
Then, think about how changes propagate
A reliable integration has to answer a few practical questions:
- What triggers an update?
- Is the change immediate or eventually consistent?
- What happens if the destination doesn't respond?
- When should the system retry?
- How do we avoid processing the same event twice?
- How do we know synchronization has stopped working?
The last question is often the most important. An integration that works 99% of the time can appear flawless for months, but that remaining 1% rarely identifies itself as a synchronization error.
A technical error becomes a commercial problem
When the problem surfaces, it can mean a sale you can't fulfill, an incorrect price, an order that never reached its destination, or an awkward conversation with a customer.
That's why a good integration isn't measured only by whether one API can send data to another. It also needs to detect discrepancies, recover from failures, and explain what happened when the numbers don't match.