What usually gets connected
- E-shop and warehouse — an order deducts stock immediately, availability flows back to the shop
- Orders and accounting — the invoice creates itself, nobody retypes it
- Warehouse and shipping — labels and tracking numbers pulled automatically
- CRM and e-mail — client correspondence visible where you actually need it
- Anything and a spreadsheet — the sheet nobody wants to give up fills itself
Three ways to connect systems
| Method | When it is used | What it involves |
|---|---|---|
| API | The system has a documented interface | The cleanest route. Data moves in real time. |
| Import / export | Older system with no API | Files move in batches, typically every few minutes or hours. |
| Straight into the database | You own the system and have access to the data | Fast but more brittle — it shifts with every system update. |
Where integrations usually fail
The connection itself is the easy part. The work is in what comes after.
- Different names for the same thing. What the shop calls a “variant” is an “item” in the warehouse and an “article” in accounting. Mapping those takes longer than the transfer itself.
- Outages. The other side will be unreachable sometimes. An integration that quietly drops an order at that moment is worse than none — it has to wait and retry.
- Duplicates. When a message is sent twice, two invoices must not appear. That is handled by giving every operation its own identifier.
- Rate limits. Most APIs cap requests per minute. The initial data transfer hits that cap reliably.
- Which side is the source of truth. When the same value can be edited in two places, sooner or later they diverge. You have to decide up front which place wins.
You can tell a good integration by the fact that nobody notices it. You can tell a bad one by something going “missing” every so often with no explanation.
What it costs
Connecting one pair of systems over an existing API starts at 15,000 CZK + (about €600) and moves up with the number of data types transferred and how complex the mapping is. Two-way synchronisation, several systems at once or migrating historical data are naturally higher. There is a fuller breakdown on the page about what custom software costs.
How it goes
- Map what flows where. It usually turns out half the transfers are unnecessary.
- Decide the source of truth. One place per value that gets the final say.
- Connect one pair. The most annoying one. The rest can wait.
- Let it run and watch it. Errors announce themselves rather than waiting to be discovered.