When a middleware pays off — and when it does not.
Between two systems a direct interface is often enough. From three systems on, the sum tips over — and the question is no longer whether but which platform. A way to decide, without a product catalogue.
Key points
Two systems: a direct interface. From three systems sharing data upwards: look at a middleware.
Choosing a middleware is an operating decision — monitoring and maintainability count for more than the fastest first connection.
Point-to-point connections grow quadratically: five systems can mean up to ten interfaces.
Anyone who does not plan for error handling is planning for data loss.
Every further system makes the interface problem disproportionately worse.
Two systems need one connection. Five systems sharing data need up to ten. Each has its own format, its own error handling, its own documentation — or none at all.
A middleware gathers those connections in one place: one system delivers once, all the others draw from it. New systems dock onto the platform instead of onto every existing system separately.
The cheapest middleware is the one you do not need.
If you are connecting exactly two systems that rarely change, with a manageable volume of data, then a well-documented direct interface is the cheaper answer. Introducing a platform to run a single connection is overhead without benefit.
The question to ask: will a third system arrive within the next two years? If so, count the cost of migrating away from the direct interface right now.
Integrations are decided in the third year of operation — not at go-live.
Most integration comparisons judge how quickly the first connection stands. What matters is the opposite: what happens in the third year of operation, when an API update to the shop system breaks the mapping — who notices, who fixes it, what does it cost?
Platforms like Alumio play to their strength exactly there: monitoring, logging, retrying failed transfers and making adjustments without touching the code. Tools like n8n score on lean automation and on running in your own data centre. The choice follows the landscape — not the supplier's portfolio.
Count connections, not systems.
Recommendations for practice
Count your connections, not your systems — from around three active interfaces upwards, a platform is worth examining.
Judge the candidates on the operating case: error detection, retries, logging, alerting.
Ask for documentation another supplier could carry on working from.
Start with the connection that causes the most manual work today.
How many interfaces does your company run — and who knows about them?
Taking stock of your system landscape shows where the manual work and the risk sit. Often it is the first overview there has ever been.
Discuss the integration→