Why integration change breaks first
Integration is where every S/4HANA workstream meets. A finance redesign alters a payload. A logistics partner needs a new B2B flow. A cutover rehearsal needs endpoints repointed. Each change is small. Together they form the largest stream of production change in the program, owned by many teams, and usually governed by a queue that was designed for something else.
Two failure patterns follow. In the first, every integration change goes to a central board. Delivery slows, teams find the queue unbearable, and they route around it with direct connections and undocumented point-to-point links. In the second, nothing is gated in the name of speed, until a changed mapping quietly drops documents during a period close and nobody can say who approved it or what was tested.
Both patterns come from the same mistake: treating all integration changes as equal. They are not. Changing a log level on a flow is not the same as changing the structure of a message that three consuming systems depend on. A control system that cannot tell these apart will always be either too heavy or too light.
This paper argues for a different design. Govern the risk of the change, not the project that raised it. Enforce rules in the tooling, not in meetings. Let evidence be a by-product of delivery, not a document someone assembles before an audit.
The position: control the contract, automate the rest
The thing worth protecting in an integration landscape is the contract between producer and consumer: the structure, meaning, timing and error behaviour of what passes between systems. Most day-to-day changes do not touch that contract. They fix a mapping bug, add an optional field nobody has to read, adjust monitoring or deploy an unchanged package. These should move quickly, with automated checks and light peer review.
A smaller set of changes does alter a contract, add a new party, carry financial documents or touch shared runtime. These deserve named approvers, rehearsal and a rollback plan. If the organisation spends its approval effort there and nowhere else, control gets stronger while the queue gets shorter.
