Adopt the standard. The default should be strong enough that deviating requires an argument you can write down.
Every standard you adopt is a set of decisions you do not have to make from scratch, a set of tools you can reuse rather than build, and a set of integrations that become easier rather than bespoke. If there is a data standard for your domain — FHIR in healthcare, FDX or the open banking standards in finance, the telecom standards, the many public-sector schemas — using it means your consumers already know your data model before they read your docs.
Being a special snowflake has a cost that lands mostly on other people: every consumer pays to learn your bespoke model, every integrator writes a mapping layer, and every tool has to be built for you alone.
The honest exceptions: when the standard genuinely does not cover your domain, when it is so heavy that conformance would cost more than the interoperability is worth, or when it is moribund with no maintenance and no adopters. Those are real. They are also cited far more often than they apply.
Where standards get messy is fragmentation — competing standards in the same domain, versions that drift, specifications that rot from neglect. That is a reason to pick carefully and to check whether the standard is actively maintained, not a reason to invent a private one.
