System of Record

A decision, written down, rather than a technical property. Any database can store a value. A system becomes the record when the organization agrees its version is the one everyone else defers to, and then enforces that agreement in how data flows. Three things usually settle which system earns it. Origination is the first: whichever tool creates and validates the value has the strongest claim. Lifecycle is the second, because a system that manages change over time with effective dates and history beats one holding only a current snapshot. Accountability is the third, since someone has to own the accuracy of that field as part of their job. Where the three point at different systems, expect friction, and resolve it deliberately instead of letting integration behavior decide by default.

Field by field, in a document, before the integration gets built. The exercise is dull and it prevents most of the arguments that follow. List the fields more than one system holds, and for each name the owner, the systems allowed to read it, and whether anyone else may write it. Expect boundaries to fall in odd places. An email address might be owned by recruiting until an offer is accepted and by the employee record afterward, which means ownership can transfer at a defined event rather than sitting fixed forever. The recruiting-to-HR boundary is the common example, and ATS vs HRIS sets out where it usually falls. Where two teams both want a field, the tie-break is who is accountable when the value is wrong.

Time first, then trust, then decisions. Reconciliation work appears that nobody planned for, because a human has to judge which of two plausible values is right. Reports built on the two sources stop agreeing, and once a leadership team has seen two different numbers for the same thing, they discount both. The failure is rarely dramatic. It shows up as a manager insisting the tool is wrong, an interface that silently overwrites corrections every night, and a growing preference for spreadsheets kept outside every system precisely because nothing can overwrite them. Fixing it afterward is expensive, since the data has already diverged and someone must pick a winner retrospectively, discarding real work on the losing side. The cheap moment to decide ownership is before the first sync ever runs.

Through permissions, direction of flow and monitoring, not good intentions. Make the non-owning copy read-only wherever the platform allows, so the field cannot be edited in the wrong place at all. Set the interface to move that field in one direction, and log every write with its source so a bad value can be traced back to where it came from. Add reconciliation checks that compare the two copies on a schedule and raise a difference as an exception instead of silently correcting it, because silent correction hides the cause. Then tie the model to a named owner and revisit it whenever either system changes. Integrations decay once the people who agreed the rules move on, and an unexplained mapping is one upgrade away from being changed by somebody guessing.