Zero retention in financial reconciliation means source workbooks and output files are purged on a short clock — hours, not quarters — instead of sitting indefinitely in a vendor’s object store. Controllers who treat “we keep your files for convenience” as a feature are buying residual liability dressed as customer success.
What question should every controller ask before uploading?
Ask this out loud in the next vendor demo: “Where do our bank and GL extracts sit after we upload them — and for how long?” If the answer is “indefinitely, for your convenience,” that is not convenience. That is an expanded blast radius.
Finance files are not product telemetry. They are evidence. The audit room already knows how to follow a trail; it should not have to invent one through a third-party vault you never intended to create.
Why does indefinite retention create audit smell?
- Access surface grows with every retained extract — more keys, more people, more “temporary” exceptions that become permanent.
- Incident response gets harder when historical ledgers still live outside your perimeter months after close.
- Training and “model improvement” claims that require forever-data turn your close files into someone else’s dataset.
- Procurement and security questionnaires catch up eventually; vague retention answers kill deals late, not early.
What should a clean default look like?
A cleaner default is boring on purpose: upload two sides, map keys and amounts, download Matched / Partially Matched / Not Matched, then let the system delete source and output files within 24 hours. Private storage. TLS in transit. Subscription-gated access. Account and job metadata for ops — not a second general ledger in the vendor’s cloud.
You already keep the export you need for the binder. The vendor does not need a permanent copy to prove the product works.
How do you evaluate vendors without inventing customers?
You do not need a logo wall. You need precise answers: purge SLA, lifecycle rules as a secondary control, whether files train models, what metadata survives, and whether public URLs ever exist for your objects. If a tool needs your data forever to “improve the model,” ask what you are really buying — reconciliation, or a data relationship you never approved.
ReconCore is built as the standardization layer for month-end matching with a hard retention clock. If that constraint matches how your auditors already think, book a 15-minute briefing — bring a real pair of extracts, leave with a clear yes or no.