The user owns the data; the platform processes it on your behalf. The agreement should say so plainly: ownership of raw data stays with the company, use is limited to the agreed scope, and structured data plus raw records can be exported by batch and date range, with a full migration on exit. Settle this before signature, not after go-live.
Three clauses to read: ownership, scope of use (can the platform use it for industry analytics or model training), and the form and deadline for handover on exit. The third is the one people skip, and it decides your negotiating position three years later.
“Exportable” has two levels: business data as spreadsheets, and raw records — sensor readings, original inspection values — exported with timestamps. The second is what actually survives a customer audit or a dispute.
The technical safeguard is separation of master data from transactional data: batches, materials and equipment carry unique codes, so an export preserves the relationships instead of producing a pile of unlinked tables. That is the same principle as the single-ledger design, applied to data governance.
A practical test: run a small export before onboarding and see whether the output could support a real customer audit on its own. If it can, the data is genuinely in your hands; if not, renegotiate the delivery format.