Reconciliation is where most treasury automation projects succeed or fail. This page explains the minimum data you need (IDs, statuses, statements), why exception handling matters, and how APIs and host-to-host work together.
Common failure modes across treasury reconciliation programs.
If the ERP/TMS ID does not match the bank ID, status updates cannot join reliably. Teams then fall back to manual matching.
If you only get "sent" vs "not sent", you cannot drive exception workflows. Granular statuses are key to reducing investigations effort.
Statements contain fields, but mapping is not defined (or ISO fields are dropped). The result is false mismatches and manual repair.
Use this as a requirements checklist with banks and internal teams.
| Data element | Where it comes from | Why it matters | Common gap |
|---|---|---|---|
| Canonical payment identifier | ERP/TMS + bank confirmation | Joins initiation to status and statements | ID changes between systems |
| Acceptance / rejection acknowledgement | API response or file ack | Detects immediate failures | Acks not stored centrally |
| Processing status updates | API polling, callbacks, portal, reporting files | Drives exception workflows | Status too coarse or delayed |
| Bank statements | Host-to-host/files, portal downloads, statements APIs | Confirms cash movement and finality | Fields not mapped to ERP/TMS |
| Return / reject reason codes | Status + reporting | Root-cause analysis and repair automation | Codes are not normalized |
| Structured remittance data | ISO messages, remittance fields, statements | Improves match rates for receivables | Data is truncated or not stored |
Related: Client requirements · ERP integrator
Two rails, one end-to-end operating model.
Statements are often best handled via established file rails (host-to-host) with clear cutoffs and batch operations.
Status APIs and callbacks can reduce latency and improve investigations workflows. This is often where teams see the clearest operational ROI.
Build a process for breaks: where they queue, who owns repair, and how data gets corrected. Automation without exception design fails.