Reconciliation Agent
Every break explained, before the desk opens.
Reconciliation is invisible until it fails, and then it is the only thing anyone talks about. The work itself has not changed in twenty years: pull four exports, line them up in a spreadsheet, and spend the morning finding the twelve rows that do not agree. This agent does the lining up. It reads the switch, the core banking postings, the settlement files and the bank statements together, matches what belongs together, and hands the desk the breaks — each one with a stated cause rather than a highlighted cell.
- Reads
- Switch, core, settlement, statements
- Matching
- One-to-one through many-to-many
- Breaks
- Classified with a cause
- Delivers
- Exceptions, not exports
- Multi-source matching
- Breaks with a cause
- Proposed adjustments
- Full audit trail
What Reconciliation Agent does
Every side of the transaction, together
Switch logs, core banking postings, scheme settlement files and bank statements are ingested in their own formats and normalised onto one shape. Reconciliation fails most often at this step, not at the matching.
Matching that survives real data
One-to-one where the data is clean, and one-to-many and many-to-many where it is not — a settlement batch against its constituent authorisations, a single credit against the fees netted out of it.
Tolerances instead of exact equality
A cent of rounding, a fee taken at the far end and a posting that lands the next business day are not breaks. They are matched within tolerance and recorded as matched, so the exception list stays worth reading.
Breaks classified, not just listed
Timing difference, fee variance, FX movement, duplicate posting, missing leg, amount mismatch. A list of unmatched rows is the start of the investigation; a classified break is most of the way to the answer.
The reason attached to each one
Every break carries which fields differed, by how much, and which side each figure came from — so an analyst opens a case knowing what happened rather than reconstructing it from four systems.
Rules learned from what the desk already does
The judgement calls analysts make repeatedly — this counterparty always settles a day late, this fee is always deducted at source — are captured as rules rather than remade every morning.
Adjustments proposed, never posted
Where a break has an obvious correction the agent drafts it and shows its working. A person approves and posts. Nothing touches a ledger on a model's say-so.
An audit trail that stands up
Every match, every tolerance applied and every exception decision is recorded with who or what decided it and when — because in reconciliation the record of the reasoning is the deliverable, not a by-product.
From four exports to one exception list
- 1
Ingest
Each source pulled in its own format — switch, core banking, settlement, statements.
- 2
Normalise
Dates, currencies, references and signs brought onto one shape before anything is compared.
- 3
Match
One-to-one, one-to-many and many-to-many, inside the tolerances the desk has agreed.
- 4
Classify
What is left is sorted by cause — timing, fees, FX, duplicates, missing legs.
- 5
Resolve
Breaks presented with their evidence and, where there is one, a drafted correction to approve.
- Banks and payment processors reconciling across a switch and a core
- Finance operations teams closing the day by hand
- Fintechs whose volumes have outgrown the spreadsheet that used to work
- Anyone whose month-end is spent explaining last month
Questions people ask
No. It drafts a correction where the cause is clear and shows the working behind it, and a person approves and posts. An agent with write access to a ledger is a control failure waiting for an audit, whatever its accuracy.
Ready to build something remarkable?
Tell us about your next big move — or your first AI integration. We will come back within one business day with a point of view, not a brochure.
Typical first response: under 24 hours · No obligation · NDA on request