
Matching items automatically and looking only at those that do not reconcile
In the reconciliations team, analysts compare account movements, data feeds and ledgers, and spend much of the day on matches that a clear rule would make on its own.
The time spent on the easy items
The items come from sources that do not speak the same language: account movements, the data feeds from counterparties and card schemes, the accounting subledger. The analysts compare them using spreadsheets and matching rules that have grown up over the years, which recognise the simple cases and leave out the rest, for example a single receipt that covers several invoices or a payment reference written differently. Open items pile up, and when the period close arrives someone has to work backwards to find out why a balance does not reconcile.
People receive only the anomalies, already explained
In the example, the agent matches the items using the bank's rules and, where the rule is not enough, proposes a match and shows which elements it relies on, such as the amount or the reference. It groups the unmatched items by type of anomaly and brings them to the reconciliations analyst together with what it has already checked. The analyst confirms or rejects the proposals and decides the adjustments, while the accounting entries stay with whoever is responsible for them.
- The agent matches the items according to the rules and proposes a match for the doubtful cases
- Groups the anomalies and explains what does not reconcile
- The analyst confirms the matches and decides the adjustments
Illustrative example
What an assisted reconciliation requires
| Item | In this case |
|---|---|
| Data | In this caseAccount movements, counterparties' data feeds, ledgers and statements, with the same identifier for every transaction |
| Modules | In this caseData Foundation, Data Intelligence, Agents |
| Who decides | In this caseThe analyst on the doubtful items; accounting adjustments are approved by whoever is accountable for the close |
| What we measure | In this caseThe share of items matched without intervention, the items open at period end and the time needed to explain a variance, compared with previous closes |
| AI Act risk class | In this caseIt works on accounting items and does not assess people, so it does not touch the creditworthiness of individuals, the high-risk use in Annex III that concerns banks. Before development, we confirm the classification all the same with the bank's risk and compliance functions |
| Where to start | In this caseWith Deployment, because the matching rules already exist, even if scattered, and the items closed in the past are cases with a known outcome on which to test the agent |
When the sources have no common reference
If the counterparties' data feeds carry no reference linking them to the account movement, and nobody can add one upstream, the agent proposes matches that the analyst has to redo by hand. In that case we fix the data first with Data Foundation, together with those who produce the feeds. With low volumes and rules that already work well, too, the gain is modest, and we prefer to say so before we begin.
Other sheets for the industry are in Banking & payments; the full list is in the use cases.
Let's start with the items that stay open
With the reconciliations team we look at the items left open at the last close, to work out together which rules are missing.