Most tools match two things.Finance problems rarely have two sides.

Feed, file or document

Direct connections, scheduled extracts, portal downloads, statements and scanned documents all land in the same pipeline.

• ERP, banking, gateway, billing and operational systems
• Statements and confirmations read as data, not filed as PDFs
• The original kept as evidence for the match it supports

One shape, many systems

Dates, currencies, entity codes and identifiers brought to one form so two sides can be compared without a manual clean-up first.

• Multi-entity, multi-currency and multi-book handled together
• Identifiers repaired where a system truncated or reformatted them
• Reference data mapped once and reused by every reconciliation

Line, header or balance

Two sides compared at the level the question needs. Comparing a line-level source to a header-level one is why reconciliations never close.

• Line, header, batch or balance, set per reconciliation
• Aggregation applied deliberately rather than by accident
• Both sides held at their native grain as well as the compared one

Resolved, not deferred

Split settlements, batch payments and part-deliveries resolved into the set of records they actually cover.

• One-to-many and many-to-many sets resolved explicitly
• Split and batch settlements unbundled to their components
• Partial matches cleared to the part they settle, not left whole

When close enough is correct

Rounding, bank charges, exchange differences and agreed variances treated as tolerances rather than exceptions, with the rule visible.

• Tolerance by value, percentage or both, per reconciliation
• Charges and exchange differences classified, not absorbed
• What was matched within tolerance shown, never hidden

Names never agree

Where identifiers differ across systems, matching uses names, amounts and dates together, and records why each match was made.

• Party name variants resolved to one counterparty
• Amount, date and narration used together where no key exists
• Every fuzzy match carrying its confidence and its reason

Classified, not listed

Differences classified into timing, error, omission, dispute and unidentified, so the list can be worked by cause instead of read top to bottom.

• Timing differences separated from genuine breaks
• Root cause attached, so the same break stops recurring
• Unidentified items named as unidentified, not netted away

A break gets older, not better

Every open difference carries its age, so a two-day item and a two-year item are never presented as the same problem.

• Ageing by reconciliation, counterparty and reason
• Items past a threshold escalated rather than carried forward
• What was written off, and on whose authority

Somebody is working it

Each break is assigned, worked and closed with a note, so the reconciliation is a live control rather than a monthly report.

• Assignment, action and closure recorded against the item
• Correspondence and evidence attached to the break itself
• What is genuinely irrecoverable separated from what is late

Rates, not instances

Testing everything reports a rate rather than a handful of findings, which is a different and more useful conversation with a board.

• Every transaction tested, not a judgemental selection
• Exceptions reported as a rate across the population
• Thousands of items below materiality that add up, surfaced

As it happens

The document that supports a match is held against the match, so testing is a read rather than a document request.

• Every match linked to the record that supports it
• A complete trail from source record to posted entry
• Sample requests answered without a new extraction

Controls that run

Reconciliations and control checks run on a schedule rather than at period end, so a failure is found in the week it happens.

• Control checks run continuously, not as a close task
• Breaks escalated on age rather than discovered at close
• The same evidence serves internal and statutory audit