🛠️ Implementing reconciliation

Choosing your bank connectivity channels

The bank connectivity channel is the technical route by which a company receives data from its bank before it can reconcile it. The choice determines which formats are available, delivery times, reliability and cost. In France, EBICS is the market standard, SwiftNet is the norm for international groups that bank with many institutions, and APIs add real-time data on top of the end-of-day statement, which arrives only once per business day.

Setting up statement delivery with your bank

A statement subscription is the service through which a bank provides a customer with the statements for an account, in an agreed format and at an agreed frequency. Much of a reconciliation’s quality is decided when these subscriptions are set up. Formats, level of detail, frequency and delivery times are negotiated and written into the contract, one account at a time.

• Scope: every account, across all entities and currencies. An account left out is a gap in your controls.
• Formats: camt.053 as the target (specify the version), with CFONB120 or MT940 as a fallback during testing; camt.054 for the detail behind batches booked as a single amount.
• Level of detail: require batch detail (one TxDtls per transaction) rather than opaque lump-sum entries. Some banks offer this as a paid option.
• Frequency and timing: the D+1 statement available by 7:00 a.m. at the latest to keep the morning batch run on schedule; intraday statements (camt.052) at 10 a.m., 1 p.m. and 4 p.m. if treasury needs them.
• References: check that the bank passes through EndToEndId, MndtId and the batch reference, not just a concatenated description.
• Incident recovery: a procedure to resend a statement (missing or corrupted file) and how far back history can be downloaded.

The reconciliation engine: matching rules

The reconciliation engine is the component that compares two or more sets of entries and produces matches between their lines. Matches are made automatically when confidence is complete, suggested to an operator when it is partial, and sent to exceptions in all other cases. The rules run in a cascade, from the strictest to the loosest.

# Rules evaluated in ascending priority order; the first rule that matches wins.
rules:
  - id: r10-endtoend-exact
    priority: 10
    match: one_to_one
    keys:
      - left:  bank.entry.end_to_end_id      # camt TxDtls/Refs/EndToEndId
        right: psp.payout.reference          # reference carried by the PSP transfer
    conditions:
      - amount_equal: true                   # to the cent
    action: auto_match                       # automatic matching

  - id: r20-payout-batch
    priority: 20
    match: one_to_many                       # 1 bank entry <- n settlement lines
    keys:
      - left:  bank.entry.remittance_info    # "SETTLEMENT BATCH 42 ACME FR"
        right: psp.settlement.batch_number
    conditions:
      - sum_equal:
          tolerance: 0.00                    # the batch invariant tolerates no gap

  - id: r30-montant-date
    priority: 30
    match: one_to_one
    conditions:
      - amount_equal: true
      - date_window_days: 2                  # settlement -> bank lag
    action: suggest                          # suggestion for a person to approve

  - id: r90-catch-all
    priority: 90
    action: exception                        # everything else goes to the workflow
    classify_by: bank.entry.btc_code         # pre-sorted by BkTxCd / AFB code

Managing exceptions: workflow and discipline

An exception is any entry the engine has not matched. It becomes an open item that is categorized, dated and assigned to someone. The setup is judged by how fast exceptions are resolved and how well the handling of each one is documented.

• Named owners, with target turnaround times by category (fees: 2 days; missing payout: 1 day, because it is cash).
• Aging visible to everyone: 0–7 days, 8–30 days, more than 30 days; anything over 30 days is escalated to the closing committee.
• Improvement loop: every recurring exception must end up either as a new matching rule or as a fix at the source (a reference missing from an upstream data flow).
• Segregation of duties: the person who reconciles does not approve their own correcting entries above a set threshold.

Managing by KPIs

The auto-match rate is the metric most often reported, but the aging of open items says more about the health of the process. A backlog whose average age stays low shows the team is keeping pace with incoming work. An aging backlog signals a painful close three months before it happens.

Tool landscape

Available tools range from ERP modules to specialized high-volume engines. The selection criteria are the number of banks and PSPs connected, transaction volumes, the need for many-to-many matching, and the team’s maturity. License cost is only one factor among them.