Skip to content
Felix Schumann
Editorial illustration about: Financial automation: Reconcile before marking it complete
Journal / 13

Financial automation: Reconcile before marking it complete

Felix Schumann·

A payment has arrived. Does that mean the invoice is fully settled and the money is already in the bank account? Not necessarily. I would model those states separately in a financial automation. The value lies in helping a team inspect discrepancies and understand exactly what each number represents.

Research checked: 2026-09-09 · Cover: AI-generated illustration

Which movement are we comparing?

Stripe’s balance report distinguishes balances, activity and payouts. Its documentation points to a separate report for payout reconciliation. This illustrates why a payment event and a bank deposit provide different views of the same process.

For a custom dashboard, I would name the sources and comparison boundaries first: invoice, payment provider and bank. Dates, amounts, currencies and references must match the chosen perspective. Otherwise an expected timing difference can look like an error.

Stripe: Balance summary report

Calculation remains deterministic

My design uses explicit rules for amounts, rounding and currencies. AI can summarize an unusual description or explain a review case. It should not decide through free-form prose that an arithmetic discrepancy has disappeared.

A purely invented example: a payment of 100 euros minus a 3-euro fee leaves 97 euros net. Expecting a 100-euro bank deposit would already use the wrong comparison. These figures are not a provider’s fee schedule; they illustrate the required distinction.

Keep uncertain matches visible

I would prioritize explicit references. Matching amounts and approximate dates can suggest a relationship without proving it. Several invoices for the same amount provide a simple counterexample.

Ambiguous cases receive a reason: missing reference, different currency, partial payment or multiple candidates. A review view presents the relevant records together. Manual decisions retain a timestamp and explanation.

The workflow at a glance

  1. 01NormalizeAmount, currency, reference
  2. 02MatchPrefer explicit references
  3. 03ReviewResolve discrepancies traceably
Our schematic illustration of the proposed design, not measured data.

What I would check before handover

Test data should include duplicate notifications, partial payments, refunds and payouts on a different day. Define acceptable states before running each case. Restarting the system must neither lose confirmed matches nor create extra entries.

In a sales application, this approach can support the transition from an order to operational review. It does not replace acceptance by the responsible accounting team. The first useful deliverable is a traceable reconciliation with outstanding differences, before automating further actions.

Sources and further reading