Guides
Reconciliation, explained plainly.
Short guides on the concepts behind the product. They describe how reconciliation works in general, and make no claims about results.
What reconciliation compares, why differences arise, and what “reconciled” should actually mean.
Reconciliation compares two or more records of the same activity and explains every difference. In its simplest form you compare the bank statement with your books: each credit and debit on one side should have a counterpart on the other.
Most differences are benign. Timing differences, such as a payment recorded today that clears tomorrow, resolve themselves. Others need action: amounts that disagree, payments with no invoice, or the same payment recorded twice.
A useful definition of reconciled is that every item is matched, or has a recorded reason and an owner. “Reconciled” should never mean that someone stopped looking.
Comparing gateway, bank and ledger views of the same money, and separating timing from error.
Payment flows have three natural witnesses. The gateway says what was captured. The bank says what arrived. The ledger says what was recorded. Each answers a different question, and each lags the others by a different amount.
In a healthy flow, captured funds equal funds received plus funds still in settlement. The in-settlement amount is a timing difference and should be visible and dated, not raised as an error.
Whatever remains after accounting for timing is a real difference: a fee, a refund, a duplicate or a missing record. Those are the exceptions a team should spend time on.
Why settlements arrive net, how to model fees, and when a difference is a fee versus a short payment.
Gateways usually deduct fees before paying out, so the amount that reaches the bank is smaller than the sum of captured payments. That difference is expected, but only if you can show the fee schedule that explains it.
A useful rule is to compute the expected fee, compare it with the actual deduction, and treat the remainder as the difference. If a 1.0% gap on a single payment matches a known fee rate, it is probably a fee. If it does not, investigate the customer’s payment.
Record fees as their own ledger lines so they never hide inside revenue. This keeps invoices matched at their full amount and puts the cost where it belongs.
The fields a normalized transaction needs, and why lineage matters more than elegance.
A normalized transaction needs a small set of stable fields: source, account, customer, invoice, payment, settlement, currency, timestamp, reference and reconciliation state. Everything else can hang off those.
Keep the raw record. Normalization is a mapping, not a replacement, and you will need the original when a mapping is wrong or an auditor asks where a number came from.
Store time with its zone, money as integers in the smallest unit with an explicit currency, and identifiers as opaque strings. These three choices remove most downstream bugs.
Typing, owning, aging and approving exceptions so nothing sits unexplained.
An exception is only useful if it has a type, an owner and a way to close. Start with a small set of types, such as amount mismatch, missing counterpart, duplicate and FX variance, and add more only when the existing ones stop being specific.
Rank probable causes using evidence you already hold: does the difference equal a known fee rate, is there a second payment nearby, does the currency explain it. Present the evidence beside the suggestion so reviewers can disagree.
Use a second approver for adjustments above a threshold, and record the reason. The goal is that anyone can later read the timeline and understand what happened without asking.
Make every financial transaction explainable.
Connect your financial systems, reconcile transactions automatically, and build a reliable source of financial truth.