Skip to main content

For controllers and finance teams

Settlement reconciliation that explains itself.

Sovyr ingests settlement files from every processor, matches them against your ledger, and hands your controller a reviewed exception queue — with the evidence behind every call.

A pilot starts with one month of settlement files and your ledger export. You keep your close process; we reconcile alongside it and you compare.

Money moves faster than the systems that account for it.

The problem

/0.1

What was authorized, what settled, and what got recorded stop agreeing.

$44.4B

Lost in U.S. retail and hospitality sales every year to payment disruption — with businesses reporting more than five major outages annually, and 63% of those landing in peak trading periods.

Study by Dynatrace, FreedomPay, and Retail Economics. Reported by Payments Dive.

Outages are the visible half. The invisible half is what happens after the money moves:

Multi-PSP timing differences

Every processor settles on its own calendar. A sale closes Friday, the payout lands Tuesday, the ledger entry sits somewhere in between — and the same transaction is in three periods at once, depending on which system you ask.

Fee and FX drift

Interchange, scheme, and processor fees are netted at rates that move. Cross-border sales settle at a rate nobody quoted you. The gross never equals the deposit, and the difference is never one clean number you can journal.

Chargeback and refund offsets

Reversals arrive weeks after the sale, netted into an unrelated payout, against an order that already closed. By the time the offset shows up, the period it belongs to is shut.

Month-end close slipping

Each of those becomes a line someone chases by hand. Close slips a day, then three. The exceptions nobody had time to chase quietly become write-offs — and nobody can say later why.

How it works

/0.2

Ingest. Match and explain. A human approves.

01

Ingest every settlement file.

Processor payout reports, gateway exports, bank statements, and your ledger — in whatever shape each vendor emits them. Sovyr normalizes them onto one transaction spine: same currency basis, same date basis, same identifiers. No CSV cleanup on your side.

02

Match deterministically, then explain what's left.

Rules settle everything rules can settle — amounts, references, dates, fee arithmetic. What survives that pass becomes the exception queue, and only there does AI read anything: it states what it believes happened to each unmatched item, and shows the source rows it read to get there.

03

Your controller approves. Nothing posts before that.

The queue arrives reviewed, ranked, and evidenced — each exception with a proposed treatment a person can accept, edit, or reject. A human approves every entry that reaches your ledger. Sovyr has no path that writes around them.

Why it's safe

/0.3

The controls aren’t settings. They’re the shape of the system.

Every guarantee below holds because of how the pipeline is built — not because a model was asked to behave. None of them has an off switch.

Deterministic matching runs first.

Amount, date, reference, and fee arithmetic are matched by rules — auditable, repeatable, identical on every run. The AI never touches a transaction a rule already settled, so the bulk of your volume is reconciled by arithmetic, not inference.

AI only ever sees the exceptions.

The model's entire working surface is the residue the rules could not clear. It cannot reclassify a matched transaction, adjust a settled figure, or reopen a closed period — those rows are never in front of it.

Every suggestion cites its source rows.

No explanation ships without the underlying records attached — this payout line, that ledger entry, this fee schedule. A reviewer verifies the claim against the evidence in seconds instead of taking it on faith.

Nothing posts without approval.

Sovyr proposes journal entries; a person accepts them. There is no autonomous write path into your ledger, no confidence threshold above which it acts alone, and no setting that turns the gate off.

Full audit trail, by construction.

Every match, suggestion, approval, edit, and override is recorded with who, when, on what evidence, and under which rule version. When an auditor asks why a number is what it is, the answer is a record, not a reconstruction.

That is the difference between a reconciliation system with AI inside it and a language model bolted onto reconciliation. One can show you why it believes something; the other can only tell you.

Methodology

/0.4

The AI-Powered Integrated Payment Framework.

Sovyr’s reconciliation work is one component of a four-part methodology for payment infrastructure — the AI-Powered Integrated Payment Framework (AIPF): transaction acceleration (TAC), financial reconciliation (FRC), payment security (PSC), and merchant integration (MIC). What’s on this page is the FRC, built as a product.

Component 2 — Financial Reconciliation (FRC)

Automated Settlement Discrepancy Detection Engine

The deterministic matching pass — continuously identifies inconsistencies across transaction stages.

Financial Reconciliation Accuracy Protocol (FRAP)

Standardized validation procedures, so consistency and auditability hold across records rather than per-analyst.

Real-Time Transaction Validation System

Discrepancies surface as they arise instead of accumulating until close.

Revenue Leakage Prevention Mechanism

The exception queue — the losses that hide in the gap, made explicit and assigned to someone.

Deployment

Deployment follows the framework’s phases: a diagnostic that establishes your baseline before anything changes, framework structuring against that baseline, pilot validation in a live environment, then full implementation with continuous monitoring.

Pilot targets — measured against your Phase 1 baseline

  • Reconciliation accuracy above 90–95% consistency.
  • 40–60% reduction in reconciliation-related discrepancies at full implementation.
  • Automated validation and monitoring covering 70–80% of transaction flows.

These are targets the pilot is designed to test against a measured baseline — not results claimed in advance.

Book a pilot

Bring one month of settlement files. See your exception queue.

We reconcile it alongside your existing close, then walk the queue with you line by line — every exception with the source rows behind it. You keep your process; you compare the two.

A person approves every entry before it reaches your ledger. Sovyr matches, explains, and proposes; the judgment and the sign-off remain your controller’s.