A modern risk platform for money movement
Every transaction scored. Every decision owned.
AI scores. Your policy decides. Your people review.
Every transaction, entity, counterparty, agent, and program on your rails gets a Skoor: a number from what was observed, with its n, and null where nothing was. One place to see it, score it, and decide, so growth in programs, rails, and countries never outruns control.
Now onboarding design partners.
The problem
Banks grow faster than their risk teams can read.
A modern bank runs its own core, its own rails, and hundreds of programs on top of them. The picture of what is safe is still assembled by hand: exports, sheets, and the memory of whoever has been there longest. Every new program, rail, or country adds another. Skoor turns that picture into a product.
AI scores. Your policy decides. Your people review.
AI built for money movement.
Built by people who have run payments and compliance in production. Skoor’s AI drafts the decision, a person signs it, and every signed decision teaches the next version.
Always on, ready to review.
Skoor works the queue around the clock: scoring every transaction as it lands, drafting the narrative, holding what must be held, and closing only what the policy allows it to close.
Trained on your policy and your record.
Like any good analyst, Skoor gets sharper with review. Every decision a person signs, and every draft a person overturns, joins the labeled set the next version is measured against.
Knows when to act, and when to ask.
The line between automated and reviewed is written, versioned, and never widens on its own. If it is suspicious, a person is always in the loop.
The platform
Risk as infrastructure, in six layers.
Each layer is a product surface with its own API, its own record, and its own metrics. Together they are the control plane a modern bank grows on.
Every transfer, return, hold, verification, card decision, and overdraft event from your rails, keyed by program, entity, account, and counterparty. Signed, idempotent, append-only. The event store is the source of truth.
Return rates against network thresholds, reserve coverage, manual-review aging, velocity, counterparty concentration, verification denial rates — per program, per entity, rolled up into a Program Skoor with the observation count on every number. Null where nothing has been measured.
Which decisions run on their own, and which wait for a person. A versioned policy draws the line. Two approvers on any freeze or limit. No threshold moves until a replayable test set shows it is safe.
Clear or cancel a held transfer, pause a card, freeze an account, send a request to a program, assemble a periodic review — through your bank's own API, with your own keys, every action attributed to the person or policy that took it.
Every disposition hash-chained with its narrative, its evidence, the policy version, and who decided. Monthly metrics, a board pack, an examiner export. The record is what lets you approve faster next time.
The same platform, opened to the businesses that bank on you: their KRIs, their open requests, their queue — in their own tools, or inside Claude. What protects the bank protects every business built on it.
Automated or reviewed
Ordinary, or the start of a pattern?
Most transactions are exactly what they look like. The few that are not look ordinary at first. Skoor scores each one against what the program, the counterparty, and the entity have done before, and draws a written, versioned line between the decisions the platform may make alone and the ones that wait for a person.
The default policy. Yours is written with your compliance and legal leads, versioned, and applied the same way to every alert until the next version. The automated side never widens on its own: a decision moves from reviewed to automated only when the test set shows it is safe to.
Your policy, applied
Your policy, applied the same way every time.
Not a habit and not a hunch: one written line, one version, one record of every decision made on either side of it.
Routine alerts the policy allows are closed by the platform and logged. Anything risky or ambiguous is drafted and put in front of a person, fast.
Every alert carries its narrative: what happened, what the evidence shows, what was checked, what is recommended, and which line of the policy applies.
The transaction, its Skoor and signals, the entity, the counterparty, the program’s KRIs, and every prior decision on the same actors, on one page, before anything moves.
The overview is a pulse on every program: KRIs against their thresholds, review queues and how old they are, actors in the hold band, and the six numbers the platform reports on itself.
Signed decisions and overturned drafts become labels. The metrics show where the line between automated and reviewed should move, and the next policy version is measured before it ships.
What proves it works
Every transaction gets a Skoor. The metrics prove the Skoor is right.
The Transaction Skoor is a number from 0 to 100, higher is better, computed from what the platform has observed: the program’s history, the counterparty’s history, the entity’s verification and screening state, velocity, amount against expected activity, and the rail. It carries its inputs, its policy version, its confidence, and the observation count behind it. Where the history is too thin, it is null.
Program KRIs roll up the same way into a Program Skoor, so the board pack and the transaction queue are reading the same number.
Six numbers the platform reports on itself.
For each Skoor band, the share of transactions later returned, disputed, or escalated. The platform works when higher bands go wrong more often, and the curve holds month over month.
Clear 17% · n=36 · Review 63% · n=8 · Hold 100% · n=2
Share of transactions and partners with a Skoor rather than null. Unscored volume is the honest measure of how much history the platform still lacks.
Transfers 100% · n=2601 · Programs 100% · n=4
Of the Hold band against labeled outcomes, per policy version. A threshold does not move until the labeled set says it should.
Precision 100% · n=2 · Recall 15% · n=13
Share of alerts closed under policy without a person, and the share a person later disagreed with on review.
Auto-closed 0% · n=983 · Overturned null · n=0
Median and tail, from event to decision, for the automated queue and the reviewed queue separately.
Automated median null · n=0 · Reviewed median null · n=0
Programs, countries, and dollars cleared this month that would otherwise have queued for a person.
2,117 transfers · $2.32M
Every number carries its n. Where nothing has been measured, the platform says null. Nothing is invented. The numbers above are read live from the demo tenant: simulated, Column-shaped data, 30-day window, as of 2026-10-08. See the metrics page.
Fit
Built to fit a bank’s stack, not to replace it.
A bank that built its own core did not do it to add middleware. Skoor is one slice: it reads what the stack already emits, acts through the API the stack already has, and leaves the ledger alone.
Signed webhooks, verified on the raw body, deduplicated by event id, safe under redelivery. No new integration surface for your engineers.
The platform runs against your test environment until you say otherwise. Actions go through your own API with your own credentials, never ours.
No response, no policy, no evidence: the answer is a hold, not an approval. The same default your authorization webhook already lives by.
Stablecoin transfers are screened in the same queue as ACH, wire, real-time, and card, with the same record and the same boundary.
Where the Skoor works
One risk score for every way money moves.
The same Skoor, the same bands, the same record, whichever rail the money takes and whichever program sends it.
Banks that own their rails.
Direct access to Fedwire, ACH, and real-time payments removes the correspondent’s buffer: no intermediary reading the wire, no next-day window to catch it. Risk has to run at the speed of the rail. Skoor scores every transfer as it lands, holds what needs holding before settlement, and keeps the record examiners expect, without putting a middleman back in the path.
Real-estate money flows.
Closings, escrow, earnest money, rent: large wires to counterparties that are new by design, such as title companies, attorneys, and sellers seen once. A Skoor that knows a closing wire from a compromised-email wire reads the counterparty’s history across every program on the bank, the entity’s verification, and the amount against what was declared, and asks a person only when those disagree.
Programs on top of the bank.
Payroll, marketplace payouts, lending, remittance: each program has its own return rates, reserve coverage, and counterparties. The Program Skoor rolls their KRIs up so the board pack and the transaction queue read the same number, and each program can see its own.
Disposition narratives are drafted by Claude and signed by a person. Claude carries SOC 2 Type 2, ISO 27001, HIPAA, and GDPR. We add the controls your examiners ask about.
Commercial Claude usage is billed by Anthropic, and by default Anthropic does not use data from commercial deployments to train its models.
Those are Anthropic's, for Claude. Anthropic Trust Center →
Growth that never outruns control.
A walkthrough is forty minutes: your rails, your programs, and the decisions your team makes by hand today, mapped onto the six layers. You leave with the line between automated and reviewed written down.
Claude is a product of Anthropic, PBC. Skoor is an independent company and is not affiliated with or endorsed by Anthropic or by any bank.