Workflow model

Turn policy into a workflow people can defend.

Clarivy uses typed, versioned workflow steps—not arbitrary automation—to keep every proof request, review gate and outcome explainable.

From signal to evidence outcome

  1. 1

    Signal received

    A customer risk engine, provider result, policy rule or analyst observation opens review work.

    Output: Scoped case event

  2. 2

    Policy evaluated

    Deterministic rules identify required evidence, review gates, owner and service objective.

    Output: Evidence plan

  3. 3

    Proof requested

    A clear, controlled request explains what is missing and provides a secure upload route.

    Output: Evidence request

  4. 4

    Evidence preserved

    Files and claims receive source metadata, hashes, extraction context and review state.

    Output: Evidence ledger entries

  5. 5

    Exception resolved

    A named reviewer verifies, rejects, follows up or escalates with a required note.

    Output: Append-only review event

  6. 6

    Outcome returned

    Clarivy sends a structured evidence state to the customer system; the customer owns the business decision.

    Output: Evidence outcome and packet

Policy Studio release loop

Administrators start from a reviewed template, use typed nodes and deterministic decision tables, run synthetic simulations, and publish an immutable version. Invalid or unsafe graphs cannot be activated.

Save draft, validate, simulate, request review, then publish a version. AI never auto-publishes; existing cases keep the policy version they started with.

Agentic regulated decision loop

Bounded agents prepare source-cited work inside the harness. Named humans own material case dispositions.

Use the institution’s systems of record

SystemAuthority retainedClarivy contribution
Loan origination systemApplication, credit decision, offer, disbursement and servicing.Non-STP review, RFI, exception resolution and a signed disposition callback.
KYB or merchant masterAuthoritative legal-entity and document records.Requirement mapping, document gaps, conflicts and reviewer disposition.
Screening and monitoring providerSearch data, match candidates, alerts and provider score.Alert triage, false-positive resolution, rationale and human approval.
Risk or fraud engineRisk models, scores, thresholds and source-system action.Explain the signal, collect the required material and return a controlled outcome.

Workflow building blocks

BlockPurposeControlled output
Signal ruleExplains why evidence work started.Severity, owner and required proof
Evidence requirementDefines acceptable formats, expiry and mandatory review.Specific evidence request
Decision tableEvaluates known facts with deterministic operators.Matched rules and review obligations
CommunicationRenders an approved email, SMS or portal message.Immutable message and delivery event
ReminderSchedules bounded follow-up with explicit stop conditions.Idempotent reminder event
Human review gateRequires a role, note and optionally a second reviewer.Verified, rejected, follow-up or escalated
Integration outcomeReturns evidence state to a customer-controlled system.Signed, replay-safe event

AI may assist

Draft a policy rule, explain a simulation trace, extract cited fields, identify a possible contradiction, or prioritize review work.

People retain authority

AI cannot publish a policy, approve an exception, close a material gate or make an adverse customer decision.

Continue reading

Integration guide

Connect risk engines, merchant portals and evidence providers to the controlled workflow.