Real-Time Decision Runtime

Every transaction scored before it clears

The Decision Runtime scores inbound transactions in under 150ms, assigns each one to a risk band, and routes it — auto-clear, review, or block — with no downstream surprises for analysts or operations teams.

< 150ms

p99 Decision Latency

end-to-end per transaction

96.7%

Queue Precision

analyst review efficiency

95%+

Recall at Alert Threshold

across fraud and AML use cases

3 Lanes

Routing Outcomes

auto-clear · review · auto-block

How it works

One decision, three possible outcomes

From arrival to routing in a single pass. Every transaction moves through signal extraction, scoring, and threshold evaluation before the decision is committed.

Transaction Arrives

payment rail / batch ingest

Signal Extraction

balance · velocity · network · timing

Risk Scoring

calibrated probability output

Threshold Evaluation

operator-set policy

Auto-Clear
85%
Review Queue
3%
Auto-Block
< 1%

Decision Bands

Three lanes. No grey area.

Every transaction is assigned to exactly one lane. Thresholds are set and versioned by your team — the runtime enforces the policy you specify.

Auto-Clear

Low-risk transactions pass through instantly. No queue entry, no analyst time, no friction at the point of payment.

Typical share~85%
Latency< 80ms
Review Queue

Borderline transactions are held for analyst review. Each entry carries calibrated risk evidence so investigators can triage without guesswork.

Typical share~2–3%
Avg. queue size~500 / window
Auto-Block

High-confidence fraud signals are stopped before settlement. No analyst involvement required — threshold is operator-controlled and version-tracked.

Threshold precision96.7%+
Recall95.0%

Capabilities

Built for production finance operations

Streaming and Batch Modes

Score transactions as they arrive on the payment rail, or submit batches for overnight processing. Both modes use the same decision engine with the same threshold policy.

Calibrated Risk Scoring

Each transaction receives a probability score that reflects actual observed frequency — not a raw classifier output. Scores map directly to risk bands without manual tuning.

Multi-Signal Fusion

Balance patterns, velocity signals, timing anomalies, and network topology are combined into a single coherent decision. No siloed rule engine to maintain separately.

Operator-Controlled Thresholds

Fraud and compliance teams set the threshold that governs when a transaction enters the review queue. Changes are versioned, gated, and reflected in the audit trail immediately.

Performance Monitoring

Decision accuracy is tracked across time windows and customer cohorts. Drift is surfaced before it reaches an examiner — not discovered during a regulatory review.

Consistent Peak-Load Behaviour

Latency and throughput hold steady across end-of-month spikes and intraday payment surges. No capacity pre-warming, no degraded mode.

Who operates it

Designed for three teams, not just engineers

Fraud Operations

  • Configure and version the threshold that governs the review queue
  • Monitor decision accuracy across transaction cohorts
  • Adjust scoring policy without involving engineering

Compliance Officers

  • Every decision carries a risk score and routing rationale
  • Review queue entries include the evidence behind the score
  • Audit trail is generated automatically — no post-hoc assembly

Platform / Technology

  • Single integration point for both streaming and batch workloads
  • Deploy within the bank's own network perimeter
  • Stable API surface with versioned decision contracts

See the runtime against your transaction volumes

We run a live walkthrough with transaction data that matches your use case — fraud, AML, or both.