Block every bad call
before it costs you.

Shield AI v2.3.1 runs 9 simultaneous detection signals on every inbound call: voice fingerprint, call cadence, CID provenance, network heuristics, NLP transcript analysis, behavioral graph, device fingerprint, geo and time-of-day patterns, and a cross-publisher graph that tracks callers across your entire publisher pool. Signals combine into a composite fraud score at 23ms p95 latency with 99.4% accuracy.

The database is updated continuously from PACER court filings, carrier intelligence feeds, and verified industry reporting. Known litigators receive P0 priority — they are blocked before any routing decision is made, with no agent exposure and no payout trigger. You can also manually add numbers to the Caller Watchlist with a reason tag.

Every inbound call is matched against national and state DNC registries in real time. Match confidence is configurable — block at 99% confidence or flag at 90% for review. Configurable per program so Medicare networks and auto networks can have different DNC enforcement thresholds without affecting each other.

Yes. Shield Rules define exactly when each action fires — for example, throttle a publisher when their fraud rate exceeds 3% over 7 days, pause at 6%, block at 10%. Rules use WHEN / AND / OR / COUNT syntax and can reference any publisher metric. Status changes are logged in the publisher timeline with the triggering rule cited.

Stage 1 — Pre-call deterministic: static list checks (litigator DB, DNC, watchlist) run in under 50ms before routing. Stage 2 — Mid-call real-time: pattern rules (velocity, cadence) evaluated 50–200ms into the call. Stage 3 — Post-call aggregation: publisher-level metrics recalculated in batches every 30 seconds. Stage 4 — AI Composite: Shield AI synthesizes all 9 signals into a final fraud score at 23ms p95.