BroadNet
For mobile network operators

Stop A2P SMS fraud at the operator edge.

Fraud detection built on machine learning, real-time monitoring, and trustworthy messaging channels. The same engine that runs in our 60+ operator firewall deployments.

Fraud intelligence
BroadNet · fraud protection
Fraud intelligence
200+
countries
1,000+
operator connections
99.99%
SLA

The platform

Six pillars of operator-grade fraud protection.

Real-time fraud detection

Proprietary AI scoring of every inbound A2P submission. Catches sender-ID spoofing, smishing campaigns, grey-route artefacts and trust-score anomalies before they hit the subscriber.

Smart analytics + pattern recognition

Behavioural fingerprinting tuned per operator. Surfaces suspicious traffic patterns the rule-pack misses, learns from your network's own ground truth, and exposes the verdict shape in a dashboard your fraud + RA teams can audit weekly.

Retrain
Nightly

Spoofing + smishing protection

Blocks impersonation attempts before they reach handsets. Verifies sender identity against the registry, filters malicious URLs at line rate, and applies content-based heuristics tuned for the regional dialect.

URL filtering
Line-rate
Languages
16
Block scope
Pre-SMSC

Global compliance baked in

International standards (GSMA IR.71, IR.81, IR.82) plus market-specific frameworks (TDRA, CST, NTRA, TRC, NCC, Ofcom) enforced before delivery. Default-deny on failed checks; audit-grade log every decision.

GSMA-aligned
IR.71/81/82
Markets
200+
Audit log
Immutable

Custom security controls

Flexible configuration of inspection protocols, alert thresholds, and filtering rules — tuned to your operator's risk profile, traffic shape and regulator obligations. Per-bind, per-sender-ID, per-route.

Policy granularity
Per-bind
Thresholds
Adjustable
Pin/veto
From dashboard

Smart reporting + alerts

Real-time threat monitoring with detailed insight into fraud attempts and the cost basis they would have carried. Data-driven analytics power weekly RA reviews and quarterly board reads.

Refresh
Real-time
Export
SIEM + CSV
Alerts
Webhook + email

What we stop

Six fraud patterns,
one verdict engine.

The fraud landscape doesn't respect category boundaries and neither do we. Every pattern below is scored by the same model on the same submission — there's no rule-pack switch to flip.

Fraud pattern

Sender-ID spoofing

Impersonators register look-alike alphanumerics to steal trust. We fingerprint each registered sender and flag look-alikes before traffic flows.

Fraud pattern

Smishing (SMS phishing)

Malicious URLs and credential-harvest content embedded in promotional A2P. URL-level inspection + content classifier block before delivery.

Fraud pattern

AIT (artificially-inflated traffic)

Bot-generated OTP request floods designed to extract revenue from CPaaS rebates. Behavioural model flags impossible-cadence patterns in real time.

Fraud pattern

Grey-route artefacts

Traffic that bypasses the SMSC via aggregator hops. Heuristic-based detection from delivery-receipt inconsistencies + DLR-truthfulness scoring.

Fraud pattern

Sender-ID hijacking

Authorised sender IDs reused for off-policy content. Per-sender-ID drift scoring spots this within minutes of the first off-policy send.

Fraud pattern

Bypass via OTT-to-SMS

OTT services rebroadcasting SMS-equivalent content via web push to bypass operator monetisation. Flagged at the signaling layer.

Where it sits

Seamless integration with the SMSC + signalling stack.

Integration

SMPP 3.4 + 5.0 in front of the SMSC

Transparent SMPP front-end. Preserves message ordering, DLR pass-through and CDR semantics. No SMSC config change required.

Integration

SS7 + Diameter signalling firewall

Inline at the STP for SS7 (MAP / SMS) and the DRA for Diameter. Inspects category-1/2/3 traffic per GSMA recommendation and enforces the cat-3 silent-drop policy several MENA regulators require.

Integration

Mirror or inline deployment

Most rollouts start in passive-mirror mode on a tap for 7–21 days while RA teams set baseline thresholds. Nodes flip to inline at 5×9 availability once the policy lock is signed.

Integration

Per-tenant policy engine

Multi-tenant on day one. Operators with multiple aggregator estates can apply different policies per bind without infrastructure duplication.

Inside the verdict engine

Two lanes, evaluated in parallel, fused into one verdict.

A deterministic rule engine and a probabilistic behavioural model score every submission at the same time. Rules give explainable guarantees; the model catches what rules can't express. Neither blocks the other — their outputs are fused, so a hard deny is always honoured and a novel attack still gets caught.

Deterministic lane

The rule engine

Hard, explainable verdicts evaluated on every submission at line rate. Rules never wait on the model and the model never blocks a rule — the two lanes run in parallel and their outputs are fused, so a deterministic deny is always honoured and a novel attack the rule-pack has never seen still gets caught by behaviour.

  • Sender-ID registry match, allow-lists and deny-lists per operator
  • GSMA IR.71 / IR.81 / IR.82 category enforcement with cat-3 silent-drop
  • Per-bind, per-route and per-sender-ID policy granularity
  • Line-rate URL extraction and content heuristics tuned per dialect
Probabilistic lane

The behavioural model

A scoring model trained on each operator's own labelled ground truth — RA disputes, complaint queues and regulator advisories — blended with anonymised signal from BroadNet's global feed. It fingerprints how a legitimate sender behaves and flags the drift a rule can't express: impossible cadence, trust-score decay, look-alike sender registration and grey-route artefacts.

  • Per-sender-ID and per-route behavioural baselines
  • Nightly retraining on your network's own ground truth
  • Drift and trust-score scoring for hijacked sender IDs
  • Explainable feature attribution behind every flag

Signals that feed the score

Six features behind every flag.

DLR truthfulness

Delivery-receipt semantics are cross-checked against real handset behaviour. Grey routes that fabricate DLRs to look legitimate diverge from the truth and score up.

Inter-arrival cadence

OTP and A2P submissions carry a natural rhythm. Bot-driven AIT floods and pumped traffic show impossible inter-arrival distributions the model detects in real time.

Latency + routing signature

Submit-to-DLR latency and hop signatures betray aggregator detours. Traffic that should be one hop but behaves like three is a grey-route tell.

Sender fingerprint drift

Each registered sender ID has a content, timing and volume fingerprint. Off-policy drift — new URLs, new templates, out-of-hours volume — surfaces within minutes of the first bad send.

Signalling corroboration

MAP and Diameter signals from the STP and DRA corroborate the application-layer verdict — SRI-SM anomalies and category-3 traffic reinforce or temper the score.

Revenue-impact weighting

Every flag carries a revenue-leakage estimate from the operator's own rate card, so fraud and RA teams triage the largest leaks first instead of chasing low-value noise.

24/7 expert support.

Round-the-clock implementation, optimisation and incident-response — staffed by the same fraud ops team that ships hot signatures within minutes of a new pattern appearing on the global feed.

Hot signatures

Pushed in minutes

Coverage

Follows the sun

Incident SLA

Sev-1 hotline

Tenancy

Per-operator + RA-scoped

Operator due diligence

Questions operators ask.

How is behavioural ML different from a static rule pack?

A rule pack encodes threats you already understand — known senders, known bad URLs, GSMA categories. It is fast and explainable but blind to anything novel. The behavioural model learns your network's own traffic shape and flags the drift a rule can't express: impossible cadence, trust-score decay, look-alike registrations. BroadNet runs both lanes in parallel and fuses the verdicts, so you get deterministic guarantees and adaptive coverage at once.

Does inline scoring add latency to A2P delivery?

Scoring runs on an in-memory path and the deterministic and probabilistic lanes evaluate concurrently, so the added latency is a low single-digit-millisecond budget rather than the sum of two passes. Message ordering, DLR pass-through and CDR semantics are preserved. Most operators start in passive-mirror mode and only flip nodes to inline once the latency budget is validated against their SLAs.

How do you detect grey-route and SIM-box-origin A2P?

Grey routes leave artefacts: fabricated or inconsistent DLRs, latency and hop signatures that don't match a direct interconnect, and content that impersonates registered brands. The model scores DLR-truthfulness and routing signature together with signalling corroboration from the STP, so traffic that bypassed the SMSC via an aggregator hop or a local SIM box is flagged even when the sender ID looks clean.

How do you keep false positives low on legitimate roaming and marketing traffic?

The baseline is per-operator and per-sender-ID, so the model measures drift against your own legitimate traffic rather than a generic template. Roaming, seasonal marketing peaks and new campaigns are learned, not punished. Every flag ships with feature attribution your RA team can audit, and thresholds are adjustable per bind so you tune precision to your regulator's stance.

What does the deployment model look like — mirror or inline?

Both. Rollouts start in passive-mirror mode on a tap at the SMSC and signalling edges for a 7-21 day learning window while your RA and fraud teams set baseline thresholds. Nodes flip to inline at 5x9 availability once the policy lock is signed. The engine is multi-tenant on day one, so operators running multiple aggregator estates apply different policies per bind without duplicating infrastructure.

How does anti-fraud protect A2P revenue rather than just block spam?

Blocked fraud is only half the value. Detected grey-route senders are routed into your A2P monetisation funnel so unbilled enterprise traffic becomes billed enterprise traffic, and AIT floods that would extract CPaaS rebates are stopped before they inflate your termination costs. Each detection carries a revenue-leakage estimate so finance can quantify recovered value.

How does it fit our SMSC, SS7 STP and existing firewall?

It sits as a transparent SMPP 3.4 / 5.0 front-end to the SMSC with no SMSC config change, and inline at the STP for SS7 (MAP/SMS) and the DRA for Diameter. It complements rather than replaces an existing firewall — many operators run it alongside their SMS and signalling firewalls, sharing the sender-ID registry and consuming the same policy engine.

Is every decision auditable for our regulator?

Yes. Every verdict is written to an immutable, audit-grade log with the fused score, the contributing signals and the policy that applied. The engine enforces market-specific frameworks (TDRA, CST, NTRA, TRC, NCC, Ofcom) alongside the GSMA IR series, defaults to deny on failed checks, and exports to SIEM and CSV for weekly RA reviews and regulator inquiries.

Real-time operator network protection
Operator-grade

Protect and monetize at the signaling layer.

In-network SMS and SS7 firewalls stop grey routes and fraud at line rate while legitimate traffic flows untouched.

<5ms
Inspection
99.99%
Uptime
1,000+
MNO links

Eighteen operators already shipping it.

Operator briefing on the deployment model, integration timeline, and commercial structure is available on request.