BroadNet
Where this fitsAI product map
Carrier AI · inside operator networks

AI for messaging infrastructure.

Machine-learning fraud detection, routing optimisation, intent classification, and sentiment analysis for messaging at scale. The same models that run inside the BroadNet network — now available as composable services for the operators and enterprises that already trust the carrier.

Talk to AI engineeringAll AI products
Training corpus
300M+
Messages per day of real BroadNet network telemetry — across 200+ countries, 1,000+ direct operator connections, and 23 years of interconnect history.
Inference p99
<50ms
Languages
16
Production models
6
Operating since
2003
No PII ever leaves the customer tenant. Models train on hashed, aggregated network features only.
Carrier-grade AI inside the operator network
AI for carriers

Carrier-grade ML, inside the network.

Line-rate fraud detection, routing optimisation, and sender-ID anomaly scoring deployed inside operator networks — trained on telemetry from 300M+ messages a day.

200+
countries
1,000+
operator connections
99.99%
SLA
Talk to AI engineering
Inside the agent console

Watch a model resolve a ticket — in the customer's language.

Intent detection, a live tool call, a RAG lookup, and a resolved answer — in Arabic and English, from one conversational model. Below is a redacted, illustrative transcript.

  • Customer
  • Tool / RAG call
  • AI agent
Example · redactedSession
session: cx_7f21•••model: bn-convo-v4 · ar+en
detected: arintent: order_status
مرحبًا، أين طلبي رقم ‎#12345؟
→ nlp.detect_lang() → "ar" · classify_intent() → order_status (0.98)
→ rag.retrieve(kb="orders", ref="#12345")
→ order_status(#12345)
← 200 OK { status:"out_for_delivery", eta:"18:30", courier:"•••••" }
Resolved
طلبك ‎#12345 قيد التوصيل وسيصل اليوم قبل الساعة 6:30 مساءً.
detected: enintent: notify_subscribe
Thanks! Can you text me when the driver is close?
→ notify.subscribe(order="#12345", channel="sms", trigger="eta<10m")
← 202 Accepted { subscription:'sub_9f2c•••' }
Resolved
Done — I'll text you when the driver is about 10 minutes away.

Illustrative example — a redacted sample transcript, not a live customer session.

Built on real BroadNet telemetry

The data only a carrier can see.

Carrier AI is trained on the same network signals BroadNet has been collecting since 2003 — DLRs, route latencies, sender-ID histories, and two-way conversational flows. It's not a wrapper on a public LLM. It's models with ground truth.

Training corpus
300M+
Msg / day
Routing coverage
200+
Countries
Direct DLR feedback
1,000+
Operator connections
Ground-truth history
23 yrs
Of telemetry
Capabilities

Six models. All in production today.

Every model below is running live inside the BroadNet network right now — catching fraud, picking routes, and tagging intents on the 300M+ messages we process every day. The same models are available as services to operators and enterprises through the Carrier AI API.

Used in SMS Firewall

ML fraud scoring

Sub-second fraud probability for every inbound A2P message. Catches SMS phishing, AIT, grey-route artefacts, and trust-score anomalies before they hit the subscriber.

Used in SMSC Gateway

Smart routing

Reinforcement-learning router that picks the best operator path per destination, time of day, and content type. Optimises for delivery, cost, and DLR truthfulness simultaneously.

Routes/sec per node
30,000
Used in Sender ID Registry

Sender ID anomaly detection

Behavioural fingerprinting on alphanumeric senders. Flags hijacked IDs, spoofed brands, and look-alike registrations before they damage your customer trust score.

Coverage
200+ ctry
Used in Two-way messaging

Intent classification

Multilingual intent tagging for inbound replies — STOP, HELP, purchase, support, escalation. English + Arabic native, plus 14 additional languages tuned for MENA business contexts.

Languages
16
Intents
240+
Used in Campaign analytics

Sentiment analysis

Aspect-level sentiment scoring across two-way conversations. Surfaces the segments of a campaign that drove negative reactions so you can fix the message, not the audience.

Aspects
32
Polarity
5-class
Latency
<50ms
Used in SMSC + voice firewall

SIM-box & flash-call detection

Carrier-grade detection of grey-route revenue leakage — SIM-box origination, flash-call abuse, and bypass-OTT patterns scored at line rate. Feeds the same A2P monetization loop the firewall enforces.

Latency
<20ms
Network integration

The models sit inside the network, not next to it.

Model decisions are written back into the routing plane in real time — no batch jobs, no nightly retrains. The firewall, SMSC, WhatsApp, RCS, and voice connectors all consume inference from the same regional clusters with end-to-end p99 latency under 50ms.

  1. Layer 01
    Edge

    Inbound A2P / Two-way / Voice

    Every message, call event, and webhook is enriched in-flight with model features — sender reputation, content embedding, route history.

  2. Layer 02
    Models

    Carrier AI inference plane

    Six production models — fraud, routing, anomaly, intent, sentiment, conversational — served from regional inference clusters with <50ms p99 latency.

  3. Layer 03
    Network

    SMS Firewall · SMSC · WhatsApp · RCS · Voice

    Model decisions are written back into the routing layer in real time — no out-of-band batch jobs, no human-in-the-loop bottleneck.

POST · api.broadnet.me · fraud scoring
live
{
  "msg_id":     "01HGN…7QF",
  "sender":     "MyBank",
  "destination":"+971 50 ••• 4567",
  "features": {
    "sender_age_d":     412,
    "route_dlr_truth":  0.998,
    "content_emb":      "<f32x768>"
  },
  "decision": { "label":"clean", "score":0.0021, "model":"fraud-v4.2" }
}
Data governance

Carrier-grade AI means carrier-grade data handling.

Carrier AI is built inside a regulated carrier — not a model shop. The same compliance posture that lets us operate as a licensed messaging provider in 200+ countries governs every byte the models touch.

Data stays in customer infrastructure

No PII ever leaves your tenant. Models train on hashed, aggregated network features — never message bodies, never subscriber identifiers.

Full model audit logs

Every inference is signed and persisted with the feature vector, model version, and decision. Exportable for regulators, internal audit, and dispute resolution.

GDPR · CCPA · HIPAA · ISO 27001

Carrier AI inherits the parent network's certifications: ISO 27001:2022, ISO 9001:2015, ISO 22301 BCMS, SOC 2, plus NIST AI RMF validation and OWASP LLM Top 10 testing.

  • ISO 27001
  • ISO 9001
  • ISO 22301
  • SOC 2
  • GDPR
  • NIST AI RMF
Hashed feature pipeline Signed inference logs · 7-year retention On-prem / regional residency available
Carrier AI · engineering intake

Talk to AI engineering.

Tell us where you want a model — in your firewall, on your two-way traffic, or fronting your support queue. An AI engineer will scope the integration, share a benchmark on a slice of your real traffic, and stand up a pilot within two weeks.

Talk to AI engineeringAll AI products
$ ai@broadnet.me · pilot scoped within two weeks