Akilo AI 3.0 — real-time telecom fraud intelligence


A deterministic scoring engine, not a black box: every alert Akilo raises can be replayed from its own recorded inputs and produces the exact same score, byte for byte — independently re-runnable accuracy via a real adversarial simulator, full decision provenance, and continuous assurance instead of a one-time benchmark. Built on a combination no other vendor in this market offers: GSMA Open Gateway identity signals, cross-MNO consortium intelligence, and MoMo-native detection — SIM swap, Wangiri, IRSF, MoMo fraud, and flash calls. Every number on this page is measured or fetched live — nothing invented.

Every decision independently reproducible11 deterministic detectors<1ms engine scoring — test-measuredk-anonymity + DP consortiumSandbox-validated, CAMARA-readyWCAG 2.1 AA — audited, 0 violations
0
Deterministic fraud patterns — one per real detector
<1ms
Engine scoring time (test-measured)
0hrs
Pilot setup target
Open
Now accepting pilot partners

What We Catch

Real-time detection across the full spectrum of telecom fraud vectors — one card per real detector in the engine.

Critical

SIM Swap

IMEI changes twice or more within 24 hours for the same subscriber

Signal: uniqueImeiCount ≥ 2 in 24h

High

Wangiri / One-Ring

Very short international calls at high frequency, designed to trigger a costly callback

Signal: intl && duration < 4s, high velocity

High

Revenue Share Fraud

High-frequency calls to premium-rate numbers

Signal: premium call velocity / 1h

Critical

International Bypass

Outbound international call rate exceeding 40% of traffic in a 1-hour window

Signal: intl ratio > 0.4 / 1h

Medium

Roaming Abuse

Over 80 roaming calls/day with a night-heavy pattern

Signal: roaming > 80/day, night-heavy

High

MoMo Burst

More than 15 distinct Mobile Money recipients within any 5-minute window

Signal: recipients > 15 / 5min

Medium

MoMo Night Fraud

Round-amount Mobile Money transactions between 2–5am local time

Signal: round amounts, 02:00–05:00

High

MoMo Multi-IMEI

Same device (IMEI) used across 3+ distinct subscriber numbers within 24 hours

Signal: IMEI across ≥ 3 MSISDNs / 24h

Medium

USSD Premium Abuse

Repeated USSD sessions to premium shortcodes at high velocity

Signal: USSD premium velocity / 1h

Critical

Subscriber Clone

Same IMSI active on different subscriber numbers within the same 30-minute window

Signal: IMSI on ≥ 2 MSISDNs / 30min

Medium

Flash Call Fraud

Bulk sequential sub-2-second calls from a single number — SIM farm pattern

Signal: duration < 2s, bulk sequential

Beyond Detection

What the platform does once a pattern is caught — real features, honestly labeled by maturity, not blended into one marketing claim.

Phase 1

Federated Intelligence Network

Tenants share aggregated fraud-signal statistics, not raw data. An indicator only ever publishes once at least 3 independent MNOs contribute to the same window — real k-anonymity, not a marketing claim — and calibrated differential-privacy noise is added before anything is exposed.

Live

Red Team Simulator + Pilot Results Report

Run a real adversarial simulation against the exact same production scoring engine your live traffic hits, then get a transparent precision/recall/F1 report — including attack patterns explicitly labeled "novel" specifically to test blind spots honestly, not to inflate a catch rate.

Beta

Agentic Sentinel

A real heuristic layer — Shannon entropy and coefficient of variation on transaction timing — flags automated, agent/bot-like fraud patterns. Explicitly tagged Beta everywhere it appears: real-world agentic-fraud training data is still limited, and we say so rather than overstate it.

Sandbox

GSMA Open Gateway (CAMARA)

SIM-swap and number-verification signals via the GSMA Open Gateway CAMARA APIs, feeding the Unified Trust Score alongside the deterministic engine and network intelligence. Sandbox-validated and CAMARA-ready today, with production access available on operator onboarding — never dressed up as a live aggregator connection before it is one.

Live

Bulk Historical CDR Backfill

Import a historical CDR archive through a subscriber-grouped, concurrent batch pipeline built for real telco data volumes — the same detection engine your live traffic hits, not a separate slower path bolted on for imports.

Live

Decision Provenance & Compliance Evidence

Every alert writes an immutable, content-hashed record: engine and ruleset version, every input feature, each detector's individual contribution. Re-run any decision from its own recorded inputs and get a byte-identical score back, or export a full Compliance Evidence Pack for a date range — logged, versioned, and reproducible, not a compliance certification.

What Makes Akilo Different

Not a claim to take on faith — every point below is something you can independently verify in the live product.

Live

Exactly reproducible, not a black box

Every alert writes an immutable decision record: the ruleset version in force, every input feature, each detector's individual contribution. Re-run any decision from its own recorded inputs and it produces a byte-identical score — a real recomputation, not a stored flag, so a tampered record is caught, not silently trusted. Most fraud-scoring vendors can't tell you why a score is what it is; ours can, on demand, for any alert.

Live

Verify it yourself, not our word for it

Run the Red Team Simulator — a real adversarial run against the exact production scoring engine your live traffic hits — and get a real precision/recall/F1 report back, computed from that run, not a published marketing number. Scheduled continuous runs (opt-in, default weekly) keep that accuracy independently re-verified over time, not a one-time benchmark, and every run's misses are disclosed exactly as recorded, never softened.

Phase 1

Three real signal layers, one platform

GSMA Open Gateway (CAMARA) SIM-swap and number-verification signals, cross-institution Federated Intelligence with real k-anonymity, and MoMo/USSD-native detection (momo_burst, momo_night, momo_multi_imei, ussd_premium) — deterministic patterns most fraud suites bolt on later, if at all — combined in the same scoring engine instead of three separate tools.

Beta

Limitations disclosed, not buried

Every feature here is labeled by real maturity — Live, Beta, Sandbox, or Phase 1 — not blended into one confident pitch. Agentic Sentinel says plainly that real-world agentic-fraud training data is still limited. GSMA Open Gateway is sandbox-validated and CAMARA-ready, not dressed up as a live aggregator connection. If a number isn't real yet, we don't show it.

Calculate Your Akilo AI ROI

Adjust the sliders to match your network. The result is a modeled estimate from your own inputs — assumptions disclosed inline.

Total active subscribers on your network

5,000,000
100,00050,000,000

Estimated fraud as % of revenue (industry avg: 2-5%)

3.2%
0.1%10%

Average revenue per user per month

$4.5
$1$20

How much fraud your current system catches

35%
5%90%

Your Annual Impact

Modeled estimate

Net Annual Savings

$0

ROI Multiple

0.0x

Payback Period

65d

Annual Revenue$270.0M
Fraud Losses$8.6M
Recoverable w/ Akilo AI$3.0M
Estimated fee (15% of modeled recoverable loss)$454K

Modeled from your slider inputs. The fee is gain-share — a percentage of documented recoverable loss, not a flat charge — typically 10-30% depending on the real deal. This is an estimate, not a measured or guaranteed outcome.

How It Works

The fraud engine scores every CDR in under a millisecond. Full request latency (network + database + scoring) currently runs 1-10 seconds depending on load — see our published benchmark methodology.

Ingest

CDRs flow in via SFTP, API, or real-time streaming from your switches.

Analyze

The deterministic fraud engine scores every call against 11 patterns in real-time; an optional ML layer is additive, never required.

Detect

Deterministic, rule-based scoring — every match is explainable, not a black box.

Block

Auto-trigger blocking rules, alert teams, and update the consortium feed.

See Akilo AI on Your Data

Get a personalized demo showing fraud detection on your network profile. No setup required.

Zero commitment. Full features. 48-hour setup target.

Start Your Free Pilot Today

Get full platform access for 14 days. See real fraud detection on your own CDR data. No credit card required. No obligations.

<1ms Scoring

Test-measured engine

11 Detectors

Deterministic, explainable

48hr Setup

Pilot target

Data Stays

On-premise option

01

Sign Up

2-minute form. No commitments.

02

Connect CDR

Secure SFTP or API drop.

03

See Results

Sub-millisecond fraud scoring on every CDR.

Free 14-day pilot · Full feature access · No credit card required