AML · sanctions · PEP · adverse media

Investigate faster.
Defend every decision.

Cut compliance costs 70%. Catch what humans miss.

Mizan AI is the compliance investigation workspace that screens names the way Gulf regulators actually think — transliteration-aware matching, explainable risk scores, and an agent that writes its own reasoning trace, so your file is regulator-ready before you are.

Book a walkthrough
14-day trial, no card goAML export ready OFAC · UN · EU · UK lists
The problem

Exact-match screening misses the names that matter.

The same person, six spellings — and six different "people" to a naive matcher. Meanwhile generic-vocabulary collisions bury real risk in noise. Both failures are structural, not operational.

The Mizan way

Weigh every name — mīzān means balance.

Mizan (ميزان) canonicalises before it compares, guards against vocabulary artefacts, and writes down how every score was weighed — in plain language, in the file. No black boxes, no noise, no surprises.

Capabilities

One workspace, from first hit to filed report

Everything between a suspicious transfer and a defensible disposition — screening, investigation, reporting — in a single ledger-clean interface.

Screening engine

Name intelligence that speaks Gulf Arabic

Canonicalises transliterations (Mohammad → Muhammad), strips diacritics, ignores corporate noise — then scores with Jaro–Winkler and token-set matching, guarded by a distinctiveness test.

NORM "SALEM AL FARSI" ⇄ "AL FARSI SALEM MOHAMMED"
similarity 0.97 · alias match
Risk model

Composite scores, factor by factor

Name, DOB, nationality, jurisdiction and behaviour combine into one 0–100 score with five bands — every factor's contribution itemised, every score reproducible from the same inputs.

DOB exact +8 · DOB mismatch −26
five bands · CRITICAL → MINIMAL
Investigation agent

The agent writes its own reasoning trace

Mizan's investigation agent reads the file the way a senior analyst would — and leaves a plain-language record of every step: what it normalised, which list it matched, which behavioural pattern it saw, and why it recommends a disposition. That trace is your audit trail.

AGENT INVESTIGATION_STARTED — 14 reasoning steps, engine v1.0
AGENT RISK_SCORED — composite 78 · HIGH
AGENT DISPOSITION_RECOMMENDED — escalate to MLRO
[ANALYST] PENDING_HUMAN_REVIEW — agent never files autonomously
Detection

Behavioural patterns, not just names

Beyond watchlists: structuring below thresholds, round-value bias, rapid pass-through, counterparty concentration, dormancy spikes and profile deviation — each a named, inspectable signal in the trace.

SIGNAL threshold_proximity +10   SIGNAL concentration +8
patterns read together · never a black-box flag
Case workflow

Queue, dossier, disposition

A risk-ranked alert queue; a dossier with score dial, factor breakdown, transaction timeline and reasoning trace; and one-click dispositions: close, request info, escalate to MLRO, file STR, freeze.

Reporting

Regulator-ready output

Filing narratives drafted by the agent, goAML-shaped XML export, and a complete event log. What you file is what the agent reasoned — nothing hidden, nothing invented.

How it works

From data to disposition in four moves

01 · INGEST

Connect your data

Transactions, counterparties, customers — via API, SFTP batch or direct file upload. Internal descriptors are filtered at ingestion, not screened.

02 · SCREEN

Mizan reads every name

Names are normalised, matched against your watchlist stack (OFAC, UN, EU, UK, PEP, adverse media, your own lists) and scored with the full factor model.

03 · INVESTIGATE

The agent builds the file

The investigation agent assembles the dossier, runs the reasoning trace and recommends a disposition. Humans review; the agent never files autonomously.

04 · REPORT

File with confidence

Export the narrative, the trace and the evidence in goAML-compatible XML — or keep the full audit trail in the case file for the examiner.

0%
fewer false positives vs exact-match screening
0×
faster average investigation time
0%
screening-engine uptime, 2026
0h
median time to first production screen
Field notes

What compliance teams say

★★★★★

We cut our alert queue from 2,300 to 640 a month. The distinctiveness guard alone removed an entire class of nonsense hits. The trace is what sold our MLRO — and later, the regulator.

YA
Yasmin Al-Amir
Head of AML, Khazana Exchange Co.
★★★★★

Our examiner asked how we could defend a cleared case. We opened the file and showed the trace: normalisation, similarity, DOB mismatch, done. She closed her notebook.

TJ
Tariq Jassim
Group Compliance, Al Fajr National Bank
★★★★★

The agent's reasoning trace turned our STR files from an afternoon's work into twenty minutes. It drafts the narrative from evidence it actually used. That's a different product category.

NE
Noura El-Sayed
Compliance Director, Dawaa Remit
Pricing

Simple plans. Serious screening.

Start free for 14 days. No card, no procurement theatre — production data the moment you're ready.

Starter

For teams finding their feet

Core screening for small compliance teams and early fintechs.

$99/mo
  • 1,000 screenings / month
  • 1 analyst seat
  • OFAC · UN · EU · UK watchlists
  • Agent reasoning trace
Start free trial
Enterprise

For institutions under the lens

Private deployment, custom lists, dedicated programme.

Custom
  • Unlimited screenings & seats
  • SSO / SAML · private cloud or on-premises
  • Your watchlists & typologies
  • 99.9% SLA · dedicated CSM
Talk to us

Every plan includes a 14-day free trial. Prices in USD, excluding VAT. Compare plans in full →

FAQ

Questions, answered

What watchlists does Mizan screen against? +
Standard plans include OFAC SDN & non-SDN, UN Security Council, EU Consolidated, UK OFSI, plus PEP and adverse-media sources. Enterprise customers add their own lists — internal blacklists, regulator advisories, or bespoke typologies — through the list manager or API.
How is Mizan different from a generic fuzzy-match tool? +
Three differences. First, transliteration awareness: the engine canonicalises Arabic-to-Latin spelling variants before comparing — the single largest source of Gulf-region false negatives. Second, the distinctiveness guard: a high fuzzy score built only from generic words ("trading", "marine", "holdings") is suppressed, because it isn't evidence. Third, explainability: every score is a sum of itemised factors, and the agent writes the reasoning trace. Nothing is a black box.
Can the agent make decisions without a human? +
No. The agent recommends dispositions and can auto-clear only the lowest-risk alerts (configurable threshold), always retaining the reasoning trace as the audit record. Anything at MEDIUM risk or above — and any escalation, STR or freeze — requires a named analyst. The event log records exactly who did what, and when.
How do we export filings to our regulator? +
Mizan generates goAML-shaped XML reports with the filing narrative drafted from the case evidence, plus a human-readable PDF of the dossier. For jurisdictions with different formats, the Enterprise tier supports custom templates. All exports carry the full audit trail (who, what, when, why) as embedded metadata.
Where does our data live? +
Cloud plans run in region-pinned environments (choose GCC, EU or US at onboarding) with encryption in transit and at rest, and are SOC 2 Type II audited. Enterprise customers can deploy fully on-premises or in a private VPC — the engine is deterministic and runs identically everywhere, with no data leaving your boundary.
What does implementation actually look like? +
Median time to first production screening is 48 hours: upload a file or call the API, map your fields to our schema, and you're screening. Our team holds a kickoff call, validates against a sample of your data, and tunes thresholds with you. No consultants, no six-month project plan.
14-day free trial

Put your queue on trial.

Sign up, upload a file, and see what your alert queue looks like when it stops shouting about "Falcon Trading LLC" and starts finding the pattern underneath.

Start free trial