Glass skyline of the Dubai International Financial Centre at dusk

Banking AI · CBUAE & SAMA delivery

Production AI for CBUAE- and SAMA-supervised banks — shipped with a model risk file that survives committee.

Fraud, AML, credit, customer intelligence. We run a free retrospective back-test on your historical alerts, disputes, or book — inside your environment — and share the lift vs your incumbent before SoW signature. Every model ships with the documentation pack supervisors expect to read.

Request a free back-testWhatsApp our banking lead

Pre-contract back-test on your alerts. Model risk file included.

AML alert queue · re-ranked

Model v2.4 · MRM-approved

Alert #AT-918341Rank 1 · 0.94
Entity graph: 3 nodes · 1 flagged PEP · Arabic-variant match
Alert #AT-918209Rank 2 · 0.81
Structuring pattern · 14-day window
Alert #AT-918047Rank 3 · 0.78
Sanctioned-jurisdiction adjacency

SHAP top featuregraph_degree_centrality · +0.31

Model ID · last validatedaml-rerank-v2.4 · 12 days ago

14
GCC regulated financial clients
41%
AML false-positive review-time cut (tier-1 reference)
32 ms
p99 fraud inference (digital-bank reference)
47 → 41
CBUAE thematic Qs pre-answered by our MRM template

The painful problem

18,000 alerts a month at 96 % false-positive. Themis-aligned review has flagged effectiveness.

The compliance floor cannot triage inside the regulatory SLA. The in-house data-science team has spent eleven months in MRM committee. CBUAE thematic review is on the calendar. The CEO has committed a customer-intelligence number in the earnings call. None of it survives a bad audit.

Failure mode 1

Model risk committee

Internal builds reach committee without the validation plan, fair-lending evidence, or challenger model the chair will demand. Eleven months of review is the median, not the exception.

Failure mode 2

Data-residency constraint

Hyperscaler shared-responsibility models do not satisfy CBUAE Cloud Computing Regulation. The data does not leave the bank. The engineering model must respect that on day one.

Failure mode 3

Alert volume

Rules-based monitoring fires faster than the compliance floor can triage. The supervisor reads effectiveness, not volume. Re-ranking on top of the incumbent is the lever that moves both numbers.

Use-case grid

Nine production plays across fraud, AML, credit and customer intelligence.

Each ships with a quantified lift on the bank's own data and an MRM evidence pack the committee can read on first pass.

AML alert prioritisation

Learning-to-rank layer on Actimize / SAS AML / Oracle FCCM / SymphonyAI. Tier-1 UAE bank reference: 41 % reduction in false-positive review time with no change to SAR rate.

41 % review-time reduction

Entity resolution & graph

Quantexa-pattern graph features with Arabic name-variant transformer. Resolves transliteration, patronymics, and beneficial-ownership chains across correspondent flows.

Real-time payment fraud

Sub-80 ms p99 inference integrated to UAEFTS / AANI / IPI / mada. Digital-bank reference: 32 ms p99, documented USD-equivalent losses prevented.

32 ms p99

Credit scoring on SME book

KSA tier-1 bank reference: AUC lift of 0.07 vs incumbent, IFRS 9 staging stability test passed on a 24-month rolling window.

+0.07 AUC

Early-warning & collections

Behavioural deterioration signals on retail and corporate books. Calibrated to your Basel III credit policy and downturn assumptions.

KYC document intelligence

Emirates ID, passport, and trade-licence parsing in seconds. Arabic OCR pipeline on Surya + PaddleOCR-Arabic.

Churn & next-best-action

CDP-agnostic feature store on Snowflake / Databricks / Teradata with SHAP-based explainability for relationship managers.

Arabic relationship-manager copilot

Self-hosted LLM grounded in your product and policy library. MSA + Khaleeji dialect handling for client briefings and call summaries.

Sanctions screening uplift

Re-rank layer on top of Bridger / Fircosoft / OFAC scanning, with fine-tuned matching for Arabic-script and transliterated names.

Integration mechanism

On top of the incumbent. Inside the bank perimeter.

We do not rip and replace. The bank keeps SAS / Actimize / Oracle FCCM / FICO Falcon / SymphonyAI. We add a learning-to-rank layer, a feature store, and an MRM evidence pack.

Entity resolution

Quantexa-pattern graph features

Feature store

Snowflake / Databricks / Teradata

Payment switch

UAEFTS · AANI · IPI · mada

Explainability

SHAP, monotone constraints

Layer 1
Source systems
Core banking (T24, Finacle, Flexcube, in-house), card switch (UAEFTS, AANI, IPI, mada), CRM (Salesforce / homegrown), KYC and document store.
Layer 2
Feature store
Snowflake, Databricks or Teradata feature store with lineage, freshness SLAs, and protected-attribute proxy tagging.
Layer 3
Inference & re-ranking
Learning-to-rank on top of Actimize / SAS AML / Oracle FCCM / FICO Falcon / SymphonyAI. Sub-80 ms p99 on the payment switch for fraud.
Layer 4
MRM artefacts
Model card, development document, validation plan, fair-lending checklist, monitoring KPIs, challenger model. Refreshed on a documented cadence.
Layer 5
Operations & monitoring
Drift detectors, retraining schedule, alert volume guardrails, and quarterly model validation reports for the supervisor.

Regulator & standards

Designed for the reviewers in the room.

CBUAE supervisory expectations. Basel III RWA discipline. IFRS 9 staging. FATCA / CRS. AML / CFT Federal Decree-Law 20. Model risk management aligned to SR 11-7 / SS1/23. We bring the evidence; the supervisor reads it.

CBUAE & Cloud Computing Regulation

Every reference architecture maps to CBUAE Cloud Computing Regulation expectations. Data-residency and key-management posture are documented for the supervisor before the bank signs the SoW.

Basel III & IFRS 9

Credit models ship with IFRS 9 staging stability evidence (24-month rolling), downturn calibration, and an RWA inflation sensitivity. The development document is structured for IRB / SA defence.

SR 11-7 / SS1/23 / MRM

Model risk file template includes development document, validation plan, fair-lending and bias review, monitoring KPIs, challenger model description, and a CBUAE supervisor-question response matrix. The same template that has cleared committee in a UAE tier-1 bank.

PCI DSS & ISO posture

Fraud inference appliance carries PCI DSS attestation. ISO 27001, ISO 27017, SOC 2 Type II in place. Sub-processor list and DPA aligned to UAE PDPL plus DIFC and ADGM data-protection regimes where the bank operates a regulated arm.

Reference engagements

Three anonymised outcomes — reachable on a reference call.

UAE tier-1 · AML

41 % false-positive review-time reduction

Learning-to-rank layer on existing AML monitoring. No change to SAR rate. Full lineage to the CBUAE supervisor questionnaire. Customer reachable under NDA.

KSA tier-1 · Credit

+0.07 AUC lift on small-business book

IFRS 9 staging stability test passed at a 24-month rolling window. Fair-lending checklist signed off by the bank's MRM committee. Validation pack cleared at first review.

Digital bank · Fraud

32 ms p99 instant-payment fraud inference

Integrated to the payment switch. Documented USD-equivalent losses-prevented figure. Model-card and PSD-style customer-impact metrics shipped on day one.

Differentiation

Brocode vs the four archetypes your evaluation team is already weighing.

The lead magnet includes a CBUAE supervisor-question response matrix that walks an MRM committee through this comparison.

Capability	Big-4 risk practice	SAS / Actimize / Oracle FCCM / FICO	Offshore SI (India / near-shore)	In-house build
Pre-contract back-test on your historical data
Model risk file aligned to CBUAE / SR 11-7 / SS1/23	Partial			Partial
Sits on top of incumbent (no replatform)
Arabic name variant entity resolution		Partial
UAE-resident engineers, VDI-only delivery				Partial
Sub-80 ms p99 fraud inference on payment switch	Partial		Partial
Production code, not advisory deliverable
Fair-lending checklist & monotone constraints	Partial			Partial
Named engineers, no offshore sub-contracting				Partial

Free download

Model Risk File for CBUAE-Supervised Banks

A 36-page template with worked examples for fraud, AML and credit models. Development document outline, validation plan, fair-lending checklist, monitoring KPIs, challenger-model framing, and a CBUAE supervisor-question response matrix. Headline figure: the average CBUAE thematic review asks 47 model-risk questions; this template pre-answers 41 of them.

Three-lines-of-defence operating model
AML / fraud / credit model evidence (worked examples)
Bias and fairness test pack
Model card and datasheet
Periodic validation playbook
CBUAE supervisor-question response matrix

MRM & delivery FAQ

The eight questions every MRM chair raises.

If yours is not here, raise it in the form below. We answer in writing before the first call.

Yes — that is the only acceptance criterion that matters to us. Every model ships with a development document, validation plan, fair-lending and bias review, stability monitoring plan, and challenger model description. The pack maps explicitly to CBUAE expectations and to SR 11-7 / SS1/23 thinking, and the average CBUAE thematic review asks 47 model-risk questions — our template pre-answers 41 of them. We share a redacted sample on the first call under NDA.

Start the conversation

Request a free back-test on your alerts or your book.

Tell us the regulator, the priority use case, and the existing platform. We come back with the back-test methodology and the lift figures from comparable engagements on the first call.

Prefer WhatsApp? Message our banking lead directly.

Continue exploring