Crayon Data Crayon Data Tangram AI
Controls & Anomaly Detection | Fully agentic | ✷
BANKING · OPERATIONS MANAGEMENT SUITE
C O N T R O L S   A I

Runs every operational control on 100% of the transaction population - across payments, cards, GL and trade -
turning periodic sample checks into continuous, evidenced assurance.

<60s breach to alert ✦ 100% of population checked ✦ 100+ controls live at ADIB
Why Controls AI exists

Thousands of transactions. Zero blind spots.

The problem

Risk is compounding, unseen

More transactions move through the bank every quarter, and only a sliver of them are ever actually checked.

  • 2–5% of transactions ever sampled
  • Days–weeks before a breach is even seen
  • 3× op-risk losses outgrowing revenue since 2019
Today, it is done manually

Spreadsheets and email chains

An analyst samples a batch, logs each exception by hand and chases it over email and calls.

  • ~800 hrs of manual control effort a month
  • Weeks of audit & inspection prep per cycle
  • Untracked exceptions, chased over email and calls
With Controls AI

Every transaction, every run

The engine checks every transaction as it happens and raises only the exceptions that need a human.

  • 100% of the population checked, continuously
  • <60s from breach to alert
  • ~80% less manual control effort
What it does · three moves

Detect it. Understand it. Close it.

01 Control Engine Run the rules, on every transaction
  • Reconciliation, data-quality and TAT controls
  • Executed on the full population, not a sample
  • Codified once, reused across departments
02 Anomaly Detection Catch what a rule alone misses
  • Rule, threshold and pattern monitoring combined
  • Surfaces known and emerging deviations
  • Classified Critical / High / Medium / Low
03 Workflow & Case Management Route it, evidence it, close it
  • Auto-assigned to the right owner
  • Maker–checker approval, every step time-stamped
  • Closure evidenced, ready for audit on demand
Design → Assure

Nine stages. One human gate.

01 Identify

Process discovery across departments — where controls should exist.

02 Assess

Live risk scoring puts every candidate control in order.

03 Define

Described in plain English, drafted by the agent into a reusable, parameterised rule — REC-xx, AUTH-xx.

04 Map Data

Core banking, payments, cards, GL and treasury, normalised into one model — table relationships discovered automatically.

05 Execute

Runs on the full transaction population, not a sample.

06 Detect

Rule and pattern monitoring surface known and emerging deviations.

07 Investigate

Auto-assigned by severity — Critical to Low — with the evidence already attached.

08 Remediate Human gate

Maker–checker approval, time-stamped and SLA-tracked end to end.

09 Report

Live dashboards for management and auditors, tagged to Basel, COSO, IIA and CBUAE, always current.

Dashboard

Ask it. It already knows.

live at Abu Dhabi Islamic Bank

  1. 1
    Ask it directly Natural language in, anomalies and trends out.
  2. 2
    Insights, always current 1,241 anomalies across 84 workflows, auto-updated.
  3. 3
    High priority, flagged Nothing waits for someone to notice.
1 2 3
Data Ingestion

Any core system, one model.

15 tables  ·  zero manual mapping

  1. 1
    Connects directly No ETL, no flat files, no fragile pipelines.
  2. 2
    The AI ERD Generator maps it Every table relationship, discovered automatically.
  3. 3
    One quality-checked model Consistent definitions and lineage for every control after it.
1 2 3
Data Modelling

Describe it. It writes the SQL.

prompt to executable SQL, in minutes

  1. 1
    Two steps, not months Data model, then control check — one flow.
  2. 2
    Plain English in What are the inward transactions pending without checker approval?
  3. 3
    Executable SQL out Generated, validated, ready to run.
1 2 3
Controls

Every control, always running.

84 controls  ·  run on schedule

  1. 1
    One card per control Checks implemented, anomalies found, last run, status.
  2. 2
    Nothing sampled Every run covers the full transaction population.
  3. 3
    Filter to what matters By status, severity, or data model.
1 2 3
ICD Dashboard

The exposure, in one screen.

₹50.7L exposure flagged  ·  10 critical

  1. 1
    Fourteen controls, one department Total controls, anomalies and exposure, at a glance.
  2. 2
    Where the breaks concentrate Department-wise anomaly counts, ranked.
  3. 3
    Critical, High, Medium, Low The criticality split, always current.
1 2 3
Anomaly Detection · Review

The mismatch, and the proof.

anomaly #1826  ·  medium severity

  1. 1
    Held until reviewed PO 202600019, flagged and open.
  2. 2
    The evidence, already assembled PO amount $24,250 against a GL amount of $24,000.
  3. 3
    Nothing to reconstruct Table, check ID and detection timestamp, captured automatically.
1 2 3
In summary

Every transaction checked, on the record, every time.

It checks 0% of the transaction population, continuously —
turns a breach into an alert in under 0s, already runs 0+ controls live
— and saves 0+ man-hours a month across all 0 pipeline stages.

It is live at Abu Dhabi Islamic Bank today, and it can be live in your bank next.

Try it first on your own sample data in the hosted sandbox.

demo.crayondata.ai/controls

Controls AI · Crayon Data

Thank you.

Scan for the live portal

QR code to the live portal