Financial crime investigator and team lead reviewing a transaction network on screen
  1. Home
  2. Use cases
  3. Banking & financial services
Banking · Fraud detection & complianceDelivered · client result

A large European bank

Banking fraud and compliance analytics: from 3-day queries to real time

Fraud detection and analytics moved from a 3-day query process to real-time intelligence across millions of daily transactions.

3 days → <1 s

complex relationship query time

85%

improved fraud detection, as reported by the bank

70%

productivity increase

Executive summary: fraud and compliance analytics

Client
A large European bank
Sector
Banking & financial services
Scale
Millions of transactions a day
Scope
Fraud detection, compliance reporting, document analysis

Context, decision, outcome

  • Context. Complex database queries took 3 business days, so real-time fraud detection was impossible. Siloed departmental data blocked a 360° customer view, and staff spent hours on repetitive data retrieval.
  • Decision. Re-architect around relationships: a high-performance graph data layer connecting departments, machine-learning anomaly detection, and a real-time pipeline for millions of daily transactions. Investigators keep the full context behind every alert.
  • Outcome. Sub-second answers to complex relationship queries that used to take 3 days. 85% improved fraud detection (as reported by the bank), 80% faster compliance reporting, 70% higher productivity and 42% higher customer satisfaction.
  • Time to value. Training sessions and technical documentation were delivered to banking staff, built around the questions investigators already ask.
The challenge

The challenge for fraud and compliance teams

In banking, the useful answer depends on relationships: between accounts, transactions, documents, customers and events. At this bank, those relationships were spread across departmental systems, and complex queries took 3 business days.

The delays hit investigation, reporting and every decision that needed a connected picture. By the time the data reconciled, the decision window had often closed, and slow decisions increased fraud exposure.

How the work was done

The journey: from 3-day queries to real-time investigation

The bottleneck turned out to be in the relationships, not in the people or the hardware.

  1. Step 01 · Analyse

    How we diagnosed the bottlenecks

    We mapped the path from transaction processing through fraud detection to compliance reporting, and traced the data flows between departments.

    We looked for silos, repeated retrieval work and the relationships teams needed to see together. That let us focus the first change on the questions where connected data would release the most capacity.

    Delivery Discovery delivered by our team.

  2. Step 02 · Design & transform

    How we built a data model around investigation

    We designed a high-performance graph database architecture optimised for complex relationship queries, and brought relationship data into a structure that suits investigation.

    We trained machine-learning models for pattern recognition and anomaly detection across transaction data, and built a real-time processing pipeline for millions of daily transactions.

    A unified data layer connected the departments around a shared view, with operating context and access controls built into the design.

    NatusAI products for this capability

    Abstract render of a time-versioned graph in which relationships keep their history
    Every transaction and relationship kept with its history, so a pattern can be followed across time.

Rollout and adoption

Abstract render of an operating picture in which alerts, transactions and their history are connected
Connected data turns fraud risk into something a board can discuss with evidence.
  1. Step 03 · Build

    What we built

    A real-time fraud detection engine with live transaction monitoring, instant alerts and automated risk scoring. Investigators see the activity, the risk pattern and the context around it.

    A document analysis platform with a natural-language query interface, capable of processing millions of documents in seconds, so analysts can ask questions across approved document collections.

    360-degree customer data integration with cross-departmental access and collaboration tools.

    NatusAI products for this capability

  2. Step 04 · Adopt

    How the teams adopted it

    Training sessions and technical documentation for banking staff, built around the questions investigators and analysts already ask, with the context needed to review a result rather than accept a black-box answer.

    NatusAI products for this capability

Delivered by the NatusAI team. Product chips show which NatusAI product provides, or is planned to provide, each capability; release status is listed under “For your technical team”.

Financial-crime investigators reviewing transaction tables and a relationship diagram together (illustrative image)
Illustrative image.
The human side

What changed for risk, fraud and compliance teams

Less waiting between a question and a decision, and more time for judgement.

  • Fraud investigators. Alerts arrive in real time with the graph and history behind them, so a pattern can be followed across hops and explained.

  • Compliance teams. Reporting is 80% faster, because the data is already connected and questions across documents can be asked in plain language.

  • Customer-facing staff. A 360-degree customer view across departments. Customer satisfaction rose 42%.

Measurable results that matter

The impact: real-time fraud and compliance analytics

3 days → <1 s

complex relationship queries, from 3 business days to under a second

85%

improved fraud detection, as reported by the bank

70%

productivity increase

80%

faster compliance reporting

42%

higher customer satisfaction

98%

document-analysis accuracy as reported by the bank, across millions of documents a month

Figures reported by the client at project close.

Financial data & investigation

Where the same foundation goes next

In most banks, the ledger, the fraud engine, the risk model and the audit trail live in different systems, stitched together by overnight batches. A connected, time-aware foundation lets you decide, detect and forecast on live data.

Reference scenario

Instant lending decisions

The problem
Granting a loan or limit means pulling an applicant’s history and web of obligations, usually a slow join across systems.
The approach
The full transaction history and counterparty network sit in one place. A scoring model evaluates them and returns approve or decline in moments.
The result
Real-time credit decisions at the point of sale or in the app, backed by the exact data used to make them.
Reference scenario

Fraud & AML detection

The problem
Fraud and money laundering show up as patterns over time, such as rings, mule chains and sudden bursts, which row-by-row checks miss.
The approach
Follow the money across hops, spot circular flows and velocity spikes, and score anomalies as transactions happen.
The result
Suspicious patterns flagged in real time, with the full network and history at hand to investigate and explain them.
Reference scenario

Balance & liquidity

The problem
Treasury needs tomorrow’s balances too, yet forecasts live in spreadsheets disconnected from the live ledger.
The approach
Forecast each account’s balance and roll it up to branch and bank-wide liquidity, then stress-test scenarios on the same data.
The result
Forward-looking liquidity you can plan against, refreshed continuously instead of once a month.

“What did this exposure look like on 30 June?” Every transaction and relationship is kept with its history, so audit, compliance and back-testing run on the same data as live decisions.

Reference scenarios for banking, insurance and payments; illustrative, not NatusAI client results. Lending and credit scoring may be high-risk under EU AI Act Annex III. See governance below.

Transferability

What this means for banks, insurers and payment institutions

Banks, insurers and payment institutions that still wait on batch reconciliation can apply the same approach, starting with one investigation path: a fraud ring, a claims pattern or a compliance report.

Start with the question that waits longest.

One investigation path, one set of relationships made visible, and a measured before and after.

Explainability comes from context.

When investigators can see the network and history behind an alert, they can defend the decision to supervisors.

Keep data where regulation wants it.

Graph, history and analytics can run as one system on the bank’s own infrastructure.

Regulation

Built for DORA, AML, Solvency II and EU AI Act governance

Financial institutions must be able to explain, reproduce and exit. The architecture keeps a named person responsible for consequential decisions, keeps the history needed to reconstruct them, and can run inside the bank’s own perimeter.

  • DORA (applicable since 17 Jan 2025): on-premises or EU deployment, a documented exit strategy, information for your register of ICT third-party arrangements, recovery evidence (verified recovery, NatusAI Gemini, on the roadmap)
  • AML / financial-crime supervision: alerts linked to the transactions, counterparties and history behind them, for explainable investigations
  • EU AI Act: creditworthiness assessment of natural persons is an Annex III high-risk use, with stand-alone high-risk obligations expected from 2 Dec 2027 (subject to final adoption). Human oversight, logging and data-governance documentation support deployer duties
  • Solvency II / EIOPA (insurers): the same controls (human oversight, logging, documented data flows and a documented exit plan) give insurers the documentation their system-of-governance and outsourcing reviews ask for. DORA applies to insurers as it does to banks
  • GDPR: processing inside the bank’s environment, documented data flows
  • Model risk: documentation to support your internal validation (e.g. SR 11-7 / OSFI E-22 for US/Canadian entities)

NatusAI supports your obligations; it does not replace your legal assessment or supervisory approval. How NatusAI supports EU AI Act, GDPR, DORA and NIS2 governance

A connected-knowledge graph: answers traceable to their facts and relationships
For your technical team
  • Delivered architecture: a high-performance graph database for relationship queries, ML anomaly detection on transaction data, a real-time processing pipeline (millions of transactions a day), a unified data layer across departments, a multi-agent document-analysis system with a natural-language query interface.
  • Reference engine pattern: transactions indexed on arrival as versioned edges between account, customer and counterparty nodes. Graph, time series and ML in one engine. As-of queries for any timestamp. One self-hosted binary.
  • Platform mapping: NatusAI Cygnus (Planned v1), NatusAI Auriga (Roadmap), NatusAI Aquila (Roadmap), NatusAI Orion (Roadmap), NatusAI Lyra (Available).
Questions

Questions from CROs and heads of financial crime

Speed, explainability, deployment and wider effects.

How fast can complex relationship queries become?

At this bank, complex queries that took 3 business days were answered in under a second, while the system handled millions of transactions a day.

Can investigators explain an AI-generated alert?

Yes. Each alert comes with the transactions, counterparties and history behind it, so investigators review the context rather than accept a black-box score.

Can it run on-premises?

Yes. The architecture can run as one self-hosted system on the bank’s own infrastructure.

Does the same approach apply to insurers?

Yes. The case was delivered at a bank, and the pattern applies wherever fraud, claims or compliance answers depend on relationships and history. The same controls support the governance documentation insurers prepare under Solvency II and DORA; NatusAI does not certify compliance.

What else improved besides fraud detection?

Compliance reporting became 80% faster, productivity rose 70%, and customer satisfaction rose 42%, thanks to a 360-degree customer view across departments.

Client results
Discuss a similar project

Which connected question takes too long today?

Start with one investigation path and make its relationships visible. In 30 minutes we’ll outline how we’d measure the change, and how it fits your DORA and model-risk framework.

  • Start with a scoped discovery phase that ends in a clear go / no-go decision.
  • Reference conversations about this case can be arranged under NDA, after a first briefing.

Or write to contact@natusai.com

A small leadership team reviewing a one-page result summary in a meeting room (illustrative image)