17 Nov 2026 By CISO Consulting team 19 views 6 min read

SAMA Counter-Fraud Framework Risk Scoring: Closing the Gap

Most institutions score customers at onboarding and refresh them annually, while the SAMA Counter-Fraud Framework expects continuous assessment. This article contrasts static segments with dynamic behavioural scoring and shows where the gap surfaces.

A fraud analyst reviewing behavioural risk dashboards on a large monitor in a modern operations room.
In this article
  1. What the framework expects
  2. How static segmentation works, and where it holds up
  3. What dynamic behavioural scoring adds
  4. Transaction velocity
  5. Peer-group deviation
  6. Channel switching
  7. Why the gap appears at the first incident
  8. Typical gaps we see in scoring design
  9. A practical path to close the gap
  10. How a Fraud Assessment helps
  11. Closing

Heads of Fraud Risk and Analytics who are implementing the SAMA Counter-Fraud Framework often find that the hardest question is not about governance or policy. It is about scoring. SAMA counter-fraud framework risk scoring expects fraud risk to be assessed continuously, yet many institutions still assign a customer a risk rating at onboarding and revisit it at the annual refresh. This article sets out the difference between those two approaches, what dynamic scoring needs in practice, and why the gap tends to become visible during the first fraud incident after implementation.

What the framework expects

The SAMA Counter-Fraud Framework (CFF) requires regulated institutions to assess fraud risk on a continuing basis rather than as a point-in-time exercise. The wording matters. A rating assigned once and reviewed yearly is a record of what was known at a moment in the past. It is not an assessment of what the customer is doing now.

We do not suggest that every institution needs the same model. The framework is principles-led, and proportionality applies. But the direction is clear: the institution should be able to show that its view of fraud risk changes when the evidence changes.

How static segmentation works, and where it holds up

Static segmentation places customers in broad groups such as retail, SME and corporate. Each group carries a default risk level, sometimes adjusted at onboarding for factors like product, occupation or channel. Monitoring rules and thresholds are then tuned per segment.

This approach has real strengths:

  • It is simple to explain to the board, auditors and the regulator.
  • It is easy to govern, because the number of segments is small.
  • It gives a stable baseline for resourcing and reporting.

The weakness is that a segment describes a population, not an individual. Two retail customers in the same segment can behave very differently. A salaried customer who suddenly begins moving funds in rapid succession through a new channel carries the same rating as before, because nothing in the model has a trigger to change it.

What dynamic behavioural scoring adds

Dynamic scoring keeps the segment as a starting point and then updates risk using observed behaviour. Three signals are the most practical to begin with.

Transaction velocity

Velocity looks at how many transactions occur, and for what value, over defined windows. A sudden rise in count or value relative to the customer's own history is a stronger indicator than a fixed amount threshold applied to the whole segment. Velocity also catches patterns that stay under static limits by design, such as many small transfers.

Peer-group deviation

Peer-group analysis compares a customer with others who look similar, for example by product, tenure, typical balance and activity. The question changes from "is this transaction large?" to "is this customer behaving unlike comparable customers?" This is where the limits of retail, SME and corporate labels become clear, since a peer group is usually much narrower than a segment.

Channel switching

Customers tend to use channels in stable patterns. A move from branch to mobile, a first use of a new device, or a change in how payments are initiated can be legitimate, but combined with other signals it is often an early marker of account takeover or social engineering. Channel behaviour is rarely captured in an onboarding score at all.

Each signal alone produces noise. The value comes from combining them, weighting them, and letting the combined score drive a defined response: step-up authentication, a hold, review by an analyst, or no action.

Why the gap appears at the first incident

Institutions generally pass the documentary stage of a framework implementation. Policies exist, roles are assigned, a risk methodology is approved. The scoring gap is different because it is exposed by events rather than documents.

After the first significant fraud case, the usual questions follow:

  1. What was the customer's risk rating at the time of the loss?
  2. Had anything in the customer's behaviour changed beforehand?
  3. Did the institution's systems register that change?
  4. If they did, what action was triggered, and by whom?

When the answer to the first question is the onboarding score from some years ago, and the answers to the second and third are "yes" and "no", the case review turns into a finding. The issue is not that fraud occurred. It is that the institution cannot demonstrate that its assessment was continuous.

Typical gaps we see in scoring design

These are common patterns in this area, described generally rather than as findings about any one institution:

  • Rating without refresh triggers. The model has no events that cause re-scoring, such as a change of device, contact details or beneficiary pattern.
  • Inherent and residual risk conflated. The score mixes the exposure of a product with the strength of its controls, so improvements or weaknesses in controls never show up.
  • Segment-level thresholds only. Rules are tuned to the segment and ignore the individual's baseline.
  • Disconnected data. Channel, device and transaction data sit in different systems, so combined signals are not available at scoring time.
  • No ownership for model performance. No one is accountable for tuning, back-testing and documenting changes to the scoring logic.
  • Internal fraud left out. Scoring focuses on customers and does not consider insider or collusion scenarios.

A practical path to close the gap

A full behavioural platform is not the first step. A staged approach is more realistic.

  1. Establish the risk picture. Define inherent and residual fraud risk by product and channel, so the scoring model has a documented basis.
  2. Map what is detected today. Identify which of velocity, peer deviation and channel switching are already monitored, and where data exists but is not used.
  3. Define re-scoring triggers. Agree the events that move a customer's risk up or down, and the response attached to each.
  4. Test coverage. Check that the controls and detection rules behave as designed, using extracts or read-only access where available.
  5. Assign model governance. Name an owner, set review cycles, and keep evidence of tuning decisions for supervisory review.

How a Fraud Assessment helps

Our Fraud Assessment is an enterprise-wide assessment of internal and external fraud risk across products, channels and processes, scored against the SAMA Counter-Fraud Framework. For the scoring question in this article, it gives Heads of Fraud Risk a structured view of where static ratings are carrying risk that should be dynamic.

The work follows four steps: scope, discover, assess and report. We agree products, channels and processes in scope, then use interviews, process walkthroughs and data review to discover how risk is currently identified and scored. We score inherent and residual fraud risk and test key controls. The assessment then closes with a register, gap analysis and prioritized remediation plan.

The deliverables are a fraud risk assessment report, a fraud risk register, a control gap analysis and a prioritized remediation plan. The register is organised by product and channel, with controls mapped to the SAMA Counter-Fraud Framework, so the output can support your submissions. The assessment runs for 4–6 weeks.

Two practical points often come up. Read-only access or extracts from the fraud system help us test detection coverage, but the assessment can start with interviews and documents. And insider and collusion scenarios are assessed alongside external fraud.

Closing

The first fraud incident after implementation should not be the moment you learn your scoring is static. If your customer risk ratings are set at onboarding and refreshed annually, a Fraud Assessment is a sound way to see the gap, size it by product and channel, and agree a sequenced plan to close it. Contact CISO Consulting to discuss scope.

Tags: SAMA CFF fraud risk scoring behavioural analytics financial crime fraud assessment banking
Readiness self-check

How ready are you? Find out in two minutes

Answer a few questions for your framework. You get a score, your biggest gaps and — if you want it — a detailed assessment from our team.

Ready to talk about your compliance?

Tell us where you stand. We will show you the shortest path to what your regulator expects.

Regulatory updates in your inbox

SAMA, NCA and SDAIA changes and what they mean for your institution — once a month.

E-mails in

We confirm by e-mail; unsubscribe any time. Manage your subscription · Past issues

Schedule a Free Assessment