JuicyScore

08/09/2026

How cluster-graph analysis strengthens modern anti-fraud systems?

In our new article, we look at the role of cluster-graph analysis within a multi-layered anti-fraud architecture. We examine how this approach has evolved, how it is implemented within JuicyScore, and where cluster-level signals can be useful — from investigations and attack mode to scoring amplification, fraud-farm blocking, and early warning.

Read the full article: https://juicyscore.ai/en/blog/cluster-graph-antifraud-device-authentication

How JuicyScore combines anti-fraud scoring, device authentication, and cluster-graph analysis to detect connected fraud patterns and strengthen operational risk controls.

03/06/2026

In a digital lender's application flow, JuicyScore found a segment – about 1% of applications – carrying 70%+ NPL90 risk. The signals were hidden in the device, not the credit file.

The lender already had a working decisioning process. What it needed was a layer that could see what happens before a borrower enters the portfolio: which device is used, how the application is filled in, and whether the connection environment looks consistent or manipulated.

We approached it in two layers:
→ Full JuicyScore API – custom model, five risk segments across the web flow
→ JuicyID – lighter vector, isolated 0.5–0.7% of applications at 70–75% risk; faster to validate and deploy

What separated risk most clearly:
🔸 Device and technical anomalies
🔸 Behavioral signals that don't match real customer patterns
🔸 Connection markers pointing to a manipulated environment

The part risk teams tend to overlook: the same signal layer flagged around 7% of rejected applications as a potential area for safer approval growth. Device intelligence isn't only about declining bad applications – it surfaces good borrowers that conservative rules can block.

For digital lenders, that's the balance that matters: cutting exposure to toxic applications without blocking the growth sitting in the rest of the flow.

⬇️ Full case study: https://bit.ly/4dYzC4J

28/05/2026

In digital lending, the riskiest applications do not always look risky in traditional credit data. Especially when the full journey happens online.

The early signals often sit outside the credit bureau layer:
in the device, connection, technical environment and user behavior.
In a recent anonymous case study with a South Asian digital lender, JuicyScore analyzed 150,000+ web-channel applications and built a custom ranking model using these digital signals.
The preliminary analysis showed 38.5% Added Gini and helped identify:
🔸 high-risk FPD segments
🔸 practical stop-markers for earlier filtering
🔸 lower-risk borrower groups that may support safer approval growth after validation

Risk signals should not only help lenders block bad applications.
They should also help approve good customers with more confidence.

➡️ Read the case study to see what changed when risk assessment moved beyond traditional credit data: https://juicyscore.ai/en/case-studies/custom-risk-model-south-asian-digital-lender-38-5-added-gini-juicyscore-signals

12/05/2026

India built one of the fastest payment systems in the world.
The RBI's latest proposal to fix UPI fraud: slow it down.

In his latest expert piece, Manish Thakwani (Head of Business Development, India & South Asia at JuicyScore) shares his perspective on where risk builds in the user journey, and what it takes to detect it earlier.

He notes that RBI's proposal is a response to a real and growing problem. But it also reveals a deeper assumption: that fraud can be stopped at the moment of transaction.

In practice, that's already too late.

By the time a payment is initiated, the risk has already formed.

The transaction is just where fraud becomes visible, not where it begins.

The earliest signals sit across telecom and behavioural layers. SIM swaps, device re-binding, number spoofing, abnormal communication patterns. These often precede fraudulent transactions by minutes or even hours.

Yet they're rarely connected to real-time payment decisioning.

In real systems, fraud prevention is adaptive:
🔸 risk is evaluated continuously
🔸 friction is applied selectively
🔸 monitoring extends across sessions, not just transactions

This is how you preserve speed for legitimate users while intervening where risk is actually present.

Read the full blog: https://bit.ly/4d6E2Hq

UPI fraud doesn’t start at the transaction. Learn why delays and controls are often too late, and how early signals across telecom, device, and behavior can improve fraud prevention.

29/04/2026

Most anti-fraud systems today only analyze the finished transaction. But a growing number of attacks now happen before the transaction is even sent — directly inside the user’s browser.

Attackers are increasingly using DOM injections and client-side hijacking (Man-in-the-Browser attacks) to silently modify forms, intercept submit events, and alter critical data — all while the session looks perfectly legitimate to your backend, WAF, behavioral models, and even biometrics.

The result? Real users, clean device fingerprints, normal mouse movements — and stolen money.
In our new article, we explain:

• How classic XSS has evolved into stealthy DOM-based attacks
• Why behavioral anti-fraud, biometrics, and antivirus solutions often miss these threats
• Which browser APIs attackers exploit most effectively (submit, Storage, Canvas, sendBeacon and more)
• What actually works today: real-time DOM integrity monitoring and client-side Zero-Trust

If you rely on traditional anti-fraud tools, this blind spot is already being actively exploited — and the trend is accelerating.

Cut through the illusion of security.

Discover where your current defenses stop seeing the attack — and what you need to close this critical gap:

How modern hijacking has shifted from the server side to the user’s browser: the attack unfolds invisibly, in real time, leaving no traces in logs. And most importantly, why traditional anti-fraud fails to detect the problem—even when the user is legitimate.

10/04/2026

The IRDAI Fraud Monitoring Framework is now in force (April 2026), setting new requirements for how insurers in India approach fraud detection and risk monitoring.

While the framework clearly defines governance, reporting, and responsibilities, it leaves a key question open: how to implement continuous fraud monitoring in practice. This is where many insurers face an ex*****on gap — especially when it comes to detecting cyber and “new age” fraud, moving beyond static Red Flag Indicators, and identifying patterns across digital interactions.

Solutions like JuicyScore, a device intelligence risk scoring solution, can support this transition by enabling real-time fraud detection, strengthening RFIs with dynamic signals, and improving visibility into suspicious behavior across channels.

When evaluating fraud detection solutions under IRDAI requirements, insurers should focus on continuous monitoring, real-time risk scoring, and the ability to detect environment-level anomalies — not just user data.

👉 Full breakdown, examples, and implementation guidance: https://bit.ly/4cgKvhn

02/04/2026

Device intelligence is becoming a system-level risk layer that connects events, behavior, and entities across the user journey.

At this level, it does not simply flag anomalies — it provides structure.

Treating risk as a set of individual or aggregated signals is limited. Such attributes can be ambiguous, reproducible, and do not always reflect intent.

When used as a system-level layer, device intelligence enables signals to be interpreted in context rather than in isolation — across sessions, accounts, and environments.

This is where device intelligence solutions differ in practice.

Solutions like JuicyScore focus not on isolated device attributes, but on how signals interact over time — combining device, behavioral, and network data into a structured view of risk.

In cases such as MoneyMan Mexico and ATM Online Vietnam, following JuicyScore integration, observed changes included:�– improved model separation after combining device and behavioral signals�– higher approval rates with stable risk metrics�– increased Gini and ROI through integrated signal analysis

These outcomes reflect a broader shift in fraud detection and risk scoring.

Modern device intelligence in fraud detection is not about detecting a single suspicious signal.�It is about understanding how signals behave, repeat, and relate to each other within a system.

This is what allows risk models to move from reacting to isolated events�→ to identifying patterns, connections, and coordinated activity.

For teams evaluating device intelligence solutions, the key question is no longer:�“Does this signal look suspicious?”
It is:�“How does this solution structure signals into a consistent, explainable risk context?”

JuicyScore is a device intelligence risk scoring solution designed around this principle — combining over 200+ non-PII signals into a system-level view of risk.

Value emerges from how signals interact, persist, and evolve over time — forming a system-level understanding of behavior.

Read more on the JuicyScore blog:�

How device intelligence evolves from signals to structured risk context – and why modern fraud detection depends on connections, not isolated attributes.

26/03/2026

Generative AI and large language models are increasingly discussed in the context of fraud prevention and risk management.

At the same time, it is important to distinguish where these technologies already deliver practical value and where their application remains limited by the requirements of real-world systems.

In our new article, we look at the current role of LLMs in fraud and risk management, including their strengths, limitations, and realistic use cases: https://juicyscore.ai/en/blog/llm-risk-management-antifraud-use-cases-limitations-en

Despite the hype surrounding LLMs, their application in risk management requires a measured and critical assessment. What are the real use cases, the key limitations, and the reasons why generative models cannot directly replace traditional fraud prevention approaches?

06/03/2026

Brazil now operates one of the most advanced Open Finance ecosystems in the world.

More than 128 million active consents connect financial institutions through standardized APIs, enabling credit, payments, insurance, and embedded services to operate in real time.

But as interoperability expands, the nature of risk changes.

Open Finance does more than enable data sharing. It fundamentally reshapes where risk lives in the financial system.
Risk no longer sits inside the perimeter of a single institution.

It moves with the transaction – across APIs, devices, and institutional boundaries.

In a new expert article, José Edson Da Costa, National Director of Business Development – Brazil at JuicyScore, examines how risk architecture must evolve to support Open Finance at scale.

The article examines:
🔹 how interoperability expands the ex*****on layer where fraud can occur
🔹 why traditional perimeter-based risk models are no longer sufficient
🔹what resilient Open Finance infrastructure requires

Brazil’s first phase of digital finance was about access.
The next phase will be defined by resilience.

Read the full article: https://juicyscore.ai/en/blog/open-finance-brazil-risk-architecture