In 2024, a single wave of AI-generated phishing scams could have drained millions from 46 Norwegian banks in one season. Instead, their shared alliance cut phishing losses by 90% year-over-year — without adding a single fraud analyst.
Banks are now fighting AI-generated fraud with AI-generated defense-in-depth. Generative AI has made phishing, synthetic identities, and account-takeover scripts effectively indistinguishable from genuine customer behavior — the grammar mistakes and dead giveaways are gone. Static, rule-based fraud engines built for last decade's scams block real customers as often as they catch real fraud. There's a better way: technology that doesn't just flag a suspicious transaction against a shared blacklist — it learns what "normal" looks like for every single account and reacts in milliseconds when something breaks that pattern.
Eika Gruppen, an alliance of 46 independent local banks across Norway serving over 850,000 customers, rolled out Featurespace's ARIC Risk Hub across its shared infrastructure. Result: a 90% drop in phishing-related losses in 2024 compared to 2023, with no material increase in false declines on genuine transactions. Here's the technology behind that number — and how it now sits inside Visa's own global fraud-prevention stack after Visa's acquisition of Featurespace closed in December 2024.
Why Behavioral AI Is Beating Rule-Based Fraud Detection
Traditional fraud systems ask one question: does this transaction match a known fraud pattern? ARIC Risk Hub asks a different one: does this transaction match this specific customer's normal behavior? Its Adaptive Behavioral Analytics engine builds an individual, continuously updating profile for every account — typical spending amounts, devices, locations, times of day, merchant categories — and scores each new transaction against that personal baseline in real time, in milliseconds, across every channel a bank offers (card, mobile app, online banking, wire transfer).
The system doesn't need weeks of retraining to catch a new scam variant, because it isn't looking for a specific known pattern — it's looking for deviation from the individual. That's what let it catch AI-personalized phishing scripts even before human fraud teams had labeled them as a new attack type.
This shift matters most for mid-size and regional banks, credit unions, and payment processors that can't staff a 24/7 in-house fraud data science team the way an HSBC can — the kind of institution an Operations or Risk Director at a 40-branch regional bank is trying to protect on a fixed compliance budget.
The 90% Number Behind Eika Gruppen's Phishing Turnaround
According to Visa and Featurespace's own reporting, Eika Gruppen's phishing-related losses fell 90% in 2024 versus 2023 after deploying ARIC Risk Hub across the alliance's shared banking infrastructure — a scale few individual community banks could achieve alone. [DATO A VERIFICAR — source: joint Visa/Featurespace newsroom release; exact absolute loss figures not published].
Danske Bank, one of the Nordic region's largest banks, separately reported measurable gains in fraud-catch rate after its own ARIC Risk Hub rollout, per Featurespace's newsroom case study — though Danske has not published a specific percentage publicly [REQUIERE VERIFICACIÓN].
Zoom out and the pattern holds across the platform's broader customer base — which today includes HSBC, NatWest Group, TSYS, Worldpay, ClearBank, and Permanent TSB: Featurespace reports up to a 70% reduction in genuine transactions wrongly declined, and in individual card-fraud deployments, reductions of genuine card-not-present declines as high as 98%. Every wrongly blocked transaction is a customer who might not come back — so cutting false positives isn't just a technical win, it's a retention and revenue one.
What Changed When Visa Bought Featurespace
Visa completed its acquisition of Featurespace in December 2024, and by April 2025 had folded ARIC Risk Hub directly into its Value-Added Services portfolio — making it available globally to any bank, acquirer, or financial institution on Visa's network, not just Featurespace's original direct clients. That distribution shift is the real story for smaller institutions: fraud AI that used to require a standalone enterprise contract with a specialist vendor is now a service layer banks can plug into through infrastructure they already use.
Featurespace also launched TallierLTM in 2023 — described as the industry's first Large Transaction Model, a generative-AI foundation model pre-trained on billions of transactions. Financial institutions call it through an embeddings API that converts a customer's transaction history into a "behavioral barcode" without exposing personally identifiable data. In benchmark testing at an industry-typical 5:1 false-positive ratio, TallierLTM improved fraud value detection by up to 71% over standard models — the same kind of leap foundation models brought to language and image tasks, now applied to spotting a stolen card three transactions before a rule-based system would have noticed.
Banks that adopted adaptive behavioral fraud detection early — Eika Gruppen, Danske Bank, HSBC, NatWest — are now several product cycles ahead of institutions still running purely rule-based engines. As GenAI-driven fraud attacks scale (Spain's Tribunal Supremo has already had to rule on bank liability for AI-assisted SMS-authentication fraud, per its April 2025 ruling), that gap between adaptive and static defenses will only widen.
The Technology Stack Behind ARIC Risk Hub
None of this requires a bank to build fraud AI from scratch.
| Tool | Role in This Deployment | Access Model | From |
|---|---|---|---|
| Featurespace ARIC Risk Hub | Real-time adaptive behavioral scoring across card, mobile, online banking and wire channels | Enterprise license via Visa VAS or direct | Custom enterprise pricing |
| TallierLTM | Generative-AI transaction embeddings for early-stage fraud pattern detection | Embeddings API, partner-distributed (e.g. via TSYS) | Custom enterprise pricing |
| Visa Value-Added Services | Global distribution layer bringing ARIC to Visa network banks/acquirers | Available through Visa network membership | — |
⚠️ [REQUIERE VERIFICACIÓN: pricing for non-Visa-network institutions and minimum transaction-volume thresholds are not publicly disclosed — banks should request a scoped quote directly.]
Who This Actually Protects — and Who Still Needs More
This kind of platform delivers the clearest ROI for banks, credit unions, and payment processors already inside the Visa network, or large enough to justify a direct Featurespace enterprise contract — regional and alliance banks like Eika Gruppen's 46 members are the sweet spot, since shared infrastructure spreads the cost across many smaller institutions. It is not a plug-and-play tool for an individual small business or a standalone fintech without an existing card-network relationship; those institutions are better served by lighter-weight fraud APIs built for lower transaction volumes.
Frequently Asked Questions
Is ARIC Risk Hub available to banks outside Visa's network?
Yes — Featurespace continues to serve direct enterprise clients such as HSBC, NatWest, and Danske Bank independently of Visa's distribution layer, though Visa's Value-Added Services portfolio is now the fastest route for Visa-network institutions specifically.
Does adaptive behavioral analytics replace a bank's existing fraud rules?
No — most deployments run ARIC's behavioral scoring alongside existing rule-based and consortium-data systems, using it to catch what static rules miss and to reduce false declines on genuine customers, rather than as a full replacement.
How long does a deployment like Eika Gruppen's typically take?
Featurespace does not publish a standard timeline publicly, and enterprise fraud-platform rollouts across a multi-bank alliance typically span several months of integration and tuning rather than weeks [REQUIERE VERIFICACIÓN — exact Eika Gruppen deployment timeline not disclosed].
Generative AI has made fraud harder to catch with static rules — and it's making adaptive, behavior-based detection the new baseline rather than the exception. The institutions moving first aren't just cutting losses; they're rebuilding customer trust in every transaction that goes through smoothly.