From 200+ to 23 Daily Alerts

About NordicBet

NordicBet is an online casino and sportsbook operating under Swedish (Spelinspektionen), UK (UKGC), and MGA licences. With 120,000+ monthly active players, they focus primarily on the Scandinavian market where responsible gambling requirements are among the strictest in the world.

The Problem

Spelinspektionen (the Swedish regulator) had recently increased enforcement of responsible gambling requirements. NordicBet needed to demonstrate proactive detection and intervention for problem gambling.

Their existing rule-based RG system was generating chaos:

• 200+ alerts daily from simple threshold triggers
• 85% false positive rate — most alerts were recreational players who happened to deposit more than usual
• 2 full-time staff reviewing alerts (mostly closing false positives)
• Real cases slipping through because the team was overwhelmed by noise

The rule-based approach was fundamentally flawed: "Deposit > €500 in a week" catches both a problem gambler chasing losses and a recreational player who got a bonus at work.

The Solution

NordicBet deployed SafePlay to replace their rule-based RG detection. The integration took 10 days:

1. Days 1-3: Connected SafePlay to their player data (transactions, sessions, support tickets, chat logs)
2. Days 4-7: SafePlay analyzed historical data to establish baselines and identify patterns
3. Days 8-10: Parallel run — both systems active, comparing results
4. Day 11: Full deployment with SafePlay as primary

SafePlay used multiple signals:

• Text analysis: LLM analyzed support chats for distress signals ("I need to win back", "last time, I promise", "my wife can't know")
• Behavioral patterns: Chasing losses, session time changes, deposit velocity spikes
• Trend detection: Not just current behavior, but trajectory (risk score increasing over time)

Results

After 6 months of operation:

Why It Worked

SafePlay succeeded because:

• Context over thresholds: A €500 deposit means nothing without context. SafePlay considers the player's history, behavior patterns, and communication.
• Text signals: Players often express distress in support chats before their behavior triggers thresholds. LLM catches these signals.
• Trend tracking: A risk score of 65 means different things if it was 30 last week vs. 70 last week. Direction matters.
• Audit-ready documentation: Every alert comes with a full explanation, making regulator reviews straightforward.