Skip to main content
Risk thresholds let you calibrate how aggressively AntifraudAI screens events. Every screened event receives a score from 0 (very low risk) to 100 (very high risk). Thresholds define the boundaries that map those scores to one of three decisions: allow, review, or block. By adjusting these boundaries you can dial AntifraudAI’s sensitivity up or down to match your business’s unique risk tolerance.

​
Default Thresholds

Out of the box, AntifraudAI ships with the following balanced defaults: These defaults work well for most businesses at launch. Once you have a few days of production data in the analytics dashboard, you can begin fine-tuning based on your observed false positive and fraud-slip-through rates.

​
Changing Thresholds

1

Open Risk Threshold settings

Log in to the AntifraudAI dashboard and navigate to Settings → Risk Thresholds.
2

Adjust the boundaries

Use the interactive slider or type exact integer values into the Allow/Review boundary and Review/Block boundary fields. The three ranges update in real time as you adjust the values so you can preview the mapping before saving.
3

Save your changes

Click Save Thresholds. New thresholds take effect immediately for all new incoming events — no deployment or restart required.

​
Threshold Strategy

Choosing the right thresholds is a balance between two competing risks: Lower block threshold (more aggressive)
  • Blocks more fraudulent transactions → fewer fraud losses.
  • Also blocks more legitimate transactions → higher false positive rate → frustrated customers and lost revenue.
  • Best for high-risk verticals (crypto, digital goods) where fraud losses outweigh customer friction.
Higher block threshold (more permissive)
  • Fewer legitimate transactions are blocked → better customer experience.
  • More fraudulent transactions slip through → higher fraud losses.
  • Best for low-risk verticals or businesses with strong post-payment fraud recovery processes.
The review band is your safety valve. Transactions in the review range are flagged for manual inspection rather than automatically blocked, letting your fraud operations team make the final call on ambiguous cases.

​
Per-Event-Type Thresholds

AntifraudAI allows you to set independent thresholds for each event type, so you can apply different levels of scrutiny depending on the nature of the event:
  • Transaction events — typically warrant stricter thresholds because a fraud loss is direct and immediate.
  • User events (account creation, login, password reset) — may use a higher block threshold since the downstream risk is lower than a payment event.
  • Custom events — inherit the global thresholds by default; override them per event type in Settings → Risk Thresholds → Custom Events.
To configure per-event-type thresholds, select the event type from the dropdown at the top of the Risk Thresholds settings page and adjust the values independently.
Changing thresholds in the production environment takes effect immediately for all new events. A misconfigured threshold could block legitimate users at scale or let fraudulent transactions through. Always test your desired configuration in the sandbox environment first and validate the results before applying changes to production.
Start with the default thresholds and let at least 72 hours of production data accumulate. Then open Analytics → Decision Breakdown to review your false positive rate and fraud detection rate, and adjust the boundaries incrementally — in steps of 5 points — until you reach an acceptable balance.