What is a SaaS Fraud Score Threshold?

The SaaS fraud score threshold is a numeric boundary at which you take actions depending on the level of risk indicated by your transaction. Usually, after a machine learning model has analyzed a transaction, a score is assigned based on the degree of resemblance to fraud. The threshold indicates what action is to be taken next, whether to approve the order, ask for a manual review, or decline it. Correct placement of the cutoff is crucial as it affects both security and the company’s revenue.

The definition of a favorable score can vary, influenced by the specific criteria adopted by each vendor. A low risk score is typically associated with transactions that do not present significant anomalies; a high score may correspond to situations requiring further review, such as discrepancies in customer or account details, an atypical transaction volume within a short timeframe, or an originating IP address that raises concerns. 

SaaS merchants set their own bands; for example, 0–30 means “approval,” 31–70 indicates “review,” and 71+ means “decline,” depending mainly on their risk tolerance, average order value, and the industry.

Commonly, SaaS companies use these approaches:

• Single fixed threshold — this technique utilizes a single fixed threshold to differentiate between approved and declined statuses, without an intermediate category.
• Tiered thresholds — scores are arranged under multiple categories or bands (e.g., “approve”, “review”, “decline”), thereby allowing the manual review aspect.
• Dynamic threshold — the cutoff level is impacted by transaction factors such as quantity, item type, or customer history.
• Segmented threshold — different thresholds are set for different segments such as customer groups, payment means, or location (e.g., geographically).
• Thresholds adjusted by machine learning — the cutoff level evolves continuously as new patterns of fraud appear, so the model has to self-adjust on the fly.

Finding the right SaaS threshold is based mainly on historical data. Here are some of the steps that you can follow to achieve that:

1. Check past transactions to locate where fraud and genuine orders have clustered over risk score ranges.
2. Determine the quantities of regular customers turned away and the quantities of fraud cases identified across cutoff levels.
3. Instead of employing a single blanket threshold, group customers by risk indicators such as order value, geography, or product category.

Merchants typically consider the balance between risk and revenue when establishing fraud score thresholds. A reduction in the authorization cutoff point typically corresponds to a greater number of orders that present intrinsic risks. This may influence the frequency of chargebacks, fee collection, and the handling of penalties. Data suggests a connection between a higher cutoff point and the process verified customers undertake for specific requirements, potentially influencing sales conversion, revenue, and customer trust.

The precision of a fixed fraud detection threshold is influenced by shifts in fraud tactics. The majority of SaaS merchants review their thresholds at least quarterly, and when major sales events, new product lines launch, or there is a substantial increase in chargebacks or false declines.

Before making any decisions on where to devote your efforts, get a few of your thoughts straight.

1.   What aspects of 退款 和 误拒 are relevant to your business’s financial operations?
2.   With a high transaction volume, is evaluating the practicality of manually reviewing every business order pertinent?
3.   Has there been a change in the customer demographic or product offerings?

决策因素：

•       Industry-specific metrics for fraud prevalence
•       欺诈预防 methods and resources available
•       Acceptance of workload by human reviewers