Skip to content

Risk weights and the Suspect Score

What the Suspect Score measures, how to tune it, and when it affects orders.

Updated

Rules give a yes-or-no verdict. The intelligence tier adds a graded one: a Suspect Score between 0 and 1 for each visit, built from signals about the network, the browser and the visitor’s history. Turn it on with Enable intelligence tier (Smart Signals + Suspect Score) under Rules → Risk weights.

Turning it on changes nothing at checkout. The score is recorded and shown, and that is all, until you also turn on Apply Suspect Score to order tagging.

What goes into the score

Six categories each have a weight from 0 to 10. A higher weight makes that category count for more.

Category Raised by Default weight
Network Tor, VPN, proxy or hosting-provider address 4
Automation Automated browser, bot user agent, AI agent, virtual machine 7
Browser integrity A browser that misreports itself; privacy tooling counts for less 5
Velocity Many IP addresses or countries for one visitor within five minutes 4
Abuse Unusually high activity for your store, or shared reputation 8
Tenure Lowers the score for a returning visitor with history 5

A visit with no signals scores about 0.27.

Three signals are recorded but carry no weight, because they also fire for honest shoppers: a mismatch between the IP address’s time zone and the browser’s, a mismatch between the browser’s reported identity and the one seen by the network, and private browsing.

Confidence is a separate figure from 0 to 1. It says how much Protect could observe about the visit and how consistent the visitor is with their previous visit. It is not a risk score.

Set the threshold

A visit at or above the threshold counts as suspect.

Threshold preset Value Flags
Aggressive 0.5 More visits, including more honest ones.
Balanced 0.7 The default.
Conservative 0.9 Only the strongest cases.

Threshold (custom) accepts any other value.

Preview before you save

The preview rescores the last 30 days of your recorded sessions with the weights and threshold on screen, and draws the result as a histogram. It samples up to 5,000 sessions and tells you how many it sampled out of the total.

The preview is an estimate: it treats every visitor as new and leaves shared reputation out.

Changing a weight or the threshold clears the preview; run it again.

Apply the score to orders

When Apply Suspect Score to order tagging is on, an order whose visit scored at or above the threshold cannot be tagged sd-clean. It is tagged sd-suspicious instead. See Order tags.

Because the two switches are separate, you can run the score for a while as a record only, check it against the preview, and apply it to tagging afterwards.

If you move to the Free plan

Sessions keep their recorded signals, but the app shows a reduced set: Tor, country, network, automation flags and five-minute velocity.

Type to search…

↑↓ navigate↵ selectEsc close