Skip to content
Documentation
All docs

Core Concepts

How Risk Scoring Works

Understanding the 0-100 risk score, the three risk zones, and the 50+ signals behind every score.

Every return RefundSentry processes is assigned a risk score from 0 to 100. The score is the single number that drives every other decision in the app — auto-tagging, workflow routing, fulfillment holds, and the customer risk profile.

The three risk zones

The 0-100 score is bucketed into three zones. Default thresholds:

  • LOW (0-30) — Normal customer behavior. Process the return as usual. Most stores see 70% or more of returns land here.
  • MEDIUM (31-65) — Worth a quick human glance. The return has one or two unusual signals (slightly fast turnaround, second return this month, etc.) but nothing definitive.
  • HIGH (66-100) — Likely abuse or fraud. Multiple signals fired — wardrobing pattern, serial returner, bracketing, suspicious timing, etc. These are the returns to hold, scrutinise, and act on.

Zone boundaries are fully configurable in Settings → Risk thresholds. Stores with very generous policies often raise the HIGH boundary to 75; stores with strict policies lower it to 55. See Tuning Risk Thresholds for guidance.

How a score is built

A score is a weighted sum of 50+ behavioral signals, evaluated in parallel for every return. Signals fall into four broad categories:

  • Velocity signals — How often this customer returns relative to their order count, the time between purchase and return, and whether the return falls inside the wardrobing window (buy Friday, return Monday).
  • Pattern signals — Bracketing (multiple sizes/colors of the same SKU), exchange churning, repeat returns of high-value items, and reason-cluster anomalies (every return reason is "didn't fit" but the customer has 12 returns).
  • Customer context — Lifetime order value vs. lifetime return value, previous refundsentry tags, manual overrides, and chargeback history.
  • Order context — Order value, discount stacking, geographic anomalies (billing/shipping mismatch), and high-abuse SKU flags.

Every signal has a configurable weight. The default weights are tuned on cross-merchant benchmark data, but you can adjust any of them in Settings → Signal weights. Disable signals that don't apply to your business model — for example, "weekend wardrobing" doesn't make sense for a B2B store.

Where to find scores

  • Dashboard — Every return shows its score, zone, and the top 3 signals that fired. Filter by risk zone with the chips at the top.
  • Return detail page — Click any return to see the full signal breakdown, each signal's weight, and the customer's history of previous returns.
  • Customer profile — Aggregated risk across all of the customer's returns, with trajectory (rising / stable / declining).
  • Customer tags in Shopify admin — Customers are auto-tagged with refundsentry:low-risk, refundsentry:medium-risk, or refundsentry:high-risk so you can use the tag in your existing returns workflow, Flow rules, or merchant operations.

What scores are not

A risk score is not a verdict. It's a confidence-weighted summary designed to focus human attention. A 78 doesn't mean "this customer is a fraudster" — it means "multiple signals fired, please look at this one before processing." False positives are inevitable in any scoring system; the goal is to make human review cheap, not to remove it entirely.

For high-stakes decisions (refusing a refund, blocking a customer), always pair the score with at least one human review. Use Workflows to automate everything below that threshold.

Next steps