How to Build an IP Risk Score
Combine VPN, proxy, Tor, hosting and reputation signals into one IP risk score. A practical guide to weighting signals, setting thresholds and tuning over time.
A risk score turns a pile of separate IP flags into one number your rules can act on. You can use a ready-made IP fraud score, or build your own when you have data to tune it. Either way, the structure is the same — here's how to think about it.
The ingredients
A good IP risk score blends signals that each capture a different kind of risk:
| Signal | What it captures | Typical weight |
|---|---|---|
| VPN / proxy / Tor | Origin hidden | High |
| Residential proxy | Deliberate evasion | High |
| Hosting ASN | Server, not a person | Medium–high |
| IP reputation | Past abuse | High |
| Geo consistency | Masked/odd location | Medium |
| Velocity | Burst behaviour | Medium |
For the underlying concepts see what is IP reputation and what is an IP fraud score.
See the raw signals behind an IP
Step 1: gather signals from one lookup
Rather than calling many sources, pull the anonymiser, hosting and reputation signals from a single IP lookup via the IP reputation check or the detection APIs. Fewer calls, consistent data.
Step 2: weight and combine
Assign each signal a weight reflecting how predictive it is for your abuse, then combine into a 0–100 value. Start simple — a weighted sum capped at 100 — before reaching for anything fancier. Residential-proxy and reputation signals usually deserve the most weight.
Step 3: set action thresholds
Map the score to actions, not verdicts:
- 0–29 → allow
- 30–69 → challenge (verification / MFA)
- 70–100 → block or manual review on sensitive actions
Step 4: tune with outcomes
This is where a custom score earns its keep. Feed back real results — confirmed fraud, chargebacks, false positives — and adjust weights and thresholds. Re-tune periodically as abuse patterns shift; a score set once and forgotten drifts out of date.
Step 5: combine with non-IP signals
IP risk is one input. Blend it with device, behavioural and payment signals for a complete picture, especially against sophisticated residential-proxy abuse.
Bottom line
An IP risk score is a weighted blend of anonymiser, hosting and reputation signals condensed into one 0–100 number. Pull the signals from a single lookup, weight residential-proxy and reputation most heavily, map the score to allow/challenge/block bands, and tune continuously against your own outcomes.