Prompt
How do I build risk-based verification rules for new users?
Latest observation
A good way to build risk-based verification rules for new users is to turn verification into a decision system: score each signup by risk signals, then apply the lightest verification needed to get acceptable confidence.
1) Define the outcome you want
Start by deciding what “verification” is protecting against:
- fake accounts
- chargeback fraud
- identity abuse
- spam/bot signups
- account takeover attempts
- regulatory/KYC requirements
Different goals imply different rules.
2) Identify risk signals
Common signals for a new user:
- Email quality: disposable domain, typos, newly created email
- Phone quality: VoIP, prepaid, recently ported, invalid prefix
- Device signals: emulator, rooted/jailbroken, browser fingerprint anomalies
- IP / network: VPN, proxy, TOR, datacenter IP, risky geo
- Velocity: many signups from same IP/device/card/email domain in short time
- Behavior: copy/paste patterns, form completion speed, bot-like navigation
- Identity consistency: name/address/DOB mismatch, geolocation mismatch
- Transaction context: high-value order, high-risk product, shipping mismatch
- History: prior bans, prior disputes, duplicate identities
3) Create a risk score
Assign weights to each signal and compute a score like:
- Low risk: 0–29
- Medium risk: 30–69
- High risk: 70+
Example:
- Disposable email: +25
- VPN/datacenter IP: +20
- New device + unfamiliar country: +15
- Multiple signups from same device: +30
- Mismatch between billing/shipping country: +20
Keep it simple at first. Use historical fraud/abuse data to tune weights.
4) Map risk to verification steps
Use step-up verification only when needed.
Low risk
- Email verification only
- Optional CAPTCHA if you see bots
Medium risk
- Email + phone OTP
- CAPTCHA
- Light device or reputation checks
High risk
- Email + phone + stronger identity verification
- Selfie/liveness check
- Government ID check
- Manual review for borderline cases
The idea: don’t over-verify low-risk users, but raise friction where risk is high.
5) Add step-up triggers
Instead of verifying everyone upfront, trigger extra checks when users do risky actions:
- first payout
- first large purchase
- change of payment method
- address change
- login from new country/device
- rapid repeated actions
This improves conversion while still managing risk.
6) Include hard rules and soft scoring
Use both:
- Hard blocks for clearly abusive cases
- known bad IPs
- banned devices
- impossible age/location rules
- Soft scores for everything else
- borderline signals that increase friction but don’t block instantly
That avoids false positives.
7) Build exception handling
Have a path for legitimate users who get flagged:
- manual review queue
- alternate verification method
- appeal process
- support escalation
This is especially important for users in VPN-heavy regions or shared-network environments.
8) Monitor performance
Track:
- fraud/abuse rate by risk band
- verification completion rate
- false positive rate
- drop-off/conversion impact
- manual review volume
- approval rate by geography/device/channel
Adjust thresholds based on measured outcomes.
9) Test and iterate
Run A/B tests or phased rollouts:
- current rules vs. new rules
- different thresholds
- different verification paths
Optimize for the tradeoff between:
- fewer bad accounts
- lower user friction
- lower support burden
10) Keep legal/privacy in mind
Make sure your verification process:
- complies with data protection laws
- minimizes data collection
- has clear user consent and disclosure
- stores sensitive data securely
- avoids discriminatory proxies or unfair geographic bias
Simple example rule set
You could start with something like:
Score =
- Disposable email: +30
- VPN/proxy: +20
- Datacenter IP: +20
- 3+ signups from same device in 24h: +25
- Country mismatch with phone/IP: +15
- Previous abuse link: +40
Actions
- 0–19: email verify
- 20–49: email + phone OTP
- 50–79: email + phone + CAPTCHA + delayed access
- 80+: manual review or ID verification
Best practice
Start with a small, explainable ruleset, then evolve toward a hybrid of:
- rules for obvious abuse
- ML/analytics for subtle patterns
- step-up verification at the riskiest moments
If you want, I can help you design a sample risk-scoring model or a decision table for your specific product.