New account fraud detection is the work of deciding, at the moment of signup, whether the account in front of you belongs to a real person, a duplicate of someone already inside, or an automated actor running at volume. The answer has to arrive in milliseconds, before the welcome email leaves and before trial credit or a signup bonus attaches to the account.
Most teams start with one signal — an email domain, an IP address, a captcha — and then spend a quarter arguing about what it means. This checklist replaces the argument with a written policy: five checks, one decision, three outcomes.
It assumes you own your signup flow and route the decision through Prynt Control. No data team required.
The five checks to run before you accept a signup
Run all five on every signup, then combine them. One check rarely justifies a block; two or three that agree usually do.
1. How many accounts already exist on this device
Fingerprint the device before the form posts. A device fingerprint survives incognito windows and cleared cookies, so a stable visitorId tells you whether this is the first account on that device or the ninth. Count accounts per device over a rolling window and read the number as a shape: one is normal, two or three is a shared laptop, dozens in a week is a policy problem rather than a user problem. How the identifier is built is covered in device fingerprinting.
2. Is the email disposable, aliased, or minutes old?
Disposable domains, plus-addressing and alias relays are cheap to produce. An alias is also a legitimate privacy choice, so treat it as weight, not a verdict: a disposable address on a clean device and a residential network can pass while skipping the bonus and the trial. The same address on a device that already made four accounts is a different decision.
3. Does the network look anonymized?
VPN, proxy, Tor exit and datacenter ranges show that the visitor chose to hide the origin network. Plenty of honest users do exactly that, and blocking every VPN punishes privacy-minded customers and remote teams. Score it instead: an anonymized network with a disposable email and a new device is a pattern; an anonymized network alone is one data point.
4. Is the client a browser, a script, or an AI agent?
Automation now includes headless browsers with patched fingerprints and AI agents that browse, fill forms and clear simple challenges. Separate the clients you want, such as a partner’s agent or a research crawler, from the ones you do not, such as signup loops. Automation markers, tampering and inconsistent client behaviour belong in the policy; see bot detection and AI agent detection.
5. What does shared reputation say?
A reputation network answers what you cannot answer alone: has this email hash, phone hash or device already appeared in abuse elsewhere? Reputation travels as hashes rather than personal data, and a first-seen identity is not suspicious by default — every real customer is new once.
From five checks to one decision
Combine the five checks into one of three answers, and record why. Here is a policy you can adapt:

| Device accounts | Network | Client | Action | |
|---|---|---|---|---|
| 1 | normal | residential | human | Allow |
| 2–3 | normal or alias | any | human | Allow, bonus once per device |
| 2+ | disposable | VPN or datacenter | human | Step-up: verify_email |
| 5+ in 7 days | any | any | human | Step-up: captcha and verify_email |
| any | disposable | datacenter | automation detected | Block |
Control answers each signup with allow, step-up or block, a reason code and a suggested step such as captcha, verify_email or card_or_passkey. Prynt never blocks on its own: your code receives the decision and chooses what the user sees. Keep the ladder shallow — allow the clean, step up the ambiguous, block the obvious — and make the block page say what the person can do next.
Watch the policy in mirror mode before you enforce it
Control starts with a 14-day mirror period: decisions are recorded, nothing is blocked. Use it to read the signups that would have been stopped, check them against support tickets and manual reviews, and move thresholds before a real customer hits a wall. Then switch on enforcement one use case at a time — signup first, then free_trial, then bonus — so a mistake in one flow cannot damage another.
What to measure after you turn it on
Track a short list and review it weekly:

- Step-up rate per use case, and how many people complete the step.
- Block rate split by reason code, so one noisy signal is visible.
- False-positive reports from support, tagged to the decision that caused them.
- Trial-to-paid conversion for allowed accounts versus stepped-up accounts.
- SMS cost per verified signup: Control caps OTP requests per device, number and network, which also limits pumping losses.
FAQ

Is a VPN a reason to block a signup?
No. An anonymized network is one signal among five. Block only when it lines up with a disposable email, multiple accounts on the device, automation or bad reputation.
Should a disposable email be blocked or stepped up?
Step it up by default. Hold the bonus and the trial, ask for email verification or a card, and keep the block for cases where the device or the reputation network already shows abuse.
Do I have to store raw emails, phones and cards to run these checks?
You can run the checklist with hashes. Control keeps phone numbers and cards only as keyed hashes, and email verification still proves the address receives mail. Document your retention choices with your legal team.
Prynt Control returns allow, step-up or block for signup, trial, bonus, login, checkout, SMS and paywall decisions, with a reason code and a suggested step your code can act on. It begins in a 14-day mirror period, so you see the decisions before anything is blocked. Start from https://pryntid.com.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.