A flagged signup lands in your queue with a score and a reason code, and that is when the real work starts. To investigate a fraudulent signup you have to turn an alert into evidence a second person can check: who the visitor is, what else the device has done, what the network says, and what happened on the page.
Most teams start with the email and stall. A disposable address proves little, and confirming it only shows the person can open an inbox they made a minute ago. Device and network evidence, correlated across accounts, is what separates one bad signup from a farm.
Prynt Identify provides that layer; Prynt Replay provides the session. Here is one analyst’s path through both, in the order the evidence arrives.
Start with the visitorId, not the email
When a signup fires, your server already holds a response: a decision, a risk score, reason codes and a visitorId. Save all four before you open anything else. The visitorId is a stable handle for that device and browser; the email and the phone number are fields the person can change in seconds.
Build the evidence note first
Open a note and paste requestId, visitorId, timestamp, decision, score and reasons. Every claim in your write-up should point at one of those fields. A reviewer who was not on shift cannot re-run your intuition, but they can check a requestId.
Count the accounts on the device
Query the visitorId against your users table, not the email. One device holding several accounts on different emails over weeks is the shape of signup farming. One account reached from many devices in a short window is the opposite shape, and it belongs in your account-takeover playbook, not this one.

The identity graph does the joining: shared device and network signals cluster accounts that never share an identifier on purpose. Read the cluster — size, age spread, whether anyone paid, and how the linked accounts were built with fraud rings. Three accounts created the same afternoon on one device is a decision you can defend; one account is a step-up at most.
Read the network signals in context
VPN, proxy, Tor and datacenter signals are context, not a verdict. A datacenter ASN on a consumer signup for a local service deserves a step-up; the same signal from a laptop behind a work VPN does not. Block on the raw flag and you reject real customers, then train support to distrust the score.
Rank each signal by what it costs you to be wrong. Blocking is expensive and hard to reverse; verification is cheap. Sort the network signals you collect, then reserve blocking for the cases where device, network and account evidence agree.
Check the device for tampering and automation
Tampering signals answer a narrow question: is this browser reporting what it actually is? Headless browsers, spoofed properties and patched fingerprints surface here. Bot and AI-agent detection answers a different one: was this typed by a person at all? Good device fingerprinting keeps those two answers separate.

Not all automation is abuse. An AI agent requesting a demo is a lead; an AI agent creating trial accounts in a loop is not. Record which one it looks like and treat the label as evidence, not as an automatic block.
Watch the session, then read the summary
Replay is where the abstract becomes concrete. Every form field is masked in the browser and masked again on the server, with no unmask option, so you can watch the whole signup without reading a name, an email or a card number.
Watch behavior instead: a form submitted faster than a person could type it, a field cleared and refilled three times, rage clicks on a button that never responded. Check JS errors and Web Vitals before you blame the user — a signup that failed during a script error is a bug report, not a fraud case.
Py summarizes the session and explains what went wrong. If your team works inside your own admin, link the session to the account with identify(userId, { accountId }) and open it in the embedded player.
Turn the finding into a decision you can repeat
An investigation ends with a policy, not a verdict. Write down which evidence combinations produce which outcome: a cluster on one device plus a datacenter ASN means block; a new device on a clean residential network means allow; anything ambiguous means step-up with email verification or a passkey. Prynt returns allow, step-up or block with a reason code and a suggested step — your code decides whether to act, and Control’s 14-day mirror period lets you watch those decisions on live traffic before anything is blocked.

Then automate the handoff: an Automate workflow can push {{trigger.decision}}, {{trigger.riskScore}} and {{trigger.reasons | json}} to Slack or your CRM in the same minute, so the next analyst starts where you finished.
FAQ
Is a VPN signup automatically fraud?
No. VPN, proxy and datacenter signals are context. A work VPN on a business signup is normal; a datacenter ASN on a consumer account with three sibling accounts is not. Use the signal to choose between allow, step-up and block.
Can I investigate a signup without reading the form data?
Yes. Replay masks every form field in the browser and again on the server, with no unmask option. You can watch pastes, retries, dead clicks and JS errors without touching personal data.
What if several accounts share one office device?
Check the network: many accounts from one office IP with separate payment instruments usually means a shared machine. Accounts that also share a fingerprint and a payment method are a different case. When the evidence is mixed, step up instead of blocking.
Try it free
Run your next flagged signup through both: Prynt Identify’s free plan covers 5,000 identifications a month with no card, and Replay’s free plan covers 1,000 sessions a month with 14-day retention. Create an account at 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.