# Spotting Repeat Fraudsters (Device Fingerprinting)
Source: https://paymentsandrisk.com/docs/fraud/detection/device-fingerprinting

Last updated 4 November 2026.

TL;DR

-   **Device fingerprinting** links orders that seem to come from the same phone or computer, so a fraudster who comes back with a new name, email and card can still look familiar. It's a match with a level of confidence, not a certainty
-   **A device match isn't a person ID. It can't tell you who was holding it**
-   Platforms can flag signs of VPNs, emulators, phone farms, remote desktop tools and anti-detect browsers. Each flag is a lead to check, because real customers can trip them too
-   Signals like **battery state**, **gyroscope data**, **TLS fingerprints** and **typing cadence** can add what transaction data alone can't, where the browser, device and permissions allow them
-   Vendors range from shared reputation networks (ThreatMetrix, Iovation) to behavioral analytics (Sardine) to managed decisions (Forter)
-   No single signal is definitive. The power comes from comparing the device, its connection, the browser and the behavior for details that don't agree

Where to go next

-   [Someone is hitting my checkout right now and I need it to stop](https://paymentsandrisk.com/docs/playbooks/survive-fraud-attack/)
-   [I don't want to buy anything yet. What can I do with the tools I have?](https://paymentsandrisk.com/docs/fraud/processor-rules-configuration/)
-   [Can device data help me win a chargeback I've already got?](https://paymentsandrisk.com/docs/chargebacks/representment/compelling-evidence/)

Your processor may already do some of this, so the question is whether you need more. Device fingerprinting is a way of recognizing the phone or computer placing an order. A small script reads details the browser gives away anyway, such as browser version, screen size, graphics chip and time zone, and builds an ID out of them. A native app SDK can read more, within what the platform permits. The point is to link a returning device, with some confidence, even when the fraudster comes back with a new name, a new email and a different card. A changed setting can break that link, and a shared family computer can make a false one ([Fingerprint's accuracy explainer](https://try.fingerprint.com/free-pdf-device-identification-accuracy-explained)).

[Stripe](https://docs.stripe.com/disputes/prevention/advanced-fraud-detection#activity-indicators) and [Adyen](https://docs.adyen.com/risk-management/) both document some of this, depending on your integration and plan.

Device fingerprinting started as an ID for a browser. Device intelligence goes further, across hardware, software, network and behavior. All of it tries to answer four questions. Is this a real device? Is a real person using it? Have we seen it before? Is someone else driving it? It gives you a read on each, not a certain answer.

This page is the signals, what each one can show, and what vendors say they offer. Putting them into rules? See [Building Fraud Rules](https://paymentsandrisk.com/docs/fraud/detection/building-fraud-rules/#device-intelligence-as-rule-inputs).

* * *

## How It Works

```mermaid
flowchart TD
    A[User Visits<br/>Site/App] --> B[SDK Collects<br/>Signals]
    B --> C[Transport Layer<br/>TLS, IP, ASN]
    B --> D[Device Layer<br/>Hardware, OS, GPU]
    B --> E[Browser Layer<br/>Canvas, WebGL, Fonts]
    B --> F[Behavior Layer<br/>Typing, Mouse, Sensors]
    C --> G[Cross-Layer<br/>Correlation]
    D --> G
    E --> G
    F --> G
    G --> H{Known<br/>Device?}
    H -->|Known Bad| I[Block or<br/>Review]
    H -->|Known Good| J[Trust<br/>Signal]
    H -->|New| K[Create<br/>Profile]
    K --> L{Signals<br/>Consistent?}
    L -->|Anomalies| I
    L -->|Clean| M[Normal<br/>Activity]

    style I fill:#b42318,color:#fff
    style J fill:#1f6f4a,color:#fff
```

### Fingerprint Types

A rough guide, not a measurement:

| Type | How It Works | Persistence | Evasion Difficulty |
| --- | --- | --- | --- |
| **Cookie-based** | Stores a token in the browser | Low (cleared easily) | Trivial |
| **Browser fingerprint** | Hashes browser attributes (user agent, plugins, fonts) | Medium (changes with updates) | Moderate |
| **Device fingerprint** | Combines hardware signals (GPU, CPU, screen, sensors) | Higher (can survive a browser change) | Harder |
| **Probabilistic/fuzzy** | Uses ML to match devices even when some attributes change | Higher (can survive cookie clearing and incognito), with some false matches | Harder |
| **Behavioral** | Learns how someone uses a device (typing, movement) | Builds over sessions | Hard to replicate at scale |

Vendors can combine several types and compare the layers, so spoofing one signal can leave inconsistencies in the others. Ask for the confidence level and known limits of any match. [Fingerprint's Smart Signals](https://fingerprint.com/products/smart-signals/) show what one vendor lists.

* * *

## Signal Categories

Platforms can collect signals across seven categories. No one category decides anything on its own. The useful reads are the inconsistencies _between_ them. What you can get depends on web or native app, the browser or OS, and permissions. So read each row below as a lead to check against your normal customers and your confirmed outcomes.

### 1\. Browser and App Signals

The traditional fingerprinting layer. Still useful, but easily spoofed in isolation.

| Signal | What It Reveals | What to Check |
| --- | --- | --- |
| **Canvas fingerprint** | Hash of rendered 2D graphics (varies by GPU, driver, OS) | A match with a known emulator or headless signature |
| **WebGL fingerprint** | GPU vendor and renderer string, 3D rendering output | `Google SwiftShader` points to a software GPU, which a headless browser, an emulator or an ordinary machine can use. `VirtualBox Graphics Adapter` points to a VM |
| **Audio fingerprint** | Audio processing output, which varies with the hardware and OS | Identical output across many sessions. Don't assume it's absent in headless, because [current headless Chrome](https://developer.chrome.com/docs/automation-and-testing/headless) shares normal Chrome's functions |
| **Font enumeration** | Installed font list | A Windows device with Linux fonts may be a spoofed user agent |
| **Navigator properties** | Plugins, languages, hardware concurrency, device memory | [`navigator.webdriver`](https://developer.mozilla.org/en-US/docs/Web/API/Navigator/webdriver) `true` means automation is driving the browser (Selenium, Puppeteer). It doesn't say why |
| **User agent** | Browser and OS identification | Easily spoofed, but inconsistencies with other signals are revealing |

Anti-Detect Browsers

[Multilogin](https://multilogin.com/antidetect/antidetect-browser/), [GoLogin](https://gologin.com/) and [Dolphin Anty](https://dolphin-anty.com/) let users isolate browser profiles and control these signals. They have legitimate uses, and fraudsters can use them too. Platforms can sometimes spot them when the browser attributes don't line up with harder-to-fake signals like TLS fingerprints, sensor data and behavior. Don't assume every one gets caught.

Looking for Basic IP Enrichment?

Want server-side IP lookups with no SDK? Geolocation, VPN detection and datacenter detection are available by API. See [Data Enrichment](https://paymentsandrisk.com/docs/fraud/detection/data-enrichment/#ip-intelligence). This section goes further, and each technique runs in a different place. WebRTC tests run in the browser. TLS fingerprints come from the server or network edge, and [Cloudflare](https://developers.cloudflare.com/bots/additional-configurations/ja3-ja4-fingerprint/) exposes them in its rules and analytics. Spotting a residential proxy by device correlation needs a device signal.

### 2\. True IP and Network Intelligence

The IP is easy to hide. Some platforms try to look past the visible one.

| Technique | How It Works | What It Can Point To |
| --- | --- | --- |
| **WebRTC leak detection** | Creates a hidden peer connection via STUN servers. Under some network and browser settings, UDP traffic can bypass a VPN tunnel and expose the real IP | VPN users whose real IP leaks through WebRTC |
| **TLS fingerprinting (JA3/JA4)** | Hashes the TLS ClientHello message (cipher suites, extensions), at the server or edge. Many similar clients share a fingerprint, and it can be missing | User agent says "Chrome" but the TLS fingerprint matches Python's `requests` library: possibly a bot. It isn't a person's ID or a hidden IP ([Cloudflare](https://developers.cloudflare.com/bots/additional-configurations/ja3-ja4-fingerprint/)) |
| **ASN/datacenter detection** | Maps IP to owning organization. Classifies as residential, mobile, datacenter, or hosting | A datacenter IP at a consumer checkout is a tell worth checking. Some real shoppers browse through VPNs or proxies, so weigh it with the device and behavior signals |
| **Residential proxy detection** | Correlates IP rotation patterns with stable device fingerprints | IP changes every request while the device fingerprint stays constant: possibly a rotating proxy. Weigh it with account history |
| **Timezone vs. IP mismatch** | Compares browser timezone (`Intl.DateTimeFormat`) with IP geolocation | IP in London, timezone set to `America/Los_Angeles`: possibly a VPN, possibly a traveler. Weigh it with account and device evidence ([SEON](https://docs.seon.io/knowledge-base/device-intelligence/residential-proxy-and-vpn-detection)) |

Several vendors sell "True IP" technology, ThreatMetrix and Sardine among them. It tries to surface the actual IP behind a VPN or proxy. Accuracy swings by technique, and by how the VPN is configured. Don't treat it as ground truth.

**IP classification matters as much as the IP itself.** It tells you who owns the address, not whether the customer is real ([SEON on IP types](https://docs.seon.io/knowledge-base/digital-footprint/digital-footprint-widgets)). Measure the risk on your own orders.

| IP Type | Example Owner | How to Read It |
| --- | --- | --- |
| Residential | Comcast, AT&T, BT | Expected for consumers. A residential proxy can look the same |
| Mobile/Cellular | T-Mobile, Vodafone | Expected for mobile. Many customers can share one IP |
| Datacenter/Hosting | AWS, DigitalOcean, OVH | Worth checking at a consumer checkout. Real customers can come through VPNs, shared networks or remote environments |
| Known VPN provider | NordVPN, ExpressVPN ranges | Real customers use these. Weigh it with device and account history |
| Known proxy service | Bright Data, Oxylabs | Worth checking, with device and account history |
| Tor exit node | Published exit node lists | Worth a closer look, and easy to check against the published lists |

### 3\. Device Integrity

These signals check the device itself. Has it been tampered with, or is it an emulator? Is someone driving it remotely? Many of them need a native app SDK. A web script sees only what the browser exposes, and it can't read native process lists or hardware IDs like the IMEI.

| Signal | What You'd Usually See | What to Check |
| --- | --- | --- |
| **Emulator detection** | Real hardware identifiers (e.g., `samsung/...`) | Generic build strings (`goldfish`, `ranchu`, `sdk_gphone`), SwiftShader GPU, missing sensors |
| **Root/jailbreak** | Stock OS, normal permissions | Presence of `su` binary, Cydia/Magisk, SELinux permissive, writable system partition. Some real users root their own phones |
| **Remote desktop** | No remote access software active | TeamViewer, AnyDesk, Windows RDP, Zoom screen sharing detected |
| **VM detection** | Real GPU, normal CPU | VirtualBox/VMware graphics adapters, low CPU core count, VM-specific BIOS strings |
| **Developer tools** | `navigator.webdriver` is false | `navigator.webdriver === true`, empty plugins array, CDP connection active. Developers and test tools set these too |
| **Frida/instrumentation** | No runtime hooks | Frida toolkit detected (commonly used to bypass security checks and manipulate API responses) |

Remote Desktop Detection Matters

Remote access scams get past ordinary device checks. A fraudster calls the victim and talks them into installing TeamViewer. Then they drive the victim's device and make the transaction themselves. It comes off the victim's real device, on their real IP. Traditional signals look clean. Spotting active remote desktop software is one of the few ways to catch it.

### 4\. Behavioral Biometrics

How someone _uses_ a device is harder to fake at scale, though automation can imitate some of it. [Behavioral Analytics](https://paymentsandrisk.com/docs/fraud/detection/behavioral-analytics/) covers how to test these.

**Typing and keystroke dynamics:**

| Signal | What It Measures | What to Check |
| --- | --- | --- |
| Typing speed | Characters per second, words per minute | Perfectly uniform timing may be automation |
| Dwell time | How long each key is held | Zero variance may be injected keystrokes |
| Flight time | Gap between releasing one key and pressing the next | Identical intervals may be scripted input |
| Segmented typing | Typing in bursts with pauses (switching to reference material) | Name or address typed in fragments. It may be someone reading a stolen data sheet, or someone checking their own details |
| Copy-paste in identity fields | Whether name, SSN, or address was pasted vs. typed | Pasted identity details. [Autofill](https://www.w3.org/WAI/WCAG22/Understanding/identify-input-purpose.html) and password managers are legitimate, so compare with your real customers |

**Mouse and pointer behavior:**

| Signal | What It Measures | What to Check |
| --- | --- | --- |
| Movement trajectories | Cursor path between targets | Perfectly straight lines or geometric patterns may be scripted |
| Micro-movements | Tiny jitter when cursor is "still" | Zero jitter may be a bot. Trackpads and assistive tools can differ |
| Click patterns | Timing, location, frequency | Uniform click intervals may be automation |
| Scroll behavior | Speed, direction, pauses | Uniform scroll with no pauses may be a bot |
| Guided movement | Cursor controlled by someone else (latency artifacts) | Possibly a remote access tool in use |

**Mobile sensor data:**

These depend on the platform. [Sensor APIs](https://developer.mozilla.org/en-US/docs/Web/API/Sensor_APIs) can be unavailable or blocked, [motion access](https://developer.mozilla.org/en-US/docs/Web/API/DeviceMotionEvent/requestPermission_static) can need a tap or a permission, and [pointer pressure](https://developer.mozilla.org/en-US/docs/Web/API/PointerEvent/pressure) has a default value on hardware that doesn't report it. Missing data isn't proof of an emulator.

| Signal | What It Measures | What to Check |
| --- | --- | --- |
| Gyroscope | Device orientation and rotation | Zero readings may be an emulator or a device on a rack. Or a phone on a table |
| Accelerometer | Movement and vibration | Perfectly static may mean nobody's holding it. A phone on a stand looks the same |
| Touch pressure | Force of finger on screen | Uniform pressure may be automation |
| Device orientation | How the phone is held | Static at ~90 degrees may be a rack, or a phone on a stand |

### 5\. Battery and Power Signals

A small category. On the web it depends on the [Battery Status API](https://developer.mozilla.org/en-US/docs/Web/API/Battery_Status_API), which some browsers support, including some on Android, and others don't. Native apps read it through their platform's own APIs. A missing reading doesn't mean an emulator.

| Signal | What You'd Usually See | What It May Point To |
| --- | --- | --- |
| Always at 100%, always charging | Battery fluctuates throughout the day | A phone farm on USB hubs. Or a laptop that stays plugged in |
| Battery level never changes across sessions | Varies naturally | An emulator reporting a static level like `0.50` or `1.00`, or a browser hiding the real one |
| Inconsistent battery between "same user" sessions | Consistent within short timeframes | Account sharing or credential selling. Or a customer with two devices |

Battery data alone is weak, and it's only useful as corroboration. Take a device that's always charging and never moving the gyroscope. It's turned up on 50 accounts. Investigate a possible phone farm, and rule out shared equipment and other honest explanations. Even together, those signals don't tell you who used the device.

### 6\. Phone Farm and Fraud Factory Detection

A phone farm is a rack of devices pretending to be independent users. Physical or virtual, doesn't matter. No single signal catches them, so don't go hunting for one. You catch farms on the composite profile. A native app SDK sees more of this than a web script.

| Signal | What a Farm Device Can Look Like |
| --- | --- |
| Battery | Always at 100%, always charging (plugged into USB hubs) |
| Gyroscope/accelerometer | Zero movement (mounted on rack, not held by a human) |
| Device attributes | Same hardware fingerprint appearing across many "different" accounts |
| Factory resets | Frequent resets (every few hours/days) to create "fresh" identities |
| Device ID cycling | IMEI, Android ID, or advertising ID changes on the same hardware, where a native SDK is permitted to see them |
| Network patterns | Many devices sharing the same IP range or cycling through a small pool |
| Behavioral uniformity | All devices show identical interaction patterns, timing, and session flow |
| No natural idle periods | Legitimate users sleep. Farm devices may operate 24/7 |
| Physical proximity | Dozens of "independent" devices co-located in the same physical space |

### 7\. Consortium and Reputation Data

The simplest signal here is the one to ask first. Has this device been linked to fraud before?

Consortium networks pool device reputation across their members. What you get is the provider's own signal, at the access and update speed it documents, not every other merchant's report ([TransUnion's Device Risk](https://www.transunion.com/solution/truvalidate/digital-insights/device-risk) is one). Ask how old a flag is and what caused it.

| What Consortium Data Can Tell You | Example |
| --- | --- |
| Device has a bad history | Flagged at 3 other merchants in the past 90 days. If the flag came from a chargeback, that's a dispute, not confirmed fraud |
| Device is brand new to the network | Never seen before, no history. Test whether that means more risk on your new accounts |
| Device has a long history | First seen 2 years ago, nothing flagged across 12 merchants. A good sign, not a guarantee |
| Device is linked to a suspected ring | Same device cluster seen in coordinated attacks across the network |

* * *

## How Signals Work Together

No single signal is definitive, so you layer them and cross-reference. It's the inconsistencies between layers that give you the strongest read.

**Example: Spotting a possible anti-detect browser**

An anti-detect browser can spoof canvas fingerprint, WebGL renderer, user agent, timezone and language. But say you also see this:

-   The TLS fingerprint (JA3/JA4) matches a Chromium build, not the user agent it claims
-   The battery API reports a static level
-   Typing patterns don't match the account's historical behavior
-   The IP is residential, but the ASN belongs to a known residential proxy provider

Any one of these alone might be explainable. All four together are a strong reason to investigate. They still don't prove it.

**Example: Spotting a possible remote access scam**

The transaction comes from the victim's real device, real IP, real location. Traditional signals look clean. But say you also see this:

-   Remote desktop software (TeamViewer) is active
-   Mouse movements show latency artifacts consistent with remote control
-   The user is on an active phone call during the session, where a native SDK is permitted to see that
-   Typing patterns show hesitation and segmented entry (possibly a victim being coached)

* * *

## Vendors

Vendors vary a lot. What they collect, how big the network is, how they hand you a decision.

### What Vendors Say They Offer

These are each vendor's own claims, read 3 November 2026, not independent results. Ask what you'd get on your product, platform and plan, including anything not listed here. Network sizes count different things, so don't rank them.

| Vendor | What Its Own Material Covers | Ask About |
| --- | --- | --- |
| [Sardine](https://www.sardine.ai/device-and-behavior) | Device and behavior signals. [It reports](https://www.sardine.ai/blog/how-hackers-are-targeting-b2b-payments-with-remote-access-tools) spotting remote screen-sharing tools with 92% precision and a 96% detection rate, which is its own figure | The population and period behind those numbers |
| [ThreatMetrix and BehavioSec](https://risk.lexisnexis.com/products/behaviosec) (LexisNexis) | Combined device and behavior | Current network size and what it counts |
| [Iovation](https://www.transunion.com/solution/truvalidate/digital-insights/device-risk) (TransUnion) | Device reputation, with its own network claims | What a flag means and how old it is |
| [Kount](https://assets.equifax.com/marketing/US/assets/payments-fraud.PDF) (Equifax) | Policies, case and risk functions, with its own network scope. Its legacy [`PC_REMOTE` field](https://developer.kount.com/hc/en-us/articles/4411149691540-Troubleshooting-Questions-for-Device-Data-Collector) says a device is enabled for PC Remote software, not that a remote session is active | The [current Orders API mapping](https://developer.kount.com/hc/en-us/articles/37233969016084-Risk-Inquiry-Service-to-Orders-API-Comparison) before relying on a legacy field |
| [Sift](https://sift.com/) | Payment, account and abuse risk | Score types and packaging on your plan |
| [Forter](https://www.forter.com/pricing-offer/) | Covered and uncovered pricing, so a guarantee is a contract choice | Whether you can write your own rules, and what your contract covers |

### Choosing a Vendor

The right vendor matches the biggest problem you've measured. Not the longest feature list. Then compare a real quote and test it on your own traffic.

| If Your Primary Problem Is... | Look At |
| --- | --- |
| **Sophisticated fraud bypassing basic rules** (anti-detect browsers, residential proxies, emulators) | Sardine, ThreatMetrix. Ask how each handles these on your traffic |
| **Needing device reputation at scale** (has this device been seen in fraud before?) | ThreatMetrix, Iovation. Ask how much of the network is in your own industry and country |
| **Remote access / social engineering scams** | Sardine, or [BioCatch](https://www.biocatch.com/account-takeover-protection/remote-access-attacks), which is aimed at banks. The 92% precision figure is Sardine's |
| **Identity verification + device signals** (synthetic identity, application fraud) | Kount (Equifax), or [NeuroID](https://www.experian.com/business/products/neuroid) (Experian) for form interaction. Ask what data each uses on your plan |
| **Multi-abuse-type detection** (payment fraud + promo abuse + content abuse + ATO) | Sift. It scores by abuse type, so ask which types your plan covers |
| **Not wanting to manage rules at all** (outsourced decisions, possibly with a financial guarantee) | Forter. A covered contract reimburses chargebacks on orders it approved, under its terms. That isn't a network liability shift |
| **Budget-conscious / just getting started** | Fingerprint, which publishes a price, or your processor's built-in tools (Stripe Radar, Adyen Protect) |

### Pricing and Accessibility

| Vendor | Price | Primary Use Case |
| --- | --- | --- |
| **[Fingerprint](https://fingerprint.com/pricing/)** | Free up to 1,000 API calls/month. Pro Plus from $99/month for 20,000 calls on web and iOS, then $4 per 1,000 more. Android has its own allowance of 500,000 | Device identification, returning visitor recognition |
| **[SEON](https://seon.io/pricing/)** | Starter $699/month (2,500 fraud checks, 10 users, 50 rules). Premium is custom | Email and phone enrichment plus device signals |
| **ThreatMetrix** (LexisNexis) | Ask for a quote | Device intelligence, shared reputation |
| **BioCatch** | Ask for a quote | Behavioral biometrics for banks, social engineering detection |
| **NeuroID** (Experian) | Ask for a quote | Form interaction analytics, application fraud detection |
| **Iovation** (TransUnion) | Ask for a quote | Device reputation |
| **Sardine** | Ask for a quote | Behavioral signals and remote desktop detection |

The Fingerprint and SEON rows come off their own pricing pages, read 4 November 2026. Check them again before you budget.

**Budget-conscious?** Fingerprint and SEON both publish a price. Fingerprint has a free plan. SEON publishes none: its Starter is **$699/month for 2,500 fraud checks**, ten users, fifty custom rules. That's roughly **28 cents a fraud check**. Fingerprint's Pro Plus works out to about half a cent a call at its included 20,000. Those aren't the same unit, because SEON's a wider fraud platform, so compare what each call or check includes. Use both as a yardstick when a quote-only vendor sends a proposal.

**Enterprise?** Expect to talk to sales at ThreatMetrix, BioCatch, NeuroID, Iovation and Sardine. Contract length and any minimum volume are questions for them. The only way to find out if you're too small is to ask: "At \[monthly orders\], is there a minimum commitment, and what would I pay in year one?" A vendor that won't quote you has answered the question. If that happens, start with Fingerprint or your processor's built-in tools (Stripe Radar, Adyen Protect).

Processor Tools vs. Standalone Vendors

On Stripe Radar or Adyen Protect? You've got some of this already. [Stripe](https://docs.stripe.com/disputes/prevention/advanced-fraud-detection#activity-indicators) documents device and activity signals through Stripe.js, and [Adyen](https://docs.adyen.com/risk-management/) has its own basic and premium scope. A standalone vendor may add more, such as behavioral biometrics, sensor data, True IP or shared reputation. Name the gap you've measured, then compare a real quote and the time your checkout can spare. See [Processor Rules Configuration](https://paymentsandrisk.com/docs/fraud/processor-rules-configuration/) for what each processor includes.

* * *

## Use Cases

### Fraud Detection

| Use Case | How Device Intelligence Helps |
| --- | --- |
| **[Card testing](https://paymentsandrisk.com/docs/fraud/types/card-testing/)** | Same device cycling through hundreds of card numbers points to a single source |
| **Multi-accounting / [promo abuse](https://paymentsandrisk.com/docs/fraud/types/promo-abuse/)** | Same device fingerprint across multiple "different" accounts. Could also be a shared household device |
| **[Account takeover](https://paymentsandrisk.com/docs/fraud/types/account-takeover/)** | New device + new location + new behavioral pattern on existing account |
| **Fraud ring linkage** | Cluster of devices with shared attributes, same IP ranges, same behavioral patterns |
| **Application fraud** | Signs to investigate: emulator flagged, pasted identity fields, missing sensor data |
| **Remote access scams** | Signs to investigate: active remote desktop software, mouse latency artifacts, an active phone call where a native SDK can see one |

### Account Security and Trust

| Use Case | How Device Intelligence Helps |
| --- | --- |
| **Step-up authentication** | Trigger MFA on unknown device (see [3DS](https://paymentsandrisk.com/docs/fraud/prevention/3ds/)) |
| **Trusted device recognition** | Returning device with clean history = lower friction |
| **Session management** | Limit active devices per account |
| **[Compelling evidence](https://paymentsandrisk.com/docs/chargebacks/representment/compelling-evidence/)** | A device match is one element of the purchase history Visa CE 3.0 needs. On its own it isn't enough |

* * *

## Implementation Approaches

### Build vs. Buy

| Approach | What You Get | What to Weigh |
| --- | --- | --- |
| **In-house** (FingerprintJS open source) | Basic device fingerprinting, full control, no data sharing | No shared reputation, and you maintain it yourself |
| **Device ID vendor** (Fingerprint) | Device IDs and the [Smart Signals](https://fingerprint.com/products/smart-signals/) on its plan | Test the accuracy it advertises on your own traffic. Ask about behavior signals and shared reputation |
| **Full platform** (Sardine, ThreatMetrix, Sift, etc.) | Wider signals and a rules engine, by product | Cost, data sharing terms, vendor dependency |
| **Processor-included** (Stripe Radar, Adyen) | Some device signals and ML scoring, by plan and integration | How much of the signal you can see and write rules against |

See [vendor selection guide](https://paymentsandrisk.com/docs/fraud/vendors/selection-guide/) for evaluation criteria.

### What to Ask a Vendor

Ask these on the vendor call, in these words:

1.  **What signals do you collect beyond basic fingerprinting?** (Behavioral biometrics, sensor data, TLS fingerprints)
2.  **How do you handle anti-detect browsers and residential proxies?** (Cross-layer correlation, not just database lookups)
3.  **How big is your consortium, and is it relevant to my vertical?** (A billion devices doesn't help if none are in your industry)
4.  **Can I write custom rules against your signals, or are decisions fully managed?**
5.  **What's the latency?** (Compare it with the time your own checkout can spare)
6.  **How do you handle privacy compliance?** (Data tokenization, consent management, GDPR readiness)

* * *

## Privacy Considerations

Check the Privacy Rules First

A fingerprint reads information from the customer's device, so these rules can apply before there's an order:

-   **EU and UK:** Reading device data can need prior consent, unless it's strictly necessary for a service the customer asked for ([EDPB guidelines](https://www.edpb.europa.eu/system/files/2024-10/edpb_guidelines_202302_technical_scope_art_53_eprivacydirective_v2_en_0.pdf)). A GDPR legitimate interest doesn't replace that consent where it's required. Ask counsel where a fraud check falls
-   **California:** The CCPA covers purpose limits, consumer rights and sharing as well as disclosure, and its security and fraud exception has conditions ([CPPA FAQ](https://cppa.ca.gov/faq.html))
-   **Elsewhere:** Check local rules

Talk to counsel before you implement it. Sharing with a vendor's consortium needs its own look.

### Best Practices

1.  **Transparency:** Disclose device fingerprinting in your privacy policy
2.  **Purpose limitation:** Use only for fraud prevention and security
3.  **Data minimization:** Collect only what's needed for fraud detection
4.  **Retention limits:** Expire device profiles (consortium retention may differ)
5.  **Consent where required:** Cookie banners, opt-in where legally necessary
6.  **Vendor data sharing:** Know what your vendor shares across its consortium, and on what terms. [TransUnion's iovation notice](https://www.transunion.com/privacy/iovation) is one to read

## Rolling It Out

**Just getting started?**

1.  Start with what your processor's tools already do (Stripe Radar, Adyen Protect)
2.  Build velocity rules ([how they work](https://paymentsandrisk.com/docs/fraud/detection/velocity-rules/)) using device ID as a dimension
3.  [Review privacy requirements](https://paymentsandrisk.com/docs/fraud/detection/device-fingerprinting/#privacy-considerations) before adding standalone tools

**Adding standalone device intelligence?**

1.  [Choose a vendor](https://paymentsandrisk.com/docs/fraud/detection/device-fingerprinting/#choosing-a-vendor) based on your primary fraud problem
2.  Integrate it and run shadow mode until your planned sample and outcome window are ready. Thirty days is a planning start, not proof
3.  Compare the new signals to your current catches and false positives

## Do next

-   [Build behavioral patterns into your review process](https://paymentsandrisk.com/docs/fraud/detection/manual-review/)
-   [Building Fraud Rules](https://paymentsandrisk.com/docs/fraud/detection/building-fraud-rules/): Using device signals in rules
-   [Running Fraud Operations](https://paymentsandrisk.com/docs/operations/running-fraud-operations/): Operational cadence

## Deeper

-   [Prevent Fraud](https://paymentsandrisk.com/docs/fraud/): The fraud hub
-   [Data Enrichment](https://paymentsandrisk.com/docs/fraud/detection/data-enrichment/): Server-side IP, email, phone signals (no SDK)
-   [Risk Scoring](https://paymentsandrisk.com/docs/fraud/prevention/risk-scoring/): Incorporating device signals into scores
-   [Behavioral Analytics](https://paymentsandrisk.com/docs/fraud/detection/behavioral-analytics/): Complementary detection method
-   [Evidence Framework](https://paymentsandrisk.com/docs/fraud/detection/evidence-framework/): How device data fits Tier 1/Tier 2
-   [Fraud Vendors](https://paymentsandrisk.com/docs/fraud/vendors/landscape/): Every vendor profile, with prices where they're published
-   [Fraud Model Feedback](https://paymentsandrisk.com/docs/fraud/detection/fraud-model-feedback/): How device signals improve ML
