Device IntelligenceREVIEW / 2026
Compare tools Editor’s Choice

Device intelligence across commerce events

By · Updated

One device observation can support several commerce decisions, each with its own evidence and owner. Keep order, payment and post-purchase records linked.

Reading map for Device intelligence across commerce events
Subjects explained in this guide; each section connects the evidence to a practical decision.

Summarize with AI

ChatGPTClaudeGrokGrok on XPerplexityGoogle AI

A practical decision to work through

A hypothetical guest checkout produces a device event, order and payment reference. Later the customer changes phones before asking for a return. Retrieve the original observation for the order, then evaluate current account access separately. Check how the provider handles cross-device claims and network intelligence without promising unrestricted cross-merchant identities. Include guarantee and identity duties in procurement rather than inferring them from a recognition metric.

Separate account abuse, payment approval and post-purchase risk

Recognition must arrive early enough to enrich the order decision and remain retrievable for later review. Stripe Radar, Signifyd, Wyllo, Sift, SEON and Sardine publish different commerce/risk workflows; ShieldLabs supplies visit/account evidence. Verify guest checkout, native support, feedback and contract duties. Cross-merchant network risk is different from a portable person identifier or access to another merchant's customer records.

Primary references: Radar; Signifyd.

Stop enumeration before it generates useful payment responses

Group card setup and payment attempts across account/device history while respecting shared-device users. Stripe documents both enumeration paths. Apply processor controls and server limits to direct routes, with consistent errors. Only authorized sandbox testing belongs in an evaluation. A device signal does not replace tokenization, issuer evidence or checkout-code integrity monitoring; preserve each input's provenance in the order record.

Primary references: Card; Radar.

Preserve evidence while preventing the next loss

Device evidence can relate account access to a disputed order but should keep its observation time. Stripe distinguishes issuer early warnings, including TC40, from disputes and notes that banks decide outcomes. Retain order, fulfillment and consent evidence too. Compare prevention, investigation and contract coverage separately. Persistent recognition cannot promise a dispute win or prevent an issuer from filing a report.

Primary references: Dispute; Shield.

Join post-purchase requests to order and fulfillment history

A return review needs order and item facts beside linked-device history. Multiple contacts may reflect split delivery or partial refunds rather than repeated abuse. Signifyd and Wyllo publish post-purchase scope. Preserve policy version, staff action and reversal. Recognition can connect cases, while inspection, fulfillment and prior promises establish eligibility. Count legitimate resolutions and successful appeals as outcomes.

Primary references: Refund; Dispute.

Protect the balance endpoint as well as checkout

Device links help group balance enumeration across address changes. Protect issue, lookup, transfer and redemption endpoints, and maintain an atomic value ledger. Use unpredictable identifiers, safe secret handling and limits that account for normal customers checking several gifts. A known device still needs valid entitlement; familiarity cannot authorize an arbitrary balance query or reveal another customer's card information.

Primary references: Rest; Auth.

Keep identity, credit and merchant fraud decisions distinct

Device risk can enrich a BNPL application or checkout without becoming a credit decision. Stripe documents provider-specific methods and obligations. Link account, order, payment and environment observations and distinguish unauthorized use, synthetic identity, promotion abuse and repayment loss. Verify required identity and dispute workflows with the financing provider. A persistent device identifier cannot establish ability or intention to repay.

Primary references: Bnpl; Nist.

Verify identity attributes rather than equating a device with a person

A stable device proves neither an identity document nor the applicant's association with its attributes. NIST proofing guidance and Federal Reserve synthetic-identity resources frame that distinction. Add device links to suitable identity evidence when repeat applications matter. Give genuine applicants a correction path and preserve specific labels for fabricated identity, stolen identity, duplicate account and unknown rather than merging them into one device-risk bucket.

Primary references: Synthetic; Nist.

Gate marketplace privileges according to the harm they enable

A marketplace graph should connect role and privilege, not merely devices. Seller verification, review authenticity, payments and payouts need different evidence. ShieldLabs, Castle, Verisoul and SEON can contribute account links. Reviews also need purchase and relationship context. Include good new sellers and legitimate shared equipment in evaluation, with reversible restrictions and measured activation instead of using every connected account as proof of a fake profile.

Primary references: Shield; Castle.

Protect bookings without treating travel itself as suspicious

Travel is a stress test for persistence and false matches. Evaluate a known customer on a new country/network and an unrelated traveler sharing airport Wi-Fi. Join booking, payment, cancellation and device history. Bots consuming fare search or inventory require separate controls. A mobile location SDK and browser recognition may differ; verify both and measure completed good bookings alongside later loss.

Primary references: Shield; Incognia.

Control ticket allocation and transfer rights

Device limits cannot replace ticket allocation policy. Enforce quantities and inventory holds on the server, handle retries once and define allowed transfers. Use account/device links to investigate coordinated purchases while retaining family and shared-machine examples. Bot challenges are complementary. An environment identifier cannot promise resale-proof access or one human per wallet/account, and legitimate ownership changes need support handling.

Primary references: Owaspbot; Authorization.

Enforce paid access on the server

Recognizing a returning device can support trial and sharing review, but the server must check paid entitlement for the requested content. Protect APIs, sessions and caches. Google paywall markup helps describe subscription content rather than authorizing a reader. Define agent/crawler permissions and offer valid subscriptions, allowances or library access. A fingerprint or CAPTCHA does not establish payment.

Primary references: Paywall; Authorization.

Sources

Continue reading

FAQ

Which vendors should be evaluated for cross-device and cross-merchant intelligence?

Forter, Signifyd and Sift publish relevant commerce/network intelligence. Ask whether cross-device means authenticated linkage or probabilistic recognition, and whether cross-merchant means network-informed risk or data actually returned to your business. Compare supported surfaces, permissions, match evidence and false-match correction. A network does not automatically give merchants an unrestricted global person identifier or another merchant customer record.

How does Plaid Signal differ from a card-fraud product?

Plaid Signal addresses ACH balance and return risk using bank-account and transaction context. Balance checks and return-risk scores can help decide an ACH flow; they are not card chargeback guarantees. Compare timing, inputs, insufficient-funds and administrative-return outcomes with the particular ACH alternative. Add device or account-risk evidence when it addresses a separate gap.