Alder Finance: Synthetic Identities and SIM-Swap Account Takeover
Alder Finance, a digital-only lender, saw synthetic-identity fraud reach 2.3% of new applications and account takeover (ATO) reach 0.4% of existing accounts last quarter. Fraudsters combine real SSNs with fabricated identities, pass knowledge-based checks, then use SMS OTP; the fraud team linked 60 SIM-swap cases. Call-center agents reset passwords after verifying 'last four' and date of birth, and 22% of ATOs started with a call. Genuine customers abandon 18% of applications at document upload. The board wants both fraud and abandonment reduced. The lender has no device reputation feed, uses a single fraud score at onboarding, and does not monitor post-login behavior. A vendor offers behavioral biometrics and device intelligence; the CISO asks for a layered design, not a single score.
1. Design a layered control architecture across onboarding, authentication, and account recovery. Explain where each layer intervenes and why a single fraud score is insufficient.
Read the worked answer
Onboarding: move to identity proofing that verifies document authenticity, liveness, and selfie-to-document match, then confirms possession and control of the phone number. Add device and network intelligence, velocity checks across applications, and consortium signals for reused SSNs and addresses. Score at multiple points, not once; a synthetic identity often looks clean at the first check and becomes suspicious when the same device applies for several people.
Authentication: replace SMS OTP as the primary step-up with phishing-resistant passkeys or FIDO2 where possible, and treat SMS as a fallback with risk checks. Bind sessions to device signals and monitor post-login behavior: new payee, sudden limit increase, password change followed by withdrawal, and unusual navigation. Step up when behavior diverges, not only at login.
Recovery: this is where 22% of ATOs begin. Stop using last four and date of birth as verification. Require a cooling period for high-risk resets, verify via a known device or in-app approval, and escalate suspicious cases to video verification with document checks. For SIM-swap indicators, such as a recent number port or carrier change, block self-service reset and route to a specialist queue.
A single score fails because it collapses independent signals into one threshold. Layering lets a good document score be overridden by bad device history, and it gives investigators evidence about which control caught the fraud.
2. The board wants fraud and document-upload abandonment reduced together. Propose specific friction trade-offs and estimate what you would measure to prove the design is balanced.
Read the worked answer
Treat friction as targeted rather than universal. Save-and-resume at document upload, image-quality feedback before submission, and clear capture guidance reduce abandonment without weakening identity proofing. For low-risk applicants, allow a provisional decision and defer a second document to first withdrawal. For high-risk signals, such as a new device, mismatched address, or velocity hit, add manual review or a selfie challenge even though it increases friction for that segment.
Do not remove liveness or document authenticity checks to improve conversion; that trades fraud for growth and will raise losses later. Instead, reduce repeat capture and drop-off. A/B test the upload experience, measure completion by device type and document type, and separate genuine abandonment from fraud abandonment.
Metrics: synthetic fraud rate by control and by applicant segment; ATO rate per 10,000 accounts; SIM-swap-linked ATOs; percentage of ATOs originating in the call center; document-upload completion rate; time to complete; false-decline rate for genuine applicants; manual-review queue volume and turnaround; and customer complaints. Set a joint target, for example reduce fraud by 40% while holding abandonment at or below 15%, and report both weekly. Watch for displacement: if onboarding hardens, fraud may move to account recovery, so monitor recovery-channel ATO separately.
3. A vendor claims its behavioral biometrics and device intelligence will solve the ATO problem. How would you evaluate and govern that claim?
Read the worked answer
Run a bounded proof of value against Alder's own data, not a vendor reference. Define the decisions the tool will inform: step-up authentication, manual review, or block. Test on historical ATO and genuine-customer samples to measure detection at fixed false-positive rates, and segment results by channel, tenure, and device type. A vendor that performs well at onboarding may not detect post-login mule behavior.
Check integration and privacy. Device intelligence must work on web and mobile, survive app reinstalls, and avoid relying on permissions customers will refuse. Behavioral biometrics should collect only what is needed, with a clear notice and retention limit; legal must review consent and employee-monitoring implications. Ask how models are updated and whether Alder can export features for its own rules.
Governance: do not let the vendor score auto-block high-value customers. Start in monitor mode, then use it for step-up and review. Track precision, recall, alert volume, and customer friction. Require model-change notification and an annual independent validation. Contract for data ownership, deletion, and exit; avoid a proprietary signal that cannot be replaced.
Decision rule: adopt if it improves detection at an acceptable false-positive rate and reduces manual review, not merely because it adds a signal. Keep SMS risk checks and recovery controls regardless, because SIM swap defeats possession factors that the vendor may not observe.