Kyc aml · Production

Brex rebuilds customer onboarding as an AI-native multi-agent system

The problem

Brex's KYC and underwriting onboarding relied on manual judgment, implicit heuristics, and fragmented tools, creating a ceiling on velocity and scalability and causing onboarding to take days.

Workflow diagram · grounded in source
1
Startup segmentation
ai_action
“Segmentation agents: These agents leverage data sources like LinkedIn (via Clay), company website, and application information to determine at the very start of the application if a startup is either already professionally invested (PI) …”
2
Document verification and classification
ai_action
“Verification & OCR agents: Specialized sub-agents automatically process and validate high-friction documents in real time. This includes an OCR & classification agent that verifies proof of address, articles of incorporation, SAFEs, and …”
3
Identity and fraud evaluation
ai_action
“Identity & fraud agents: Beyond simple identity checks, these agents evaluate behavioral signals and anomalies. A dedicated fuzzy-match agent resolves name mismatches on IDs (e.g., "Johnny" vs. "John"), which has successfully reduced man…”
4
Underwriting financial profiling
ai_action
“Underwriting (UW) agents: These agents reconstruct a company's financial profile by automatically qualifying and mapping applicants to UW segment policies.”
5
Decision synthesis and routing
routing
“A final decision agent synthesizes evidence, confidence scores, and Brex policy into a single outcome. When confidence is high, decisions are made instantly. When confidence falls below a defined threshold (e.g., when data sources confli…”
6
Human analyst escalation review
human_review
“the case is escalated to a human analyst”
7
Supervised signal feedback
feedback_loop
“Those human decisions are then fed back into the system as supervised signals, continuously improving calibration and accuracy over time.”
Reported outcome

Brex's AI-native multi-agent onboarding now processes most eligible businesses in minutes, with card application auto-approval growing from 0% to 40%, an 85% reduction in business address RFIs, and a 70% reduction in manual identity reviews.

Reported metrics
Manual identity reviews reduced70%
business address RFIs reduction85%
Customer onboarding speedmost eligible businesses onboard to Brex in minutes
Onboarding time vs. prior processonce took multiple analysts and several days now happens in minutes
Reported stack
ClayLinkedInOCR
Source
https://www.brex.com/journal/rebuilding-onboarding-ai-native
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Brex's AI-native multi-agent onboarding now processes most eligible businesses in minutes, with card application auto-approval growing from 0% to 40%, an 85% reduction in business address RFIs, and a 70% reduction in…

What tools did this team use?

Clay, LinkedIn, OCR.

What results were reported?

Manual identity reviews reduced: 70%; business address RFIs reduction: 85%; Customer onboarding speed: most eligible businesses onboard to Brex in minutes; Onboarding time vs. prior process: once took multiple analysts and several days now happens in minutes (source-reported, not independently verified).

How is this kyc aml AI workflow structured?

Startup segmentation → Document verification and classification → Identity and fraud evaluation → Underwriting financial profiling → Decision synthesis and routing → Human analyst escalation review → Supervised signal feedback.

Related kyc aml cases