← Cases
United StatesFinance and insurance2025 – ongoing

Brex

AI expense submission, policy and audit tools in Brex spend management

Brex reported a 94% policy compliance rate after rebuilding its spend management product on Claude

WorkingVendor case studyVendor ↗

Problem

Finance teams wrote complex expense rules by hand and checked claims after the money was spent. Employees often learned of a policy breach only after a violation.

Approach

Brex began testing AI models in 2023 and chose Claude in Amazon Bedrock for its spend management product. Employees submit expenses in conversation, and the system uses calendar, location and spending context to categorize them. Finance teams upload policy documents instead of writing rules, and an AI workbench groups related expenses and flags policy concerns.

Results

94%policy compliance rate among Brex customers, against a 70% industry standard, per AnthropicVendor case study[1]
100%share of transactions reviewed by AI anomaly detection at line-item level, per AnthropicVendor case study[1]
$56.5 millionannual salary savings across Brex customers, per AnthropicVendor case study[1]

Takeaway

Moving policy checks from after the spend to before it is where Brex placed its compliance gain.

Categories

Business function
Finance, risk and legal
AI technology
Generative AI
Evidence
Vendor case study

Sources

Source summaries

[1]Anthropic25 Sep 2025

Anthropic describes Brex's rebuilt spend management product with conversational expense submission, policy interpretation and AI auditing. It gives a 94% compliance rate against a 70% industry standard, full line-item review and $56.5 million in annual salary savings across customers. The page carries no date; the Internet Archive first saved it on September 25, 2025.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026
Open case page ↗