Expense Policy Agent: Get an AI agent to enforce your policy across every entity

Every expense your team submits lands with the same question implicitly attached. Is this in policy? Almost always, the answer is a guess. Not because the policy doesn't exist, but because nobody reads a 14-page PDF for a $47 lunch.

Here's what actually happens. An approver opens the dashboard. Scans the amount. Remembers roughly what the policy said. Clicks approve. Moves on. Multiply that across hundreds of expenses a week, and what you have isn't enforcement. It's rubber-stamping.

Most expenses get a five-second glance. Out-of-policy spend slips through at month-end. And the ones that do get flagged ping-pong between approver, finance, and submitter until no one's quite sure who owns the decision.

Your finance team deserves better than this. Every expense evaluated the moment it's submitted, across every entity your business operates in, without a PDF in sight. That's what Expense Policy Agent does.

Expense Policy Agent is your always-on AI reviewer, built directly into Airwallex Spend. It reads your policy. It evaluates every card expense and reimbursement your team submits. It flags exactly what's wrong, and cites the rule that caught it.

Built for global finance, from day one

Global finance doesn't run on a single policy. It runs on a master policy, layered with entity-specific rules, stretched across multiple currencies and languages. Most expense tools force a choice: flatten that complexity into one-size-fits-all rules, or spin up separate instances per region. Neither are scalable, and that’s where every other AI expense agent on the market hits a wall.

Airwallex Expense Policy Agent was built to solve for this from day one.

• Entity-specific rules: Maintain a master policy for fundamentals. Layer local rules on top. Your UK subsidiary runs different per-diem rates than your US office, and the agent knows.

• Multi-language: Receipts and policies in French, Chinese, Japanese, Spanish, Arabic, read and evaluated natively. Flags surface in the user's own dashboard language.

• FX-aware: Policy limits compare against historical exchange rates, not raw conversions. A hotel in euros hits the right regional cap, every time.

A ¥12,000 client dinner in Tokyo isn't the same as a $120 one in Austin. Your policy knows that. Now your reviewer does too.

One global rule-set, fluent in your entities, currencies, and languages.

How Expense Policy Agent works

Upload your existing policy, or generate one from best practices. The agent converts it into enforceable rules. From that point on, every expense your team submits is evaluated automatically. Compliant expenses clear with a green badge. Out-of-policy items are flagged with a reason and a direct link to the policy section that applies.

You're always in control. Approvers can dismiss any flag. Admins can surface overrides and create lasting exceptions. Test mode lets you validate the agent on your policy before rolling it out company-wide. Hidden notes let you add internal exceptions only admins see, like "The US$50 dinner limit doesn't apply to the Sales team during client closing week." And your policies, flags, and business data are never used to train any shared model.

Your employees learn policy in the moment. Most expense tools punish mistakes after the fact. Submit, wait two weeks, get a terse rejection from a frustrated approver, learn the rule through embarrassment. Expense Policy Agent flips that. When an employee submits an expense, any flagged items are surfaced instantly in clear, easy-to-understand language. It provides a direct link to the exact section of the policy that applies, so your team absorbs the rules in the moment they need them.

Proof from 150,000 evaluations

Three months of early access testing clearly shows the agent is as reliable as a human approver.

• Up to 73% of expenses need zero human review: Most expenses are routine, compliant, cleared automatically. Things like coffee, office supplies, standard travel bookings.

• 99.4% verification alignment: When the agent marks marks an expense as verified, human approvers agree almost all the time. Which means agent-verified expenses can safely be fast-tracked.

• Up to 13 concurrent policies: Sophisticated organizations manage complex global hierarchies. By starting small and carefully refining exceptions, they safely scaled Expense Policy Agent from local entities to cover their entire global footprint.

What's here today, and what's coming next

Expense Policy Agent is now rolling out globally to Grow and Accelerate customers. Baseline checks run from day one with no setup required. Upload your own policy, and the agent starts enforcing your specific rules across card expenses and reimbursements.

Two more releases are already in motion:

Approval mode. Today the agent flags and verifies. Soon it will move into your Spend workflows directly, auto-approving compliant low-risk expenses and escalating flagged items to humans.

Embedded guidance. "What's my dinner budget when travelling to London?" Answered instantly, with city-specific caps, powered by your own policy.

Enforce every rule, on every expense, everywhere.

Policy enforcement used to mean reviewing a sample, hoping your approvers remembered the rules, and finding out at month-end which ones they didn't. On Airwallex, finance teams are freed up to do the work that actually moves your business forward. Every expense is reviewed by an agent the second it's submitted, and only exceptions are routed when a human needs to see it.

Policy finally lives where spend lives on Airwallex.

Already on Airwallex? Go to Expenses → Settings → Expense policy agent to upload your policy to start enforcing your company's rules on every expense