	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Expense Audit Agent

Expense Audit Agent

Author: Venkata Sudhakar

ShopMax India's sales and operations teams submit hundreds of expense claims each month - travel, meals, accommodation, and equipment. Most are legitimate, but some exceed policy limits, lack receipts, or get submitted outside the allowed time window. A manual review process is slow and inconsistent. An automated expense audit agent that checks every claim against the expense policy and flags violations for human review makes the process faster, fairer, and harder to game.

This tutorial builds an Expense Audit Agent using ADK and Gemini. The agent checks each expense claim against policy limits, receipt requirements, and submission windows, classifies each as approved, flagged, or rejected, and produces a clean audit report for the finance team.

The below example shows the expense audit workflow for ShopMax India employee claims.

from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService

# --- Expense policy limits (Rs) ---
POLICY = {
    "travel_daily": 3000,
    "meal_per_day": 1000,
    "accommodation_per_night": 5000,
    "equipment_single_item": 10000,
    "max_days_to_submit": 30,
    "receipt_required_above": 500,
}

# --- Expense claims for review ---
CLAIMS = [
    {
        "claim_id": "EXP-001",
        "employee": "Deepak Sinha",
        "city": "Mumbai",
        "category": "travel",
        "amount": 2800,
        "has_receipt": True,
        "days_since_expense": 12,
        "description": "Cab to client site and back",
    },
    {
        "claim_id": "EXP-002",
        "employee": "Nisha Kapoor",
        "city": "Delhi",
        "category": "meal",
        "amount": 1450,
        "has_receipt": True,
        "days_since_expense": 8,
        "description": "Team lunch with 3 clients",
    },
    {
        "claim_id": "EXP-003",
        "employee": "Ajay Menon",
        "city": "Bangalore",
        "category": "equipment",
        "amount": 13500,
        "has_receipt": False,
        "days_since_expense": 45,
        "description": "External monitor for home office",
    },
    {
        "claim_id": "EXP-004",
        "employee": "Priya Singh",
        "city": "Hyderabad",
        "category": "accommodation",
        "amount": 4800,
        "has_receipt": True,
        "days_since_expense": 5,
        "description": "Hotel stay - client visit Hyderabad",
    },
]

# --- Tool: audit expense claims ---
def audit_expense_claims() -> dict:
    """Audit all pending expense claims against company policy.
    Returns:
        Audit results with violations and recommendations per claim
    """
    results = []
    for claim in CLAIMS:
        violations = []
        # Check amount vs category limit
        limit_key = {
            "travel": "travel_daily",
            "meal": "meal_per_day",
            "accommodation": "accommodation_per_night",
            "equipment": "equipment_single_item",
        }.get(claim["category"])
        if limit_key:
            limit = POLICY[limit_key]
            if claim["amount"] > limit:
                over = claim["amount"] - limit
                violations.append(f"Amount Rs {claim['amount']} exceeds {claim['category']} limit Rs {limit} by Rs {over}")
        # Check receipt
        if claim["amount"] > POLICY["receipt_required_above"] and not claim["has_receipt"]:
            violations.append(f"Receipt required for claims above Rs {POLICY['receipt_required_above']} but not submitted")
        # Check submission window
        if claim["days_since_expense"] > POLICY["max_days_to_submit"]:
            violations.append(f"Submitted {claim['days_since_expense']} days after expense (max {POLICY['max_days_to_submit']} days)")
        # Determine status
        if len(violations) == 0:
            status = "APPROVED"
        elif len(violations) == 1 and "exceeds" in violations[0]:
            status = "FLAGGED"
        else:
            status = "REJECTED"
        results.append({
            "claim_id": claim["claim_id"],
            "employee": claim["employee"],
            "category": claim["category"],
            "amount": claim["amount"],
            "status": status,
            "violations": violations,
        })
    approved_total = sum(r["amount"] for r in results if r["status"] == "APPROVED")
    return {"total_claims": len(results), "approved_total": approved_total, "results": results}

# --- Agent setup ---
audit_tool = FunctionTool(func=audit_expense_claims)

agent = LlmAgent(
    name="expense_audit_agent",
    model="gemini-2.0-flash",
    instruction="""You are an expense audit agent for ShopMax India.
Call audit_expense_claims to audit all pending claims.
Present a table: Claim ID | Employee | Category | Amount | Status | Issues.
Then provide:
  - Total approved amount (ready to pay)
  - FLAGGED claims: recommend manager approval with specific reason
  - REJECTED claims: explain why and what the employee needs to resubmit
Keep the tone professional and factual. Format Rs with commas.""",
    tools=[audit_tool],
)

session_service = InMemorySessionService()
runner = Runner(agent=agent, app_name="expense_audit", session_service=session_service)

import asyncio

async def run_audit(query: str) -> str:
    session = await session_service.create_session(app_name="expense_audit", user_id="finance")
    from google.genai.types import Content, Part
    msg = Content(role="user", parts=[Part(text=query)])
    result = ""
    async for event in runner.run_async(user_id="finance", session_id=session.id, new_message=msg):
        if event.is_final_response() and event.content:
            result = event.content.parts[0].text
    return result

response = asyncio.run(run_audit("Audit all pending expense claims and tell me what can be paid, what needs review, and what must be rejected."))
print(response)

It gives the following output,

Expense Audit Report - ShopMax India

Claim  | Employee      | Category      | Amount   | Status   | Issues
-------|---------------|---------------|----------|----------|--------
EXP-001| Deepak Sinha  | Travel        | Rs 2,800 | APPROVED | None
EXP-002| Nisha Kapoor  | Meal          | Rs 1,450 | FLAGGED  | Exceeds daily limit by Rs 450
EXP-003| Ajay Menon    | Equipment     | Rs 13,500| REJECTED | Exceeds limit by Rs 3,500 + no receipt + late submission (45 days)
EXP-004| Priya Singh   | Accommodation | Rs 4,800 | APPROVED | None

APPROVED - Ready to pay: Rs 7,600 (EXP-001 + EXP-004)

FLAGGED - Manager review required:
  EXP-002 (Nisha Kapoor): Meal Rs 1,450 exceeds Rs 1,000 daily limit by Rs 450.
  If this was a client entertainment claim, manager may approve the overage.

REJECTED - Employee must resubmit:
  EXP-003 (Ajay Menon): Three violations - amount over limit, no receipt,
  and submitted 45 days after expense (deadline is 30 days). This claim
  cannot be processed. Employee should seek pre-approval for equipment
  above Rs 10,000 before purchase.

Connect this agent to your expense management system via a Cloud Function that triggers when new claims are submitted. Integrate with your HR system to validate that the employee is active and the cost centre is valid. Store audit logs in BigQuery for quarterly compliance reporting and pattern analysis to identify employees with repeat policy violations.

Send your comments, suggestions or queries regarding this site to [email protected].