Expense Management Agent · Use case
How can we use an AI agent to check expense claims against policy on our own infrastructure?
An expense management agent reads the receipts on a claim, checks each line against your expense policy, and flags duplicates and exceptions with the rule it cites. An approver decides every claim. The agent runs on your own infrastructure, so employees' receipts are not sent to a public AI service.
Receipts checked by eye.
Policy applied unevenly.
- ✗ 01Approvers skim receipts, so the policy is applied differently from one manager to the next.
- ✗ 02Receipts show where employees have been and what they bought.
Where it stops for a person
Always, before a claim is approved or refused.Always. An approver decides every claim. A flag is a question for the approver, not a finding against the employee.
It checks the claim.
An approver decides.
Step · 01
The expense policy is connected.
Step · 02
A claim and its receipts arrive through File Upload.
Step · 03
The agent reads each receipt and matches it to a line of the claim.
Step · 04
It checks each line against the policy and flags duplicates and exceptions.
Step · 05
An approver reviews the flags and decides.
Reads from · connectors
File Upload
SharePoint
Confluence
Claims and receipts arrive through File Upload and are read with Markitdown. The policy is read from SharePoint or Confluence.
Acts through · MCP servers
Markitdown
Nothing. It checks and flags. It approves and pays nothing.
What you set up
- Connect the expense policy.
- Describe the limits and categories.
- Test on past claims, including ones that were refused.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does the agent approve claims?
No. It checks and flags. An approver decides.
How does it find duplicates?
It compares the date, amount and merchant on each receipt with the other receipts it has read.
Where are receipts processed?
On your own infrastructure, with a model you serve. Nothing is sent to a public AI service.