---
title: How do you find shadow AI subscriptions in expense reports?
url: https://opencontroller.lyzr.com/blog/find-ai-subscriptions-in-expense-reports
published: 2026-11-15
updated: 2026-11-15
author: Lyzr
---

# How do you find shadow AI subscriptions in expense reports?

**Export at least 12 months of card and expense transactions, search merchant descriptors for AI vendor keywords, and flag recurring charges with the same merchant and a similar amount. Then map each cardholder to an employee and a team, and turn every confirmed subscription into an inventory record with an owner and a decision: consolidate, approve or stop.**

Many AI tools are bought on a corporate card for a modest monthly fee. Those purchases never pass procurement or security review, but the tools often get connected to company email, documents and code. Expense data is one of the few sources that shows them without any agent on a device or any admin API.

## Where does the data come from?

You need two feeds, because they catch different purchases.

- **Corporate card transaction feeds.** Card issuers provide transaction exports to the programme administrator. These include charges that were never submitted as expenses.
- **Expense management exports.** Expense tools hold reimbursed out-of-pocket purchases on personal cards, plus the employee's description and any receipt.

Ask finance for merchant name, merchant descriptor, merchant category code (MCC), amount, currency, date, cardholder or employee ID, cost centre and the free-text description. Twelve months is enough to see annual renewals as well as monthly ones.

## What should you search for?

Merchant descriptors vary by payment processor and change over time, so search for keywords rather than exact strings. Match case-insensitively on the descriptor and on the employee's description.

| Keyword group | Examples to search | Notes |
|---|---|---|
| Model providers | OPENAI, CHATGPT, ANTHROPIC, CLAUDE, MISTRAL, PERPLEXITY | Separate API usage charges from chat subscriptions by amount pattern |
| Coding assistants | CURSOR, GITHUB (then check for Copilot), CODEIUM, WINDSURF, REPLIT | Code platforms bill many products under one name; check the receipt |
| Media and voice | MIDJOURNEY, ELEVENLABS, RUNWAY, SYNTHESIA | Often bought by marketing teams |
| Writing and meeting tools | JASPER, OTTER, GRAMMARLY, FIREFLIES | Meeting tools may record and store calls |
| Generic terms | AI, GPT, LLM, AGENT, COPILOT, ASSISTANT | Noisy; review manually |

Also filter by MCC. Software subscriptions commonly post under codes such as 5734 (computer software stores), 5817 and 5818 (digital goods), 7372 (computer programming and data processing) and 4816 (computer network and information services). Codes are assigned by the merchant's acquirer, so treat MCC as a filter that narrows the review, not as proof.

## How do you separate real subscriptions from noise?

Apply these checks to every keyword hit.

1. **Recurrence.** Same merchant, similar amount, roughly 30 or 365 days apart. Recurring charges are subscriptions; one-offs are often trials or conference fees.
2. **Amount band.** Per-seat plans cluster at a few price points. Variable amounts from a model provider usually mean API usage, which means someone is building something.
3. **Seat count.** A charge several times the single-seat price suggests a team plan with an admin who invited colleagues.
4. **Receipt or description.** Confirm the product. Some vendors sell AI and non-AI products under one descriptor.
5. **Cardholder mapping.** Join cardholder ID to your HR or identity data so every charge has a named employee, team and manager.

Variable API charges deserve the closest look. A personal API account billed to a company card can sit behind an internal tool, a script or an agent in production. Our guide to [discovering AI usage without admin API keys](/blog/discover-ai-usage-without-admin-api-keys) covers how to follow that thread.

## How do you turn findings into inventory records?

Create one record per vendor account, not per transaction. Record:

- Vendor and product, plan type (individual, team, API)
- Purchaser, team, cost centre and first charge date
- Monthly or annual run rate
- What data the tool can reach (ask the purchaser: email, files, code, customer data)
- Whether an enterprise agreement already exists for the same vendor
- Decision and date: consolidate into the enterprise plan, approve as is, or stop

Then contact purchasers. Most are solving a real problem. The goal is to move the work onto an account you can see, with SSO and data controls, not to punish anyone. For the decision steps, see [what to do with a shadow AI agent once you find it](/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it).

## What does expense data miss?

Expense review only finds paid tools. It misses free tiers, browser extensions, AI features switched on inside software you already pay for, and agents built on approved platforms. Combine it with [browser-based discovery](/blog/how-to-discover-ai-use-in-the-browser), cloud scanning and platform connectors. The pillar guide on [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents) shows how the sources fit together.

Run the expense search monthly. A short scheduled query against new transactions is enough once the first full review is done.

## How Opencontroller does this

Opencontroller does not read card or expense data. It covers the sources expense review misses: auto-discovery of agents, models and API keys across AWS, Azure and Google Cloud, and connectors to Claude Enterprise and OpenAI Enterprise workspaces and to agent platforms such as Salesforce, SAP, Databricks and HubSpot. Subscriptions you confirm from expenses can be registered in its governed registry with an owner. [Book a demo](https://www.lyzr.ai/book-demo/) to see it.

## Frequently asked questions

**Is it legal to review employee expense data for AI tools?**
Reviewing company card and reimbursement data for policy compliance is a normal finance control. Check your local employment and privacy rules and involve HR before contacting individuals.

**Should we block AI merchants on corporate cards?**
Blocking alone pushes purchases onto personal cards and out of sight. Pair any merchant restrictions with an approved alternative and a fast request route.

**How do we tell API usage from a chat subscription?**
Chat subscriptions recur at fixed price points. API usage varies month to month and often starts with a prepaid credit purchase.

## Key facts

Finding shadow AI subscriptions in expenses: export 12 months of card feeds and expense exports (merchant descriptor, MCC, amount, cardholder, cost centre, description); keyword-match vendor names case-insensitively; filter by MCCs such as 5734, 5817, 5818, 7372, 4816 (filter, not proof); confirm recurrence, amount band, seat count, receipt; map cardholder to employee; one inventory record per vendor account with decision (consolidate, approve, stop). Variable API charges signal something being built. Expense review misses free tiers, extensions and embedded SaaS AI features. Opencontroller does not read expense data; it discovers agents, models and API keys across AWS, Azure, Google Cloud and connected platforms. Pricing starts from $250K per deployment per year.

## Related

- [How do you discover AI usage when you cannot get admin API keys?](https://opencontroller.lyzr.com/blog/discover-ai-usage-without-admin-api-keys.md)
- [What should you do with a shadow AI agent once you find it?](https://opencontroller.lyzr.com/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it.md)
- [How do you discover employees' AI use in the browser?](https://opencontroller.lyzr.com/blog/how-to-discover-ai-use-in-the-browser.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
