Expensed SaaS: The Software Budget Hiding in Expense Reports

Somewhere in last month's expense reports there's a $20 ChatGPT subscription and a $15 Canva Pro plan. Probably a few more like them, too. Each one got approved in about four seconds, because nobody is going to hold up a reimbursement over $20. Add up every one of those across a year and across every team, and you'll find a software budget that never went through IT or procurement.

That's expensed SaaS. It's one of the easiest kinds of shadow spend to miss, because each line is small and it's spread across hundreds of people. It's also one of the easiest to fix, once you know how to pull it out of your data.

This guide walks through exactly that, step by step. You'll learn how to find the software in your T&E data and clean up the vendor names so you can see what you're really paying for. Then we'll cover how to move the renewals you want to keep onto a path you actually control.

What is expensed SaaS?

Expensed SaaS is any software subscription an employee pays for personally, or on a corporate card, and then claims back through an expense report. It skips the usual buying process entirely. Nobody raises a request and nobody reviews the vendor. Finance sees it only as a line in a T&E report, often coded as "office supplies" or just "other".

In dollar terms it looks tiny. Zylo's 2026 SaaS Management Index found expensed spend makes up only 3.7% of total SaaS spend. The catch is that it accounts for 45% of applications. So nearly half the tools running in a typical company came in through an expense report, and Zylo puts the average cost of those expensed apps at nearly $2 million a year per company.

It's growing fast, too. The same index found that expense-based SaaS spend jumped 267% year over year, and ChatGPT is now the single most expensed app. AI is the big driver here. Eight of the top 50 most expensed apps are AI-native, and the MIT research on the "shadow AI economy" found 90% of employees use personal AI tools for work, while only 40% of companies have bought official subscriptions.

Here are the top ten most expensed apps from Zylo's latest list, by transaction volume:

Some of those are tools you'd happily pay for centrally. Others, like Spotify and iCloud, probably shouldn't be on the company's tab at all. Either way, you can't make that call until you can see them.

Why expensed SaaS is a problem for finance

One $20 subscription is fine. Trouble starts when it multiplies across a few hundred people.

You pay for the same tool several times over. When five people in five teams each expense their own Canva plan, you're paying retail for five seats. A single team plan would almost certainly cost less, and you'd get admin controls with it.

Renewals happen on autopilot. Personal subscriptions renew every month or every year, on the employee's card, with no one checking whether the tool is still needed. When the employee leaves, the subscription often keeps going, and so do the reimbursement claims if the card is corporate.

Your budget is wrong. Software expensed through T&E shows up in travel and expense lines, not in your software budget. So the IT budget looks under control while real software spend is higher than anyone has reported.

Your data is exposed. Zylo found that 59% of expensed apps have poor or low security ratings. And IBM's 2025 Cost of a Data Breach research found that one in five organisations reported a breach due to shadow AI, with high shadow AI use adding $670,000 to the average breach cost. A personal ChatGPT account with customer data pasted into it is precisely the kind of thing that ends up in those numbers.

Nobody owns the contract. A personal subscription is agreed to by an individual clicking "I accept". The company has no negotiated terms and no say over how its data is used.

Step 1: Pull the software out of your T&E data

Start with a full export of expense lines and corporate card transactions for the last 12 months. You want the merchant name, amount, date, employee, department and the expense category for every line. Most expense tools can give you this as a CSV in a few clicks.

Then filter it down to software. There are a few ways to catch it, and it's worth using all of them together because each one misses something.

Filter by merchant category code (MCC). Every card transaction carries a four-digit code that describes the type of merchant. The ones that usually point to software are listed below, with descriptions from Citi's merchant category code reference:

Search merchant names for known software brands. Look for names like OpenAI, Anthropic, Canva, Notion, Zoom, Figma, Adobe, Atlassian, Slack, Dropbox and Grammarly. Also search for patterns software vendors use on statements, like "*.com", "SaaS", "subscription", "monthly" and "annual".

Look at the categories people pick. Pull every line coded to software, subscriptions, office supplies, miscellaneous or other. That's where most software ends up when there's no obvious category for it.

Look for the recurring pattern. Software is the only thing most employees expense for the exact same amount every month. Group by employee and merchant, and flag anything that shows up three or more months in a row at a similar amount. This catches the tools your keyword list misses.

At the end of this step you'll have a list of every likely software charge, which is usually longer than anyone expects.

Step 2: Normalise the vendor names

Card statements are messy. The same tool can show up under a handful of different merchant strings depending on the card network and the billing entity. Until you clean those up, you can't see how many people are paying for the same thing.

Here's what that looks like in practice:

Example strings for illustration. Real statement descriptors vary by card issuer and region.

A few rules make this much quicker:

1. 1.Strip the noise first. Remove reference numbers, asterisks, payment processor prefixes like "SQ *" or "PAYPAL *", and suffixes like "INC" or "LLC".
2. 2.Build a mapping table. Keep a simple two-column sheet of raw merchant string to clean vendor name. Every time you clean a new one, add it. After the first pass, most new lines will match automatically.
3. 3.Split vendors that sell more than one product. OpenAI, Atlassian, Adobe and Google all sell several products. Use the amount and the employee's role to work out which one it is.
4. 4.Tag each clean vendor with a category. AI assistant, design, project management, storage, communication and so on. This is what lets you spot overlap in Step 3.

Once the names are clean, pivot the data by vendor. For each tool, you want to see how many people are paying for it and what it adds up to over a year.

Step 3: Decide what to do with each tool

Now you can see the full picture. Sort the vendor list by total annual spend and go through it with IT and the department heads. Every tool falls into one of four buckets.

AI tools deserve their own look. If 40 people are each expensing a personal ChatGPT or Claude plan, that's a clear signal they need an AI assistant for their work. A business plan gives you data controls and admin visibility that personal plans don't, and it usually works out cheaper per seat at that scale.

Personal apps are the easy wins. Things like Spotify or personal iCloud storage turn up on Zylo's most-expensed list, and most companies don't intend to pay for them. A clear line in your expense policy fixes that quickly.

Step 4: Route renewals into a sanctioned path

Cleaning up once is useful. Keeping it clean is the part that saves money year after year. That means giving people a better way to get software than putting it on their card.

1. 1.Update your expense policy. Make software non-reimbursable unless it's been approved first. Say it plainly in the policy and in the expense tool itself, so people see it at the moment they submit. Give a short grace period so nobody gets caught out.
2. 2.Give people an easy way to ask. If the only alternative to expensing is a long procurement form, people will keep expensing. Let them ask in the places they already work, like Slack or Teams, and give them an answer quickly.
3. 3.Build a renewal calendar. For every tool you centralised, record the renewal date and who owns the decision. Set a reminder 60 to 90 days before each renewal so someone decides whether to keep it while there's still time to negotiate or cancel.
4. 4.Move renewals onto a PO or a controlled card. Centralised tools should be billed to the company on one invoice against a PO, or charged to a virtual card locked to that one merchant. Either way, they leave the expense system for good.
5. 5.Re-run the search every month. New tools will keep appearing. A monthly check on software MCCs and your mapping table catches them within weeks, while they're still one or two people.

The goal is for the approved path to be faster than the workaround. Once it is, most people stop looking for one.

How Blackbee AI handles expensed SaaS

The four steps above work, and you can run them in a spreadsheet. The hard part is doing it every month, forever, while new tools keep turning up. Blackbee AI is an agentic Intake-to-Pay platform that sits above your ERP and does this continuously.

It finds the software for you. The Spend Intelligence Agent watches card and expense data alongside your invoices. It recognises software merchants and groups the messy statement names into one vendor. Then it flags when several people are paying for the same tool or when a new AI app starts showing up.

It connects the data that usually sits apart. Expense tools and card feeds rarely talk to your ERP. The Integration Agent pulls them together so software spend is visible in one place, and posts clean, validated decisions back to your ERP.

It makes the approved path the easy one. The Intake Agent picks up software requests straight from Slack, Teams, email or a form. Someone asks for a tool the way they normally would, and the request gets checked against what you already own and your policy before anyone pays for anything.

It keeps an eye on renewals. The Contract Intelligence Agent reads your software agreements and surfaces renewals before the notice window closes, so a tool doesn't quietly roll over for another year.

Of course, all of this is within your explicit control. You choose which steps Blackbee AI handles, and it does only those, and does them well.

Bring your software spend into the light

Expensed SaaS is small line by line and big in total. The good news is that it leaves a clear trail in your T&E data, so you can find most of it in an afternoon. Keeping it under control is about giving people a faster, approved way to get the tools they need.

If you'd like to see how Blackbee AI spots expensed software and routes new requests through an approved path, book a 30-minute demo.