Dot grid
Answer
>
Hidden AI spend

Where AI spend hides: corporate cards, AI credits and software you already buy (2026)

The AI spend finance misses is either assembled by hand or buried inside contracts you already pay for. Here's how to find both. Last updated: November 2026.

Team Aleph
Shaping the future of AI-native FP&A
Share to
Table of contents
Subscribe to the 10X Finance Blog

Get FP&A best practices, research reports, and more delivered to your inbox.

Thanks — check your inbox.
Something went wrong. Please try again.

Hidden AI spend, the part that goes uncounted, is either assembled by hand or embedded in software you already buy. Hand-assembled costs are coding tools, agents, GPU hardware and the infrastructure under them; embedded costs arrive as AI bundled into seats, AI credits or per-action charges inside existing contracts.

To find all of it, search 12 months of general ledger, accounts payable and corporate-card data for direct AI vendors, then review your largest software renewals for new AI features, credits and tier upgrades, and count the increase as AI spend even when the vendor name hasn't changed. That second step is where most hidden AI spend turns up.

Bottom line: Your AI spend total is only as complete as the sources you searched. Platform-reported costs such as cloud usually get counted; coding tools, agents and AI inside existing software contracts usually don't, unless someone owns finding them.

Which AI costs are most likely to be missed?

The ones a person has to assemble. Research by Mavvrik with Benchmarkit, cited in our budgeting for AI spend guide, asked companies which categories they include in AI cost reporting. Costs a platform reports for you make it in far more often:

Cost categoryShare of companies including it in AI cost reportingHow it usually reaches finance
Public cloud67%Reported by the cloud platform
Hosted AI models53%Reported by the provider's billing console
Data platforms52%Reported by the warehouse or data platform
AI coding tools42%Assembled by hand from software, card and expense data
AI agents36%Assembled by hand from agent platforms, workflow logs and usage
GPU hardware36%Assembled by hand from capital and cloud spend

Source: Mavvrik with Benchmarkit, 2026, as cited in Aleph's budgeting for AI spend guide. Percentages are the share of companies that include each category in their AI cost reporting.

The pattern is the point. Coding tools, agents and hardware rarely arrive as a neat "AI" line item. If nobody owns the job of finding them and connecting the usage data, they don't make it into the total, and the number you give the board is too low.

Turn AI spend into a number you can defend. Download the guide: Budgeting for AI spend

How do you find hidden AI spend in the GL, AP and card data?

Start broad, then narrow. A first pass across 12 months of transactions catches most direct AI spend:

  1. Build a vendor list. List the AI vendors you know about, then add the categories you might not: model providers, chat assistants, coding assistants, meeting and note-taking tools, writing and image tools, and AI features sold as add-ons by existing vendors.
  2. Search vendor names and descriptions. Search AP vendor names, invoice line descriptions and card merchant names for the vendor list and for terms like "AI", "GPT", "tokens", "credits" and "copilot". Expect false positives and review them by hand.
  3. Pull every corporate card and expense feed. AI tools are often bought on a card by the person who wanted them, and expensed under software, subscriptions or office costs. A model provider's monthly invoice paid on a corporate card is often booked as a single vendor line with no detail behind it.
  4. Check cloud marketplaces. Some AI tools are bought through a cloud provider's marketplace and show up inside the cloud bill rather than as a separate vendor.
  5. Ask the people who use AI most. Run the list past IT, procurement and the heaviest AI teams. You're looking for missing tools and subscriptions, not trying to allocate every shared dollar.

Then write down what you counted, what you left out and why, and who owns the number. Our guide to getting spend data from AI platforms covers the usage exports for Claude, ChatGPT and Cursor, which you'll need to reconcile against what the ledger booked.

How does AI hide inside software you already buy?

It arrives as part of a contract you already have, so it never looks like an AI purchase. Software vendors are adding AI features, credits and usage charges to products you buy today, in three common ways. As of November 2026:

  • Bundled into the seat. Notion includes its AI features in its Business and Enterprise plans, so the subscription may carry no separate AI line item. The AI cost is real; it's inside a higher per-seat price.
  • Metered through credits. Figma gives each seat a monthly allowance of AI credits that depends on the plan and seat type, with extra credits sold on top. Track the included allowance and any additional usage charges.
  • Charged per action. Salesforce prices Agentforce through Flex Credits that each agent action consumes. The cost moves with how much the agents do, under your contract's rates and allowances.

The FinOps Foundation's guide to FinOps for SaaS describes the same shift: license contracts that now include add-ons billed on consumption, such as AI credits, and capacity add-ons of credits or tokens that expire if they aren't used. Vendors change these terms often, so confirm them in your own contracts rather than relying on any published summary.

How do you review renewals for embedded AI?

Start with your renewal calendar and work down from the largest contracts. For each renewal coming up in the next 12 months, compare the new terms with the last contract and ask four questions:

  1. Did the price per seat rise with an AI feature? If the increase pays for AI, count the increase as AI spend.
  2. Did the contract add an AI credit pool or allowance? Record the included credits, the overage rate and whether unused credits expire.
  3. Is any feature now charged per action or per use? Estimate expected volume, because that line will move with usage.
  4. Did a tier upgrade come with the renewal? Upgrades sold "for the AI features" are AI spend even if the vendor calls them a plan change.

Watch auto-renewal clauses, too. The FinOps Foundation warns that auto-renew terms often renew at the current purchase level whatever the actual use, sometimes with price increases nobody negotiated. Know each contract's termination notice window, often 30, 60 or 90 days, so you can review the AI terms before the contract renews on its own.

How do you find AI costs that nobody assembles?

Give each hand-assembled category an owner and a data source. These are the costs no single invoice reports:

  • Coding tools. Search engineering software subscriptions, card spend and expense reports. Coding assistants are often bought per developer by engineering managers.
  • Agents and workflows. Pull run counts and cost from agent platforms and workflow logs, plus the model usage they generate under each API key or workspace.
  • GPU hardware and supporting infrastructure. Ask engineering which cloud resources, data platforms and storage exist to serve AI workloads, then tag them so they show up next month without another search.

The six sources finance should collect, and where each one lives, are listed in our guide to allocating AI costs across departments. Once the costs are found, classify them by what caused each one, not by vendor: our guide to whether AI spend is COGS or OPEX covers how.

What should an AI spend inventory include?

One row per AI cost, with enough detail that someone else could check it. A spreadsheet works fine to start. Record these fields for each item:

  • Vendor and product, including the parent contract for embedded AI, such as the AI credit pool inside a design tool subscription.
  • Type: seat or subscription, metered usage, or embedded in another contract. The type tells you which levers apply later.
  • Annual cost, and for embedded AI, only the share or increase the AI drives.
  • How it's paid: invoice, corporate card or expense report, and the GL account it lands in.
  • Owner and cost center: the person who can answer questions about it and the team it belongs to.
  • Renewal date and notice window, so the renewal review happens before the contract renews.
  • Data source: the provider export, workflow log or contract that backs the number.

The inventory is also where the in-and-out decisions live. If you chose not to count a small embedded feature, write that down next to it, so next year's search starts from the same rules.

How do you stop AI spend from hiding again?

Change how AI gets bought and booked, so next year's search is shorter. Four habits do most of the work:

  1. Route new AI tools through procurement intake, even when they're cheap enough for a card, so each one gets an owner and a cost center on day one.
  2. Give each team its own API key or workspace with the model providers, so usage arrives already tagged.
  3. Add an AI question to every renewal checklist: did this contract add AI features, credits or per-use charges?
  4. Reconcile provider usage exports to the ledger every month, so a new line shows up within weeks rather than at budget season.

Worked example: from $1.2 million to $1.4 million

The budgeting for AI spend guide walks through a company that identified $1.2 million of AI spend directly: about half employee operating expense, 30% product development, 15% customer-delivery cost of revenue and 5% capitalized implementation work.

A deeper search then found about $200,000 more in shared infrastructure and AI costs embedded in existing software contracts. The total became about $1.4 million, which means roughly one dollar in seven had been missing from the first answer.

The board-ready version says both numbers: what was identified directly, and what the search for hidden spend added. That's more credible than a single total, because it shows the board how the number was built and where it could still move. The figures are illustrative, but the size of the gap is a useful reminder: the first total is rarely the full one.

How far should you dig?

As far as the decision needs, and no further. The guide sorts companies into three levels:

  • AI is a small line item. A reasonable total from GL, AP and card data, checked with IT and the heaviest users, is enough. Don't allocate every shared infrastructure dollar.
  • AI is a material part of the budget. Break the spend into its sources and trace each one back to where the cost lives.
  • AI is core to how you operate. Run the full hunt for hand-assembled and embedded costs, and repeat it every quarter.

The watch-out at every level is false precision. A directionally right total is acceptable; implying a granular understanding you don't have will damage trust with the board for little gain. Say what you counted and what you didn't.

Who should own the AI spend inventory?

Finance should own the total, with named owners for each source. Finance runs the GL, AP and card search and reconciles provider exports to the ledger. Procurement owns the renewal review, because it sees contracts before they're signed. Engineering owns coding tools, agents and AI infrastructure, because those costs are created by engineering decisions.

Agree a cadence. A full inventory once a year before budget season, a renewal check every month, and a quarterly refresh of hand-assembled costs is a sensible starting point. The inventory then feeds the plan: our guide to forecasting AI spend starts from one reconciled total, and our guide to budgeting software for FY2027 covers the tools that hold the plan together.

Common mistakes when finding AI spend

  • Searching only for AI vendors. Embedded AI sits under vendor names you already know, so a vendor search alone will miss it.
  • Booking usage bills as one vendor line. A metered AI invoice booked as a single line tells you what you paid, not which team, product or workflow drove it. Reconcile it to the provider's usage export.
  • Ignoring the renewal calendar. AI terms added at renewal are the easiest hidden spend to catch, and the easiest to miss.
  • Drifting into security. Unapproved AI tools are a security question too, but for budgeting the job is to count the spend and assign an owner.
  • Presenting one precise number. Show what you found directly and what the deeper search added.

Count your AI spend with Aleph

Aleph connects your general ledger, AI provider usage and HRIS to Excel, so AI spend refreshes on its own and every number traces back to a source, an owner and a date. Get the full method in our budgeting for AI spend guide, plan the year with our annual budgeting process guide, or book a demo to see it on your own data.

See your FP&A upside in 15 minutes   Discover how top finance teams eliminate manual work and unlock value with Aleph.  
Subscribe to the 10X Finance Blog

Get FP&A best practices, research reports, and more delivered to your inbox.

Thanks — check your inbox.
Something went wrong. Please try again.

Frequently asked questions

Any cost the company incurs to use, build or run AI. That includes seats and subscriptions, API and model usage, coding tools, specialist AI apps, the infrastructure that runs AI workloads, agents and workflows, and the AI share of software you already buy, such as AI credits or AI-driven price increases at renewal.

Yes, when they add cost. If a renewal raises the per-seat price to pay for AI features, adds an AI credit pool or introduces per-action charges, count the increase as AI spend even though the vendor name on the invoice hasn't changed. If AI comes free with no price change, note it but don't count it.

Pull 12 months of card and expense data and search merchant names and descriptions for known AI vendors and terms such as "AI", "GPT" and "credits". Then move recurring AI subscriptions onto invoices or a tagged card so each one carries an owner and a cost center going forward.

AI credits are a prepaid or included allowance of AI usage inside a software subscription. Vendors such as Figma give each seat a monthly credit allowance by plan, and sell extra credits on top. Track the included credits, the overage rate and whether unused credits expire.

Check renewals for new AI terms every month, refresh hand-assembled costs such as coding tools and agents every quarter, and run a full inventory once a year before budget season. Companies where AI is core to the product should refresh usage-driven costs monthly.

Discover Aleph today

A 30-minute demo on your numbers, not ours.
Screenshot of an income statement spreadsheet comparing revenue, cost of revenue, and operating expenses for Jan 25 and Feb 25, alongside a sidebar menu with options including 'Income Statement,' 'Analyze with AI,' and other budget categories.
Dotted grid