How to Track SaaS and AI Spend in Real Time

Software spending used to be relatively predictable. Businesses paid for a set number of licences, knew their monthly costs, and could plan their budgets accordingly. But AI tools and usage-based pricing are changing that. Costs can rise unexpectedly as employees use more tokens, run more automated workflows or adopt new AI tools, often before finance teams even see an invoice.

Spendesk’s latest research shows just how unpredictable these costs can be. AI spending among European businesses grew 340% over 12 months, while companies with sustained AI spending saw their monthly bills fluctuate by approximately 61%. In some cases, the highest monthly bill was around eight times the lowest. Traditional monthly reporting simply isn’t enough to manage that level of variation.

To stay in control, finance teams need visibility into what they’ve already spent, what they’re committed to spending, and how much budget remains. The most reliable way to get it is to connect data sources in a sensible order, starting with those that show money spent and committed, then adding those that show usage and hidden tools.

Finance teams typically start with corporate and virtual card transactions and the finance, enterprise resource planning (ERP), or accounting feed. Card data can often arrive through webhooks or event streams as activity happens, while accounting feeds commonly sync nightly and invoices are usually reconciled monthly. Next, single sign-on (SSO) and identity provider (IdP) logs show which apps people use, and these commonly sync nightly. Direct vendor API connectors then add consumption data for usage-based AI tools, often in near real time depending on the tools. Browser or network discovery agents come last, catching tools that never pass through procurement. Together, these sources feed a single live dashboard with alerts on spend thresholds, newly detected apps, and upcoming renewals. The rest of this guide covers spotting unexpected costs early, setting controls, and forecasting variable costs.

Key takeaways

• Usage-based AI pricing makes software costs behave more like a metered service than a fixed subscription.

• Month-end reports show what has already happened, so volatile AI costs need earlier signals.

• Tracking budget, actual spend, and committed spend together shows how much budget is genuinely left.

• Alerts that escalate as spending approaches a limit give budget owners time to act before overspend.

• Forecasting AI costs as a range of scenarios, and reforecasting often, keeps budgets closer to reality.

Why finance teams manage AI spend differently from traditional SaaS spend

Traditional SaaS often follows a seat-based pricing model. If your company buys 100 licences at €20 per employee each month, finance can reasonably expect a €2,000 bill. There may be changes, but the cost is relatively stable.

Finance teams see a different cost pattern with usage-based AI, where costs may change according to:

• number of prompts;

• API calls;

• automated workflows;

• agents;

• tokens consumed;

• models selected;

• data processed;

• or other measures of usage.

A team can stay within its licence count while its consumption and eventual bill rise, so finance needs to treat AI spending more like a metered service than a conventional subscription.

Fixed subscriptions versus usage-based billing

Finance teams need to distinguish between two types of software cost.

Fixed or relatively predictable SaaS

Examples include products where the main cost is based on:

• number of users;

• plan tier;

• contract value;

• or a stable recurring subscription.

These costs can usually be forecast with reasonable confidence.

Predictable isn’t the same as efficient. Comparing the seats paid for with active users in SSO logs shows unused licences that can be reclaimed, and it exposes duplicate tools doing the same job, ideally before the next renewal.

Variable or usage-based software

Here the bill changes depending on consumption. Your budget might say €5,000, but actual usage could lead to €3,000 one month and €12,000 another.

Median monthly variation in AI spending sat within a relatively narrow 55% to 65% range across European markets, according to Spendesk’s research. That suggests volatility is largely a feature of the billing model rather than one country’s buying behaviour, which makes traditional annual budgeting less effective as the only control.

Why monthly finance reporting is too slow for volatile AI costs

Month-end reporting tells finance what has already happened. For stable expenses, that’s often sufficient, but for volatile spending, it may be too late.

Imagine an AI service normally costs €2,000 per month. During the first week of the month, a team launches an automated process that dramatically increases consumption. If finance only discovers the increase when the invoice arrives or during month-end close, the cost has already happened.

Real-time visibility changes the timing. Instead of asking “Why was the bill so high last month?”, finance can ask “Why is spend increasing this week?” That creates an opportunity to investigate before unusual spending becomes the new run rate.

Why shadow IT and shadow AI hide software spend

Missing visibility usually starts with SaaS sprawl, because tools and AI assistants can be bought in ways that never reach procurement. Employees might sign up on a company card, through an expense claim, or via a free trial that converts to paid. It’s a common pattern, not a failing of employees or finance, and it means the cost often arrives late or as an unexplained charge.

Catching these purchases means reviewing card and expense-claim data for software merchants, watching SSO logs for newly used apps, and using discovery agents for tools outside SSO. The aim isn’t to ban new tools. It’s to route them into a normal purchase request, so useful ones get an owner and a budget instead of becoming maverick spend.

Keep one central SaaS inventory

A central inventory lists each tool with its owner, cost, contract term, renewal date, and seat count. Spreadsheets often serve well while the stack is small. As tools multiply, a manually updated sheet drifts from the card, SSO, and accounting data, so the inventory works best when it refreshes from those feeds.

What finance teams should monitor in real time

Real-time visibility doesn’t mean watching a dashboard all day; it means the right information is available before the accounting period closes.

At minimum, finance should understand:

• actual spend to date;

• approved or committed spend;

• budget remaining;

• major vendor movements;

• recurring payments;

• department-level spend;

• unusual transactions;

• and spend approaching predefined limits.

For variable software and AI, finance should also look for sudden changes in spending velocity. A team that normally spends €500 a week with a vendor and suddenly spends €2,000 is a more useful early-warning signal than a variance discovered several weeks later.

Budget versus actual spend

One of the simplest controls is a live comparison between budget and actual spend. Budget is what the business intended to spend; actual is what has already been spent.

If a department has a €20,000 quarterly technology budget and has already consumed €18,000 halfway through the quarter, the information becomes actionable. Finance and the department can investigate:

• whether the additional spend is expected;

• whether another cost has increased;

• whether the budget needs to change;

• or whether spending should slow.

Finance should understand every budget variance while there’s still time to act, without assuming each one is bad.

Committed spend before invoices arrive

Committed costs complete the picture alongside actual spend, because a company may already be committed to costs that haven’t yet appeared in accounting data. Examples include:

• approved purchase requests;

• signed SaaS contracts;

• annual subscriptions;

• recurring card payments;

• purchase orders;

• and expected invoices.

A useful spend view therefore needs to distinguish between actual spend, which is money already spent, and committed spend, which is money the company has agreed or expects to spend and which determines the available budget once those commitments are considered. Without committed-spend visibility, a department can appear comfortably below budget even when most of its remaining budget has effectively already been allocated.

Renewals, especially auto-renewals, are commitments waiting to happen, so it makes sense to track them before they land in the budget. A renewal calendar records each contract’s renewal date, notice period, and owner, which turns a surprise charge into a scheduled decision. With that in place, the budget owner has time to run a quick seat-usage check and decide whether to renew, renegotiate, or cancel before an auto-renewal locks in another term.

Spend thresholds and escalation alerts

A budget is more useful when something happens as spending approaches it, so finance can set different responses at different thresholds. For example:

• 50% of budget: informational

• 75%: notify the budget owner

• 90%: finance reviews remaining commitments

• 100%: additional spend requires approval

The right thresholds will vary by business, and the principle is more important than the exact percentages: don’t wait for overspend before communicating that overspend is approaching.

This matters especially for volatile technology spending, and Spendesk’s research argues that usage-based costs need an additional control layer, such as spending limits and real-time alerts, rather than relying solely on a fixed budget.

Alerts flag spend once it’s moving, while preventive controls can act before money moves. An approval workflow on purchase requests for new software, plus a procurement gate before a new vendor is added, gives each tool an owner and a business purpose from day one. Issuing subscription virtual cards per vendor, each with its own limit, can keep every recurring charge visible, capped, and easy to cancel. Where configured, a card that reaches its limit can stop further charges without anyone chasing the vendor.

Department and project-level accountability

A company-wide AI bill isn’t enough to manage AI spending, because finance needs context. Where possible, assign spending to the following:

• department;

• team;

• project;

• cost centre;

• budget owner;

• and business purpose.

This changes the conversation. Instead of “Our AI spend increased 40%.”, you can understand “AI spend increased because engineering launched a new automation project.” Those are very different situations, and the second may be entirely justified.

Good spend visibility means understanding why the money is being spent, and that goes beyond identifying waste.

How to forecast variable AI costs

You may not be able to predict a usage-based AI bill precisely, but you can make it more manageable.

Establish a baseline

Measure typical weekly or monthly spending.

Track the range

Don’t forecast only from the average, and understand your lower and higher spending months.

Identify usage drivers

Ask what changes the bill, whether it’s:

• employee adoption?

• API consumption?

• agents?

• automation?

• specific projects?

Separate recurring and experimental spend

An established AI workflow and a three-month pilot shouldn’t necessarily share the same forecast assumptions.

Build scenarios

Instead of a single number, model expected usage alongside low and high usage.

Reforecast frequently

For a fast-growing category, a forecast created six months ago may no longer reflect reality.

A practical real-time spend visibility framework

Finance teams can use a simple five-layer model.

Layer 1: Visibility

Can we see software and AI payments consistently across cards and invoices?

Layer 2: Ownership

Does every significant vendor have a department and cost owner?

Layer 3: Budget

How much did we plan to spend?

Layer 4: Commitments

How much have we already spent or committed?

Layer 5: Intervention

What happens if spending moves outside expectations?

That final layer matters because a dashboard that tells you costs went over budget isn’t a control; it’s a record. Finance teams need a process that defines who investigates and what action follows, including any alerts.

Real-time visibility without real-time micromanagement

Finance can avoid becoming the department that approves every €20 purchase. The goal is to create guardrails around spending while still allowing teams to operate. The European Spend Report 2026 shows why those guardrails are becoming more important. AI bills can move dramatically from month to month, and rapidly increasing adoption means finance teams can no longer assume software behaves like a predictable fixed subscription.

“How do we stop AI costs changing?” is a natural question, but finance can’t control how third-party vendors price their products. A question finance can act on is “How quickly can we see that our spending has changed, understand why, and decide what to do?” That’s the foundation of real-time spend control.

Explore budget management for greater visibility over budget versus spend, or spend controls and approval workflows to put guardrails around company purchases.

Frequently asked questions about tracking SaaS and AI spend in real time

Which data source gives the most complete picture of SaaS spend?

No single source is complete on its own. Card feeds show payments close to when they happen, but they miss invoiced contracts and say nothing about usage. Accounting exports capture invoices and commitments, but they typically arrive later. Single sign-on (SSO) logs show who uses which app, but not what it costs, and they miss tools outside SSO. Combining all three, plus vendor usage data for AI tools, closes most of the gaps.

Does real-time spend visibility replace the month-end close?

No, the month-end close still produces the accounting record. What earlier visibility changes is the quality of that close, because fewer surprises reach it and most variances are already understood by the time you sit down to reconcile.

Should usage-based AI tools have a separate budget from fixed SaaS subscriptions?

Often, yes. Tracking them separately keeps the stable, forecastable part of your software budget from being distorted by variable consumption, and it makes each category’s owner and assumptions clearer. The right split depends on how much of your bill is usage-based.

How often should finance review usage-based AI spend?

Match the frequency to how volatile the vendor’s bill is. Weekly checks may suit fast-moving, usage-based tools, while monthly review is usually enough for stable subscriptions. Treat that as a sensible starting point, not a rule, and adjust as you learn how each tool behaves.