SaaS Spend & Procurement Statistics 2026

SaaS statistics become misleading when the denominator disappears. An annual portfolio average is not a market forecast. A license that has no observed activity is not the same as an application used below purchased capacity. A list-price discount is not the same as savings against a prior contract, and a 93-day software buying cycle does not contradict a two-day requisition-to-purchase-order benchmark when the clocks start and stop at different events.

For CFOs, procurement leaders, FP&A, finance operations and IT-finance teams, the useful question is not which external number looks closest to the budget. It is which measure can test an internal decision. That requires a reconciled spend perimeter, contract and renewal evidence, license and consumption data, and a clear separation between committed value, invoiced spend, actual usage and forecast exposure.

The 2026 SaaS spend and procurement benchmark table

The table below keeps each population, measure and period attached. Values from SaaS-management and procurement vendors describe their customer or connected cohorts. Survey values describe respondents. Gartner values are forecasts, not observed company budgets.

First fix the denominator: seven distinctions that change the answer

This boundary matters at market scale. Gartner’s July forecast puts 2026 worldwide IT spend at $6.369 trillion and software at $1.468 trillion, with forecast growth of 14.2% and 15.5% respectively. Those modeled vendor-sales totals are not SaaS-only and cannot tell one company what its subscription budget should be.

SaaS spend is large, skewed and not a universal budget target

Zylo’s mean and median are the clearest warning against a single “average company” figure. Its average annual SaaS spend is $55.7 million; its median is $20.6 million. Our calculation puts the gap at $35.1 million. The same index reports an average of 305 applications and a median of 240. Large portfolios pull the mean upward, and the customer base is not a random sample of every company.

Per-employee numbers also need their statistic labels. Vertice reports an average of $9,324 per employee in Q2 2026. Zylo reports a median of $9,455. The values look similar, but mean and median are not interchangeable, and the cohorts and spend-discovery methods differ. Finance should calculate its own annualized software cost per active employee, then segment it by role, business unit and application category before comparing.

The internal perimeter should start with more than the contract repository. AP, corporate cards, employee expenses, procurement records, identity systems and vendor invoices can each reveal different parts of the population. The spend-data readiness tests explain how to bound and reconcile those sources before a category total or savings claim is trusted.

License utilization and waste: build a review pool before a savings case

Zylo reports that only 54% of licenses were active in the measured period and assigns $19.8 million in annual cost to unused licenses in its cohort. Vertice uses a different split: 14% fully unused and 51% used below 50% of purchased capacity. The combined 65% is a license classification, not a statement that 65% of spend can be removed.

A finance-grade review needs four stages. First, confirm the entitlement and activity window. Second, classify zero activity, intermittent activity, low-capacity use, wrong tier, duplicate function and legitimate reserve capacity separately. Third, check contract rights and service dependencies. Fourth, record realized savings only when a renewal, cancellation, downgrade, credit or redeployment changes the economic outcome.

This is why the repeated claim that “30% of SaaS spend is wasted” should not be used as a current universal benchmark without recoverable provenance, population and definition. Even a verified zero-use seat can be non-cancellable during the term. Conversely, a heavily used product can still be economically poor if the price, tier or outcome is wrong.

Decentralized purchasing makes the control tail larger than the spend share

Zylo’s survey and portfolio release say business units control 81% of SaaS spend while IT directly manages 15%. It also says the expense-based share of SaaS spend increased 267% year over year. That is relative growth in a spend channel, not a 267-percentage-point increase and not proof that every expensed purchase is unauthorized.

Decentralization changes the discovery design. A contract repository misses card and expense purchases. An AP ledger can miss free-to-paid conversions and embedded AI features. An identity feed can find access but not contractual commitment. The finance technology model therefore needs stable vendor, application, owner, cost-center, contract, user and consumption identifiers across systems. The finance technology reference architecture provides the system-role context for that record design.

The operating rule is to make the request path easier than bypassing it while keeping a low-friction route for small, low-risk purchases. Every exception still needs an owner, an evidence trail and a point at which security, privacy, commercial and budget checks become mandatory.

Renewals and contracts turn calendar discipline into a financial control

An average of 211 renewals a year is roughly four renewal events per week. That volume makes a single annual clean-up unrealistic. Zylo also reports that 79% of surveyed IT leaders encountered a price increase at renewal. Incidence does not reveal the size of the opening ask, the final negotiated increase or whether scope changed, but it does show why late review is expensive.

Vertice reports a 72.3% average across software categories for contracts containing auto-renewal clauses in Q2 2026. A separate Vertice measure says 72% of tail-spend renewals proceeded without prior review. These are different denominators: one measures clause prevalence across categories, while the other measures review behavior in a defined tail population.

For each contract, track the cancellation deadline, renewal date, price-notice date, committed quantity, true-up or true-down rights, usage window, owner and review tier. Start material reviews at 120 or 90 days, not when the invoice arrives. The source-to-pay control model shows where intake, contracting, purchasing and downstream evidence should change hands.

Pricing and negotiation: list discount is not realized savings

Vertice’s price series reached 16.4% in June after 12.1% in April and 14.2% in May. Zylo reports an 8% year-over-year increase in average annual SaaS spend. Subtracting one rate from the other would be invalid: the first is a proprietary vendor-price measure, while the second is total portfolio spend growth affected by price, seats, mix, usage, discovery and customer composition.

Vertice also reports an average 33.8% discount from list price, with category averages ranging from 19% for design software to 44% for ERP. This is the distance between list and negotiated price in its data. It is not necessarily savings against the current contract, a comparable market quote or the budget. The ERP cost framework shows why licenses, implementation, integrations, support and internal change cost must stay separate.

Tropic’s customer-renewal data gives a more specific AI-pricing example. Initial AI-related renewal asks were 20% to 37%, compared with historical uplifts of 3% to 9%. Across deals with flexibility, Tropic reports roughly a 55% reduction in those asks and a final average uplift near 12%. A buyer still needs to test whether the final deal added users, products, features, consumption rights or longer commitment.

A defensible renewal bridge starts with the prior recurring run rate, then isolates list or unit-rate change, seat and tier movement, usage or overage, added and removed modules, term, credits, taxes and currency. Procurement can then label avoided cost, price reduction, cost reduction and budget variance separately.

AI software spending needs four cost views, not one “AI budget”

Zylo reports 108% year-over-year growth in AI-native application spend and says 78% of surveyed IT leaders encountered unexpected AI or consumption charges. Ramp’s connected token cohort shows why fixed subscriptions and metered use must be separated: the April 2026 monthly median was $2,246, the average was $140,842, the 90th percentile was $73,030 and the 99th percentile was $831,338.

Ramp also reports a $46 median monthly token cost per employee, with the middle 50% ranging from $3 to $352. Premium models represented 45.8% of tokens but 55.9% of cost. These figures describe businesses connected to Ramp’s Token Spend Management product, not every corporate AI user, yet they show the forecasting risk created by model choice, automated workflows and heavy-tail usage.

KPMG’s Q2 2026 survey adds the control view: 42% of 2,145 senior leaders reported only partial visibility into AI spend, 23% struggled with usage-based costs and 33% struggled with cost structures or token economics. The FinOps Foundation’s 2026 practitioner survey, based on 1,192 respondents representing more than $83 billion in annual cloud spend, says 98% now manage some AI spend and 90% manage SaaS costs or plan to within a year.

Finance should maintain four linked views: seat subscriptions, embedded AI add-ons, token or API consumption, and the cloud or data infrastructure that supports AI workloads. Each needs a budget owner, unit rate, commitment, allocation key, alert threshold and outcome measure. The finance systems integration map is the next step when invoice, usage and ownership evidence sit in different applications.

Procurement cycle benchmarks require matching start and end events

Vertice’s Q2 2026 averages rise with contract value: 54 days under $10,000, 73 days from $10,000 to $50,000, 81 days from $50,000 to $100,000 and 93 days above $100,000. Its endpoint is the employee request through the signed software deal. The public data comes from Vertice’s processed-spend cohort and does not disclose a deal count for each band.

APQC’s median is two calendar days from purchase requisition received to purchase order released, based on 1,250 all-company responses. That clock starts later and ends at a different control point. Neither source invalidates the other. Combining them would hide intake, sourcing, security, legal, contract and approval stages inside one number.

Measure timestamps for request accepted, scope complete, budget confirmed, security complete, legal complete, commercial agreed, approval complete, signature and PO release. Segment new purchases from renewals and route reviews by value, data risk and business criticality. The requisition control boundaries help define what evidence must be complete before downstream cycle time starts.

A 2026 SaaS control set for CFOs, procurement and IT-finance

The common design is a small central standard with distributed accountability. Procurement owns commercial process and contract evidence; IT or software asset management owns application and entitlement visibility; security and legal own their reviews; finance owns the spend perimeter, budget treatment and realized financial result; business owners remain accountable for need and outcome.

How to use these statistics

This article is current through August 25, 2026. It gives priority to direct platform, transaction, contract and usage data for observed behavior; APQC for a tightly defined process benchmark; named surveys for visibility and practice; and Gartner only for forecast market context. Vendor customer cohorts are named because their figures are not universal company benchmarks.

For an internal benchmark, preserve the source, population, statistic type, period, numerator, denominator, endpoint and calculation. Refresh this page when a source issues a new annual index, changes a definition, replaces a public page, revises a forecast or publishes a later quarter. A benchmark should be retired when its provenance or denominator can no longer be recovered.