Most people searching for IT spending benchmarks in 2026 are not shopping for a statistic. They are preparing a budget, and they need a defensible answer to one question: is what we plan to spend reasonable for a company like ours? That question is answered with four ratios — IT spend as a share of revenue, the run-versus-change split, cost per employee or per site, and cost per unit of service delivered — each normalised to something your business can be compared against. A single number ("the average company spends X% of revenue") is not a benchmark, it is someone else's invoice. This guide covers which ratios to track, where defensible figures come from, how to build them from your own ledger, and how to put them in front of a board without starting an argument about whose data is right.
A note on what this post deliberately does not do: it does not publish "average IT spend" figures. Published averages mix company size, industry, capex treatment, offshore mix and shadow IT to the point where the number stops describing anything you can act on, and a benchmark copied from a blog post is not a benchmark. What follows is the method for producing numbers specific enough to defend.
Why "IT spend as a percentage of revenue" is the most quoted and least useful number
Every board pack includes it, and it is the first ratio to argue about — because it compresses everything into one figure and hides the three decisions that actually move cost.
Revenue is not a stable denominator. A company that grew through acquisition carries duplicate ERP instances and overlapping licences; a company that shrank still carries the headcount it had at the peak. The same spend produces wildly different percentages across two firms of identical size depending on whether they own their hardware, whether they run a warehouse, and whether engineering sits in the IT budget or the product budget.
The percentage is still useful for one thing: trend. Your own number, measured the same way each year, tells you whether spend is growing faster than the business. That is a real signal. Comparing your percentage to an industry average is where it falls apart.
Use it as a movement indicator, not a verdict.
The four measures worth benchmarking
Run versus change
Split total IT spend into the cost of keeping the lights on — infrastructure, licences, support contracts, the help desk, monitoring, security operations — and the cost of changing the estate: projects, migrations, new systems, automation.
The split tells you what your organisation can actually absorb. A company where run cost consumes nearly everything has no capacity for the work the board keeps asking for, and the honest conversation is about the run estate, not about the next project. A company with an unusually high change share is either mid-transformation or accumulating technical debt it will pay for later.
This is the most useful ratio in the pack, and the one almost nobody reports.
Cost per employee, per site, or per endpoint
Normalise spend over a unit the business already understands. Per-employee works for office-based companies; per-site works for retail, clinics and plants; per-endpoint works when the mix of devices is the real cost driver.
The value is in comparison over time and between your own business units, not against a market average. Two divisions of the same company with the same headcount and very different IT cost per head is a conversation worth having — usually about local contracts, local vendors, or a site that has quietly built its own stack.
Cost per unit of service delivered
For each significant service, define the unit and price it: cost per resolved ticket, cost per release shipped, cost per onboarding, cost per invoice processed, cost per gigabyte stored and served. Then track it.
Unit cost is where efficiency shows up before the P&L does. Ticket volume rising while headcount stays flat shows in cost per ticket long before anyone proposes a new hire. And when a vendor proposes an outsourced service, unit cost is the only basis on which two bids can honestly be compared — see how effective blended rate differs from cost per unit of delivered work in our guide to enterprise IT pricing benchmarks.
Concentration and commitment
Two structural measures that predict next year's problem: how much of your spend sits with your largest vendor, and how much of it is locked into multi-year commitments.
High concentration is not automatically bad — one strategic partner can be cheaper and better coordinated than five specialists. But an estate where a single vendor holds most of the spend and every contract renews in the same quarter has no negotiating room. Track the top three vendors as a share of total spend and put the renewal calendar next to it. If three renewals land in one quarter, that is a budget risk you can still fix a year ahead.
Where defensible numbers actually come from
There are four sources worth citing in a board pack, in descending order of usefulness.
Your own general ledger, mapped to a consistent taxonomy. This is the only source that describes your company; everything else is context for it. The work is unglamorous: agree the categories once, map every invoice to them, and keep the mapping stable so year-over-year comparisons mean something.
Your vendors' invoices and contracts. Not the rate card — the invoice. Effective rate, after discounts, credits and true-ups, is what you actually pay, and it is usually different from the number in the proposal.
A peer cohort you can name. Five to fifteen companies of similar size, sector and geography, where you can compare definitions as well as numbers. A dirty comparison against three companies you know well beats a clean comparison against an anonymous average, because at least you can ask "does that include your ERP team?".
Published benchmark studies, read carefully. Analyst studies and industry association surveys are useful for structure — which categories a complete benchmark should contain, which ratios the market tracks — and for a sanity check on direction. Read the methodology before the number: what is in scope, what is excluded, whether internal staff are included, how capex is treated, and which year the data describes. A study that excludes internal staff will make your own number look enormous by comparison.
Building the benchmark from your own ledger
Start with a complete category list, because the most common error is an incomplete one. At minimum:
- Software and SaaS subscriptions, including the tools bought on cards
- Cloud infrastructure and platform consumption, including committed-use true-ups
- Data centre, colocation and hosting
- Hardware: endpoints, servers, network, and the refresh cycle behind them
- Telecom and connectivity, including the circuits nobody remembers ordering
- Internal IT staff fully loaded: salary, benefits, on-costs, training
- Contractors and staff augmentation, split by engagement type
- Managed service and support contracts, with scope
- Security tooling and security operations
- Backup, disaster recovery, and the DR test that never happens
- Project spend, kept separate from run
Then apply two rules. Capex and opex must be normalised — annualise hardware bought outright, or state plainly that the comparison excludes capex. And shared cost must be allocated deliberately: licence spend that serves product engineering is still IT spend, even when the product budget pays for it.
Normalising before you compare
Four factors explain most of the gap between two companies with the same revenue and very different IT cost. Adjust for them before drawing any conclusion.
Scale. Fixed costs — the ERP instance, the security stack, the on-call rotation — do not shrink with headcount. Smaller companies will always look expensive per employee in a way that says nothing about efficiency.
Geography and delivery mix. An estate delivered from a lower-cost region costs less per unit of work, and a company with an offshore or nearshore mix will look cheaper on run cost without being better run.
Ownership. Owning hardware, running your own data centre, or operating a private cloud shifts spend from a vendor line to a depreciation line. Neither treatment is better, but comparing them without adjustment is meaningless.
Regulatory load. A regulated company carries audit, evidence, segregation-of-duties and retention costs that an unregulated one does not. Those are not waste, and benchmarking them away produces a budget the auditors will return.
Reading the result when you are the outlier
Being above the cohort is not automatically a problem, and being below it is not automatically good — an unusually low IT cost per employee is often the signature of an under-invested estate, and its bill arrives later as a security incident or a failed audit.
When you are above: decompose before you cut. Separate structural cost (regulation, ownership, geography) from discretionary cost (vendor terms, licence sprawl, duplicate tools, unmanaged SaaS). Structural cost is not a savings target. Discretionary cost usually is, and the fastest wins are licence rationalisation and renegotiated terms rather than headcount.
When you are below: identify what you are deferring. Endpoint refresh, backup testing, security coverage, and the documentation that lets someone else operate the estate are the usual candidates. Deferred maintenance is a real liability even when it never appears on the balance sheet.
The useful output of benchmarking is not a number to hit. It is a short list of specific decisions, each with an owner and a saving or a risk attached.
Using the benchmark in a vendor negotiation
A benchmark built this way changes the conversation. Instead of asking for a discount off a rate card, you arrive with your own cost per ticket, cost per endpoint, and the effective rate you currently pay after discounts.
Three moves follow. Ask for pricing against your unit rather than the vendor's package — cost per resolved ticket, cost per managed endpoint, cost per release. Ask what is excluded and price the exclusions, because the gap between two bids is usually scope, not rate. And align the term with your renewal calendar so you are not negotiating while a second contract is already mid-flight.
For the structure of an outsourced agreement and where the cost actually sits inside it, see our managed IT services practice. For the CFO-level framing of what the investment returns, this breakdown of managed IT ROI covers total cost of ownership, risk reduction, and the costs that never show up on an invoice.
A one-page pack the board will accept
Keep it to one page and four blocks.
- Total IT spend split run versus change, with the year-over-year movement on each
- The four ratios, each shown for the previous two years on the same basis
- Vendor concentration and the twelve-month renewal calendar
- The three decisions you are asking for, each with a cost, a risk and an owner
Everything else — the ledger mapping, the cohort definitions, the normalisation adjustments — belongs in an appendix that answers questions rather than raising them. The appendix is what makes the number defensible; the one page is what makes it actionable.
Doing this once a year is not enough
A benchmark built once becomes a historical document within a quarter. Rebuild the ratios monthly from the same taxonomy, put them beside the budget, and the annual exercise turns into a monitoring tool: licence growth visible before the renewal, cloud consumption drift visible before the true-up, cost per ticket climbing before anyone proposes a new hire.
That is the difference between a benchmark you present and a benchmark you run.