← Writing

The 3.7 Percent Problem

IT budgets are set to grow 3.7 percent next year while funding for agentic AI grows 31.8 percent. Every AI dollar in your 2027 plan is coming out of something else. Here's where the money is really coming from, and a one-page way to choose before the budget chooses for you.

Gartner asked CIOs what 2027 looks like and came back with two numbers that don’t fit in the same budget. IT budgets are projected to grow an average of 3.7 percent. Funding for agentic AI is expected to grow an average of 31.8 percent, faster than any other technology line in the survey.

3.7%projected average growth in IT budgets for 2027Gartner, 2026
31.8%projected average growth in funding for agentic AIGartner, 2026

Gartner calls the gap a “funding mismatch.” I’d put it more plainly. When one line grows more than eight times as fast as the budget it sits in, the difference comes out of the other lines. Every AI dollar in your 2027 plan is a swap, and something else gets less so that AI can get more.

That part is arithmetic, and nobody in the budget meeting will argue with it. What gets skipped is writing down what the something else is.

The raise is already spent

Start with the 3.7 percent, because it’s smaller than it looks. Consumer prices rose 3.4 percent over the twelve months through August. Gartner makes the same point in its own write-up of the survey: after inflation, many organizations will see only marginal gains in available funding.

Then look at what it costs to keep what you already own. Microsoft raised list prices on July 1. Microsoft 365 E3 went from $36 to $39 per user per month, an 8 percent increase. Office 365 E3 went from $23 to $26, which is 13 percent. The frontline F3 plan went from $8 to $10, which is 25 percent. Existing customers move to the new prices at renewal, so if yours falls in the next twelve months, that increase is in your 2027 budget.

That’s one vendor and one product family, and the increase on the E3 plan is more than double the growth rate of the budget that pays for it.

Four growth rates, one budget
The budget grows like prices. AI funding grows like nothing else in it.
CONSUMER PRICES, 12 MONTHS TO AUGUST 20263.4%AVERAGE IT BUDGET, 20273.7%MICROSOFT 365 E3 LIST PRICE, JULY 20268%FUNDING FOR AGENTIC AI, 202731.8%
Sources: U.S. Bureau of Labor Statistics, Consumer Price Index, August 2026. Gartner, 2027 CIO and Technology Executive Survey. Microsoft, 2026 Microsoft 365 pricing update. Bars drawn to scale.

The takeaway: the 3.7 percent is spoken for before AI shows up. Plan next year’s IT budget as flat, and plan every AI line as a transfer from somewhere else.

The swap is already happening, mostly by accident

Futurum surveyed 1,636 technology decision makers this fall and found that 46.9 percent are already over budget on AI. What the overspenders do next is the useful part. About one in six, 17.2 percent, pause or shrink the AI initiative. Another 43.3 percent absorb the overrun and settle it later, and 23.1 percent pull the money from a business unit’s budget outside IT.

46.9%of enterprises report AI spending over budget in the second half of 2026Futurum, 2026

“Absorb and settle later” is a swap with the name left off. The money still comes from somewhere. It gets decided late in the year by whoever is short, and the line that loses is usually the one that’s easiest to cut, which is rarely the one that’s worth the least.

Futurum also asked what goes first when the gap gets closed inside IT. Of the 297 organizations doing that, 60.9 percent cut external contractors and consultants before anything else. I’m a consultant, so I noticed.

BCG sees the same motion from the buying side. Its May survey of 423 IT buyers in North America and Europe found funding being reallocated to AI “from nearly everything else,” with spending intent for IT services down 19 points and CRM and ERP turning negative.

An unplanned swap eventually shows up as cancelled work. In Zylo’s survey of 218 IT leaders, 61 percent said they were forced to cut projects because of SaaS cost increases they hadn’t planned for.

“Absorb and settle later” is a swap with the name left off.

The takeaway: roughly half of companies are already over budget on AI, and most of them are covering it after the money is gone. Make the swap in the budget, while you still get to choose what loses.

Most of it isn’t in the CIO’s budget

The next number is what turns this into a CFO problem. BCG’s Applied AI Index, a survey of 1,330 senior executives published last week, puts AI spending at 3.3 percent of revenue, with more than 80 percent of it sitting outside enterprise IT. Zylo’s license data says much the same about software in general. Business units control 81 percent of SaaS spend, and IT directly manages 15 percent.

80%+of AI spending sits outside the enterprise IT budgetBCG, 2026

So the 3.7 percent describes the smaller part of the picture. Most AI swaps are being made inside marketing, operations, finance, and customer service budgets, by leaders who each see one line. In most enterprises there’s no rule for who adds them up. Gartner found that 73 percent have no plans to develop rules for who owns technology costs.

I wrote this summer about why your CFO should be your AI champion. This is the budget-season version of that argument. The CIO can see the IT lines, and only finance can see all of them.

The takeaway: an AI budget review that only covers IT is looking at less than a fifth of the spend. Ask finance for every AI line in the company on one page before anyone approves a new one.

The best argument against all of this

The fair objection has two parts: budgets aren’t really flat, and AI pays for itself. Both have evidence behind them. BCG’s buyers expect 2026 IT budgets to come in 5.8 percent higher, up from the 3.6 percent they had planned. Forrester found that 82 percent of technology decision-makers expect a budget increase in 2027. And in BCG’s index, nearly half of companies say AI is generating meaningful value.

Take the 5.8 percent over the 3.7 and the problem doesn’t change, because neither one is close to 31.8.

“AI pays for itself” is the stronger claim, and it’s true only when someone collects. Accenture surveyed 750 executives at large companies and found that less than one dollar in five of token spend can be traced to a quantified financial outcome. Seven percent use a chargeback model for AI costs. In Gartner’s survey, 36 percent of respondents believe they’ll meet the cost-reduction targets they’ve been handed, and 13 percent report significant value from AI tools so far. Deloitte asked 1,434 finance leaders how large AI and technology investments usually get approved, and 23 percent said by executive or board mandate, with no measurement process attached.

Under 20%of AI token spend can be traced to a quantified financial outcomeAccenture, 2026
7%of large companies charge AI costs back to the teams that incur themAccenture, 2026

I spent fifteen years on the consulting side of large technology programs, and the failure I saw most often was the one where the system worked and the money never came back. The savings were in the business case. Nobody’s budget was ever reduced by them, so the old system kept running next to the new one and the company paid for both. AI is walking into the same trap with a larger number attached.

The takeaway: “self-funding” is a claim that someone’s budget gets smaller on a date. If the business case doesn’t say whose budget and which date, plan as if the AI line is all cost.

Name the swap

The fix fits on one page, and it’s the same discipline I’d apply to any large technology investment. For every AI line in the 2027 plan, write down what it replaces.

The Swap Sheet

  1. The AI line. The amount, the business owner, and the one workflow it changes.
  2. What it replaces. A named line somewhere in the company: an application, a contract, a vendor, an open role, a project. Put a dollar figure on it.
  3. The switch-off date. The day the old thing stops being paid for. No date, no savings.
  4. Whose budget shrinks. The owner of the replaced line, the amount, and the quarter it comes out.
  5. The number and the check. The one operating number the AI line has to move, where that number stands today, and the date someone looks again.

Sometimes nobody can fill in line two, and that’s a useful answer. It means the line is an experiment. Experiments deserve funding, from a small pool with a cap on it and an end date on each one. Gartner’s advice is harsher than mine. It tells CIOs to require every AI investment to show business value within 90 days or face termination, and to eliminate the “zombie AI” projects that keep consuming resources. Forrester’s 2027 planning guides say to cut AI initiatives that lack governance, clear ownership, success criteria, or a defined path to scale.

When a team is stuck on line two, these are the places I’d send them first:

  • Licenses nobody uses. Zylo’s analysis of more than 40 million licenses found that organizations leave an average of 36 percent of their SaaS licenses unused.
  • Renewals. Every contract that renews in 2027 is a chance to drop a tier or a vendor. The denominator post covers how to walk into that conversation.
  • Work that’s finished and still funded. When I inventory a technology portfolio, I expect to find initiatives that are done and still drawing budget, and the same initiative funded twice in two business units.

There’s one thing I wouldn’t offer up as a swap, and that’s the senior people who check the work. AI lines raise the volume of work that needs checking. Jamming the Funnel covers what happens when production scales and review doesn’t.

Before your budget is final, ask for the swap sheet on the three largest AI lines in the company. If line two comes back blank on all three, you’ve learned what your 2027 AI budget is: a set of experiments funded by cuts nobody has chosen yet.

The money is coming out of something. You should be the one who picks what.


How RLK Can Help

I work with CIOs and CFOs to decide which line in next year’s budget pays for AI, before the budget makes that call for them. Start a conversation.


Sources

Ryan King

About the author

Ryan King

Fifteen years in technology strategy at McKinsey and Deloitte. Now running RLK Consulting: enterprise-caliber tech strategy, one strategist doing every hour of the work. Over $10B in documented value capture across 50+ engagements and 12 industries.