📉   Top 2026 CEO Trends   📉
Download Report
Cover image reading How Much Should a Company Spend on AI, with a diagram of budget tiles leading to a result and the Journey AI logo

How much should a company spend on AI? BCG found that the companies investing most in AI commit about 1.7% of revenue to it, against 0.8% for laggards. For a company of 25 to 500 people, about 1% of revenue is a sensible ceiling to test against. The split matters more than the total: most of the money should go to changing how people and processes work, not to tools, and it should be released in stages, each tied to proof.

Most CEOs get this backwards. They approve seats for an AI assistant, see little change, and conclude the budget was too big or too small. Usually it was just pointed at the wrong things. This post gives you a reference number, a way to split it, and a step-by-step method to build your own AI budget for next year.

What do companies actually spend on AI?

An AI budget is every dollar and every hour your company puts into AI: licenses, usage fees, build work, data cleanup, training and the time of the people who own the results. Most companies only count the first one.

The best public benchmark comes from BCG. Its 2026 research found that the top AI investors commit 1.7% of revenue to the technology, against 0.8% for laggards. BCG's argument is that the bigger risk today is spending too little, not too much.

In the middle market, budgets are rising fast. RSM's 2026 survey of 1,030 leaders at companies with $30 million to $10 billion in revenue found that 58% plan to invest $1 million or more in AI this fiscal year, and 84% expect to spend more next year. In a Zapier survey of 715 US professionals, 52% of managers said their organizations spend more than $100K a month on AI-related tools.

Treat these as context, not targets. These surveys include much larger companies than most of our readers run. A 120-person company does not need a seven-figure AI line to make progress. It needs a number it can defend and a clear idea of what each dollar buys.

Why is the total the wrong place to start?

Because AI value comes mostly from changes that never show up on a software invoice. BCG's rule of thumb is 10% algorithms, 20% technology and data, and 70% people and processes. That is a description of where the effort and the value sit. Many budgets invert it: most of the money goes to tools, and close to nothing goes to the work around them.

The gap shows up in the data. An Economist Impact survey of 639 senior decision-makers found that only 38% have a dedicated budget for AI development, and just 16% use structured internal training. Zapier's research found that untrained workers are six times more likely to say AI makes them less productive.

So when a CEO asks how much to spend, the better question is: what are we paying for, and how will we know it worked?

What should an AI budget pay for?

Split your AI budget into five lines. Each one has a typical mistake attached.

  • Seats and licenses. The AI assistants your team uses every day. Cheaper than most CEOs expect. The mistake is buying seats for everyone and calling it a strategy.
  • Usage fees. What you pay per task when an agent or automation calls an AI model. Small at first, and it grows with volume. The mistake is not setting a ceiling.
  • Build and integration. Connecting AI to your CRM, ERP or inbox so it can do real work. The mistake is funding this before the workflow is defined.
  • Data cleanup. Fixing the broken fields and missing records that stop any automation from running. The mistake is skipping it and blaming the AI.
  • People and process change. Training, workflow redesign and the time of the person who owns each result. This should be the biggest line. In most budgets it is the smallest, or missing.

A useful test: if you deleted every software line, would anything be left? If not, you have a tool budget, not an AI budget.

How should your AI budget change as you mature?

This is where most budget advice stops being useful, because it gives one split for every company. At Journey we see four stages of AI maturity: Scattered, Experimenting, Operating and Compounding. (We cover how to place your company in our post on the AI maturity model.) The right budget looks different at each one.

  • Scattered: people use AI on their own, often on personal accounts. Spend little and consolidate. Replace the scattered subscriptions with one approved tool, and fund fluency for you and your leadership team first.
  • Experimenting: pilots are running, but none has moved a number. Fund two or three bets, each with one owner and one metric. Put most of each bet's money into redesigning the workflow, not into software.
  • Operating: some AI workflows run in production and someone is accountable for them. Fund the owners, the run costs and the integrations that turn a working pilot into the normal way of doing the job.
  • Compounding: AI is part of how the company works. Fund new bets from the savings the earlier ones produced.

The common error is budgeting two stages ahead. A Scattered company that signs a large build contract is paying for an Operating-stage asset with no owner and no defined workflow to run it.

How do you build an AI budget, step by step?

Here is the method we use with CEOs. It takes about a week of part-time work, mostly with your finance lead.

  1. Count what you already spend. Pull every AI subscription, seat and usage bill from the last 12 months, including ones expensed on personal cards. You may find more tools than anyone chose.
  2. Set a reference number. Multiply annual revenue by 1%. For a $40 million company, that is $400,000. This is a ceiling to test against, not a target to hit. If your plan comes in well below it, that is fine, as long as you can show what each dollar buys.
  3. Price the seats. This line is usually small. OpenAI lists standard ChatGPT Business seats at $20 per user per month billed annually. A hundred seats is $24,000 a year, about 6% of the $400,000 reference. Buy seats for the teams with a defined use, not for the whole company on day one.
  4. Fund two or three bets, not twenty. Choose them by business impact. Our post on how a CEO should prioritize AI projects walks through a scoring method. Give each bet one owner and one financial metric.
  5. Release each bet in three tranches: What, How, Build. Fund "What" first: a small, fixed amount to define the work and measure today's baseline. Fund "How" next: redesigning the workflow and cleaning the data it depends on. Fund "Build" last, and price it only once the first two are done. You cannot price a build well before you know how much is broken.
  6. Put about 70% of each bet against people and process. Inside every bet, budget the owner's time, training for the team that will use it, and the workflow redesign. If the software line is the biggest line in a bet, look again.
  7. Write the kill and scale rules before you spend. For each tranche, name the number that must move and the date you will check it. If it moves, release the next tranche. If it does not, stop and move the money.
  8. Decide what pays for it. Companies rarely fund AI from new money alone. RSM found that companies raising AI spend most often plan to cut business intelligence and analytics (43%), cybersecurity (41%) and external consulting (40%). Be careful with the second one. BCG notes that traditional cost levers can generate savings of 5% to 25% within 3 to 12 months, which can then fund AI work.

The tranche structure is the part most budget guides we reviewed leave out. It turns one large, uncertain number into a series of small decisions, each made with better information than the last.

How do you know you are spending AI money badly?

Watch for these signs in your next budget review:

  • You pay for AI seats that nobody can tie to a specific job.
  • Pilots have been running for more than a quarter with no metric attached.
  • There is no line for training or workflow redesign.
  • A build was priced before anyone mapped the workflow it automates.
  • Usage fees have no monthly ceiling and no owner.
  • All AI spending sits in the IT budget, so no business leader feels responsible for the result.

Any one of these is fixable in a month. Three or more usually means the budget was approved as a tool purchase, and it needs to be rebuilt around a few bets with owners.

Where should you start this quarter?

Start with step one. Pull the last 12 months of AI spending onto one page, next to the two or three business results you most want to move next year. The gap between those two lists is your real AI budget conversation.

Then build the plan in tranches and keep the first tranche small. You are buying information before you are buying software. If you want a second pair of eyes on where your company sits and where AI money would pay back first, Journey offers a free CEO AI audit.

Frequently Asked Questions

How much should a company spend on AI?

BCG found that the companies investing most in AI commit about 1.7% of revenue, against 0.8% for laggards. For a company of 25 to 500 people, about 1% of revenue is a useful ceiling to test a plan against. What matters more is the split: most of the budget should fund people and process change, released in stages tied to results.

What percentage of an AI budget should go to training and change?

BCG's rule of thumb is 10% algorithms, 20% technology and data, and 70% people and processes. That does not mean 70% must be spent on training courses. It means most of the effort and money should go to redesigning workflows, training the team that will use the tool, and paying for the time of the person who owns the result.

How much do AI licenses cost per employee?

Less than most CEOs expect. OpenAI lists standard ChatGPT Business seats at $20 per user per month billed annually, so 100 seats cost about $24,000 a year. Seats are rarely the biggest cost. Build work, data cleanup and workflow redesign usually cost more, and they are where the results come from.

Should we budget for AI all at once or in stages?

In stages. Fund each AI bet in three tranches: define the work and measure the baseline, redesign the workflow and clean the data, then build. Price the build only once the first two are done, and set a metric and a check date for each tranche. If the number moves, release the next tranche. If not, stop and move the money.

Where does the money for AI usually come from?

Rarely from new money alone. RSM found that middle-market companies raising AI spend most often plan to cut business intelligence and analytics, cybersecurity and external consulting budgets. BCG notes that traditional cost levers can produce savings of 5% to 25% within 3 to 12 months, which can then fund AI work. Be cautious about cutting security to pay for AI.

Find where AI pays in your business.

Get Started
Blue Pipe
We use cookies to ensure you get the best experience on our website.
Got it