MMOOZZGG.com

AI investment

How much to spend on AI: benchmarks for a company

6 min read2 sources

How much to spend on AI? Two benchmarks answer it: in Russia, companies with the biggest AI effect spend 13–17% of their IT budget on AI, and worldwide 28% of organizations spend more than 10% of their ICT budget. A share of budget sets the frame, and the exact sum comes from a calculation on your company’s processes.

On this page · 5 sections

Benchmarks: Russia and the world

The Russian benchmark comes from a 2025 industry study: a survey of 150 technical directors from 16 industries. Companies with the biggest effect from AI spend 13–17% of the IT budget on it and get 6.6% of EBITDA, against 15.8% in their expectations.

The world benchmark comes from the McKinsey survey: 1,719 participants from 97 countries, May–June 2026. 28% of organizations spend more than 10% of their ICT budget on AI, and 60% expect AI investment to grow next year.

The studies use different bases: one takes the IT budget, the other the ICT budget, which also includes telecom. So the numbers show the order of magnitude, not a norm for your company.

How to use the benchmark: take last year’s IT budget and count 13–17% of it. If the sum is much more than you are ready to spend, start with one process. If it is less, check whether the company already pays for AI through department subscriptions.

A share of budget gives no effect by itself. Per McKinsey, 37% of organizations see an AI contribution to EBIT, and about 6% get at least 5% of EBIT from it: money works where it is counted by process.

What the budget depends on

The sum depends on four things. Write them down before you talk to contractors, or the budget will be built from their offers.

1. How many processes lose money. One process with routine work needs a pilot; five processes in different departments need a program for a year.

2. Data readiness. If the CRM and the accounting system are filled in, AI connects to them. If not, part of the money goes to putting the records in order.

3. Ready-made or custom. Typical tasks have subscription services, your own processes need AI agents, and closed data needs a model on your own servers.

4. Scale. 44% of organizations in the McKinsey survey scale AI across the whole company. Scale costs more than a pilot, and it is planned after the first number, not before it.

Company size affects the sum less than the number of processes. A small company with one expensive process may need more than a large one that starts with a pilot.

People, solutions, data

People. The main line is the person responsible for AI. In a small or mid-size company this is a fractional AI director: with us from $999 a month, 5 hours a week, with no new hire.

Solutions. Our deployment formats, with the price and term fixed in the contract:

1. Pilot: one AI agent on one process, from $1,499, 14 days, with the metric in the contract.

2. AI department: 3–5 AI agents on shared data, from $4,999, 4–6 weeks.

3. Turnkey business: several departments and a management panel, from $18,000, 2–3 months.

4. Support after launch: from $480 a month, cancel in any month.

Some tasks are covered by ready services or by your own development with agents: 32% of organizations in the McKinsey survey dropped the purchase of at least one product because they can build it themselves. This is also a budget line, compared with a purchase by the price per year.

Data. Preparing the CRM and the accounting system, exports and access rights go on a separate line. What makes up the AI budget in a company is covered in detail.

How to spread it over the year

It is easy to split the year into three stages: check, first process, scale. Money for the next stage is allocated based on the results of the previous one.

1. Start of the year: the AI solutions map. It shows the processes that lose money, the solution and effect for each, and the order of deployment. For the whole company the map costs from $93,000 and takes 7 weeks.

2. Then a pilot on the process where the result is easiest to measure. Its one-time price and metric are in the contract.

3. Then scale: an AI department or the next processes, if the pilot metric reached the target.

Monthly costs are counted from the launch month: the person responsible for AI, support and model fees by the number of requests. The earlier in the year the launch, the more of these payments fall into that year.

Set a reserve for mistakes openly: some pilots will not reach their target, and the money moves to the next process.

60% of organizations in the McKinsey survey expect AI investment to grow next year. For your plan this means one thing: build the second year’s budget on the results of the first, not in advance.

How to check that the money works

Money works if each solution has a metric: what it was, what it should become and how to measure it. For example, the share of requests the agent handled without a person, or the reply time to a client.

Once a month, compare the savings with the calculation. If the number is below plan, you see the cause: people do not pass requests to AI, the rules are out of date, or the process has changed.

At company level the check is the same as in the studies: the share of EBITDA or EBIT that comes from AI. Russian companies with the biggest effect get 6.6% of EBITDA, and worldwide about 6% of organizations credit AI with at least 5% of EBIT.

Solutions that did not move the metric within two or three months after launch get a review: change the rules, change the process or switch them off.

How to count payback before the start and check it after launch is covered on a separate page.

Sources and fact-checking

Checked by the editors on 14 October 2026.

  1. Expert, 8 December 2025 (expert.ru)
  2. McKinsey, The state of AI in 2026, August 2026 (mckinsey.com)

Found a mistake? Tell us

Publications

Mike Mozg

Agentic AI use cases at AT&T: nearly 500 processes now run on agents

AT&T no longer uses AI only to help staff work faster: it is rebuilding whole business processes around agents. SiliconANGLE reported on 14 November 2026 that nearly 500 agentic AI use cases at AT&T are already in production. Here is how the company rebuilds them, where a human still checks the agent, and what other large companies can take from it.

News4 min
Contact us