Why invest now
Companies worldwide are raising their AI investment. In a McKinsey survey of 1,719 people from 97 countries (May–June 2026), 60% of organizations expect AI investment to grow next year, and 44% already scale AI across the whole company.
Among large companies, 40% scale AI agents: programs that handle requests, reconciliations and reports in work systems on their own. An agent works inside a process, so the process owner decides on it, not only the IT department.
For Russia, the estimated effect is also large. By an industry study of 2025, AI could give the economy $95 billion to $150 billion a year by 2030, up to 5.5% of forecast GDP.
A market estimate does not say how much a given company will get. Only a calculation based on its processes gives the answer: where time and money are lost and what part of this work AI will take over.
Waiting also costs money. While the decision is put off, manual work on requests, reconciliations and reports costs the company the same amount every month.
The giants spend hundreds of billions of dollars on AI, but their spending goes to compute capacity and models, not to the processes of an average company. How much the world’s largest corporations spend and on what is on the page about the capital spending of the giants.
How companies with an effect differ from the rest
The gap between those who adopt AI and those who earn from it is large. By McKinsey, 44% of organizations scale AI, 37% see at least some AI contribution to EBIT, and only about 6% credit AI with at least 5% of EBIT.
Russian numbers show the same picture from another side. By an industry study of 2025 (a survey of 150 technology directors from 16 industries), companies with the strongest effect spend 13–17% of their IT budget on AI and get 6.6% EBITDA, against 15.8% expected.
These data lead to three conclusions for your company.
1. The effect comes from a budget planned around processes, not from a one-off purchase of subscriptions. Companies with a return spend a visible share of their IT budget on AI and plan it for the year.
2. The return is measured in money: a share of EBIT or EBITDA, savings per process. Without such a number, a company does not know which group it is in.
3. Scale is not the same as results. About 6% of organizations see a visible contribution to profit, against 44% that scale AI: what decides is not the number of solutions launched but which processes were chosen.
The difference between the groups starts before the first contract: with the choice of the process, the metric and the person responsible for the result.
You can test yourself with three questions. Does each AI solution have a metric in money? Is there a person responsible for it? Do you know how much the company spends on AI a year? If the answer to even one question is no, you have nothing yet to prove a return with.
What to invest in: processes, data, people
AI money splits into three items. Too much weight on one of them turns investment into spending without a return: a model without data makes mistakes, and a process without an owner stands still.
Processes. Invest where the company loses time and money: requests, support, reconciliations, reports, purchasing. By MIT NANDA, only 5% of custom AI tools reach real use, so start with the process and its losses in money, not with the choice of a tool.
An example: sales managers spend hours moving requests from messengers into the CRM. The investment here is an AI agent that creates deals on its own, plus order in the CRM it works with.
Data. AI works on what sits in the CRM, in 1C, a common Russian accounting system, and in email. If half the fields are empty and requests from messengers never get recorded, the first money goes into cleaning up the data.
People. You need a person responsible for AI and process owners who accept the agents’ work. In a company this is the AI department: people and rules, not just a contractor.
A separate item: ready-made or your own. In the McKinsey survey, 32% of organizations dropped the purchase of at least one product because they can build it themselves with agents. How to choose between your own AI and a ready-made solution is covered separately.
How much to budget
There is no single sum, but there are guides. The Russian companies with the strongest AI effect spend 13–17% of their IT budget on it, by an industry study of 2025.
Worldwide, 28% of organizations spend more than 10% of their IT budget on AI, by the 2026 McKinsey survey. The studies use different bases, IT and ICT, so compare the numbers as an order of magnitude, not exactly.
Check the budget share with a bottom-up calculation: what the routine in your processes costs, what part AI will take over and what implementation and support cost. If the calculation does not match the share, trust the calculation.
An example of a bottom-up calculation: five sales managers spend half the day moving requests and answering standard questions. The budget for this process is counted from the cost of that half day with taxes, not from a share of the IT budget.
The sum includes one-off costs (the check, the pilot, implementation) and monthly costs (the person responsible for AI, support, model fees). Monthly costs are counted from the month of launch, not from the start of the year.
The first-year sum is easy to split into two parts: the check and the first process with a metric, then scale based on the pilot results.
Guides on budget share and how to calculate are on the page about how much to invest in AI. What makes up the annual budget is covered in the section about the AI department.
How not to lose your investment
By the 2025 MIT NANDA report, 95% of organizations got no measurable return from generative AI. Look for the cause in decisions made before the start: what to implement, how to measure and who is responsible.
1. A metric before the start. What it was, what it should become and how to measure it. The metric goes into the contract with the contractor. How to calculate payback before launch is covered separately.
2. A check before you buy. AI due diligence, a full check of the company before AI investment, shows the processes, data, systems, costs and solutions already bought. After it, it is clear what to invest in and what not.
3. One process first. A pilot on one process with a metric shows a number in weeks, not in a year. The next step comes when the first one has given a result.
4. Control over suppliers. If a solution works only with one contractor or on one foreign model, access can be cut or the price raised. Keep the rights to the code, the data and the access with the company.
5. Stop in time. Once a month, compare savings with the plan and take money away from solutions that did not reach the goal.
Where to start
It is too early to start by choosing a model or a contractor. First you need an answer to where AI will return money in your company and how much.
1. A 30-minute review, free. We learn what the company does and where it hurts most.
2. A check by department: processes, data, systems, costs, solutions already bought.
3. The AI roadmap: for each process the solution, the price, the effect in money and the order of implementation. The roadmap also includes the option to do nothing if a process needs no AI.
4. A pilot on the first process from the roadmap, with the metric in the contract.
5. Scale: the next processes based on the pilot results.
A company that has already bought AI needs the same path: there are subscriptions, bots and contractors, but no overall picture. The check shows which purchases bring money and which can be switched off.
This order does not promise a place among the 6% of companies with a visible AI contribution to EBIT. It gives what you cannot get there without: a calculation before the start and a check by numbers after launch.