Why AI isn't increasing profits, McKinsey 2026 data

Almost every company already uses AI, and still profit rises for only a handful. McKinsey's State of AI survey from late August 2026 shows that 88 percent of organizations use AI regularly, and the gains show up at individual desks, yet only about 6 percent of companies see a significant impact on their financial result. That gap, confirmed by Deloitte's report from the same period, is the headline for boards: simply having AI in the business is no longer an advantage, because everyone has it. The advantage comes from something else, redesigning the process so AI takes work out rather than sticking to an old flow. That is an organizational decision, not a license purchase.

What happened

In late August 2026 McKinsey published its annual State of AI report, based on a survey of 1,719 professionals and business leaders. The results form a single picture. AI is in regular use at 88 percent of organizations, and 64 percent of respondents say AI is enabling their innovation. At the same time only 39 percent of firms report any EBIT impact from AI at the enterprise level, and a significant impact, defined as at least 5 percent of EBIT attributed to AI, shows up at about 6 percent of organizations. McKinsey calls that group high performers.

The key observation is what sets those 6 percent apart. According to McKinsey, high performers are roughly three times more likely to fundamentally redesign their workflows while adopting AI, they scale solutions faster, they introduce transformation practices earlier, and they invest more. The rest of the market is stuck at an earlier stage, with nearly two-thirds of firms yet to begin scaling AI across the organization. AI is working at people's desks, but not in the financial result.

Deloitte's State of AI in the Enterprise 2026, based on 3,235 leaders across 24 countries and six industries, describes the same gap from the deployment side. Worker access to sanctioned AI tools grew by half in a year, from under 40 to about 60 percent. But among the people who have that access, fewer than 60 percent use AI in their daily work, and that share has not changed year over year. Deloitte calls this the move from ambition to activation and notes that agentic AI adoption is outpacing the development of oversight. Companies where senior leadership shapes AI governance, rather than only a technical team, capture more business value.

Two independent studies, McKinsey and Deloitte, say the same thing at the same time. AI adoption is now widespread, but value is concentrated among a few, and the dividing line is not the technology, it is how a company rebuilds work.

Why it matters for business

For the past two years, simply switching AI on signaled that a company was modern. This data closes that chapter. If 88 percent of organizations use AI, then having it no longer sets anyone apart from the competition. The board's question shifts from "are we using AI" to "is AI changing our cost and revenue", and for most companies the answer, for now, is no.

The reason is mechanical. When a worker saves a few minutes drafting an email or summarizing a document, that gain stays with them as more comfort or a small buffer of time. It does not automatically become lower cost or higher revenue for the firm, because the process around them did not change, the same number of people handle the same number of cases in the same way. Profit appears only when the company redesigns the whole flow, removes work from it, moves people to higher-value tasks, or handles more cases without growing the team. That is the difference between AI as a convenience and AI as a lever on the result.

For a smaller company there is good news in this. The high performers' edge is not access to better models, since the same models are available to everyone. It is the decision to rebuild a specific process, name its owner, and measure the effect. A smaller firm has an advantage here, a shorter decision path and a simpler flow that is easier to redesign end to end. The barrier is neither technological nor budgetary, it is organizational, which puts it within reach of any company regardless of size.

Business use cases

Handling inquiries and complaints end to end

Business problem: the support team is drowning in repetitive inquiries, response time is climbing, and hiring more people raises cost without improving quality.

AI solution: an assistant that reads the ticket, drafts a reply from the knowledge base and customer history, and for simple cases resolves it on its own within set rules, routing only hard or risky cases to a person.

Data and processes to put in place: a well-ordered knowledge base, ticket categorization, access to customer history, clear rules for when AI answers and when it hands over, plus escalation and a record of decisions.

Potential effect: shorter response time and more cases handled without growing the team, which turns minutes into the real cost McKinsey is talking about.

Risk and limitation: a wrong model answer when the knowledge base is incomplete, which is why quality control and a confidence threshold below which a case goes to a person are needed.

Quoting and sales preparation

Business problem: salespeople spend hours preparing quotes, configurations, and answers to inquiries, leaving less time for talking to the customer and closing deals.

AI solution: an assistant drafts an initial quote and answers to common questions from the price list, the history of similar deals, and customer data, while the salesperson verifies, negotiates, and closes.

Data and processes to put in place: a current price list and discount rules, a base of earlier quotes, data from the sales system, and the principle that a person approves the final quote.

Potential effect: more inquiries handled in the same time and a shorter sales cycle, which raises revenue rather than only the salesperson's comfort.

Risk and limitation: stale pricing data leading to a wrong quote, and the temptation to hand the model discount decisions that should stay with a person.

Finance operations and document flow

Business problem: accounting and administration spend a lot of time manually transcribing invoices, matching payments, and preparing repetitive reports.

AI solution: the model reads documents, classifies them, matches them to orders and payments, and drafts reports that a person checks and approves.

Data and processes to put in place: a structured document flow, integration with the accounting system, coding and approval rules, and control over who approves items above set thresholds.

Potential effect: a shorter monthly close and less manual work, while a person keeps responsibility for the books.

Risk and limitation: errors in financial data are costly, so verification and a full audit trail for every model action are required.

Marketing, content, and market research

Business problem: a small marketing team cannot keep up with content, market analysis, and materials, and outsourcing is slow and expensive.

AI solution: an assistant drafts content, summarizes source material, and organizes research, while the team edits, fact-checks, and decides what to publish.

Data and processes to put in place: brand and tone guidelines, reliable sources, a fact-checking step before publication, and a clear split between what the model drafts and what a person approves.

Potential effect: more material from the same team and a faster response to market events.

Risk and limitation: the model can produce content that sounds credible but is unverified, so without fact-checking the risk of error and lost trust rises.

What companies can do now

Instead of buying more tools, start with one process that genuinely costs time or money. Pick an area where the work is repetitive and measurable, such as handling inquiries, quoting, or invoice processing, and treat it as a first full project, not a loose experiment.

The next step is documenting the process and assigning an owner. Someone in the company has to be accountable for getting the deployment done and have the mandate to change how the work is done. Deloitte shows that firms where leadership holds oversight of AI capture more value, so the decision to redesign a process should sit at the management level, not be left to IT alone.

The third step is data. AI works on what it can read, so putting the knowledge base, price list, customer history, or documents in order is often more important than the choice of model. Without clean data, even a good model produces results you cannot trust.

The fourth step is redesigning the flow, not bolting on a tool. The point of a deployment is to take work out of the process, hand the model the repetitive tasks, and leave decisions, exceptions, and quality control to a person. If, after deployment, people do the same as before plus operate a tool, no value appears.

The fifth step is measurement. Pick one hard metric, handling time, cost per case, number of inquiries handled, or time to close the month, and measure it before and after. Without that number you cannot tell real value from the impression of being modern.

Risks

The biggest risk is dispersion. A company launches a dozen pilots, each adds a little comfort, but none is driven to the scale where it changes the result. McKinsey shows that nearly two-thirds of organizations have not begun scaling AI, and that is exactly the trap.

The second risk is shadow AI, the use of tools outside the company's control. Deloitte notes that agentic AI adoption is outpacing governance, so it is easy to end up with company data flowing into external models without oversight and responsibility for the result blurred.

The third risk is the absence of a process owner. A deployment without a person accountable for it and empowered to change how work is done ends as a tool nobody uses, or uses in their own way, with no effect on cost.

The fourth risk is a lack of quality control over output. A model can produce an answer that sounds credible but is wrong. Without a confidence threshold, verification, and an audit trail, such errors enter the process and erode trust in the whole deployment.

The fifth risk is cost rising faster than the effect. Without measuring a hard metric, it is easy to keep a deployment that costs more than it returns, only because it looks modern.

Key takeaways

AI adoption has stopped being an advantage, because almost everyone has it. Per McKinsey, 88 percent of organizations use AI, but only about 6 percent see a significant impact on profit. The difference is not a better model, it is redesigning the process, led by leadership and measured by a hard metric. For a smaller company that points to a concrete path, choose one process, put the data in order, name an owner, rebuild the flow so AI removes work, and measure the effect. It is an organizational decision within reach of any company regardless of budget, and it, not the technology, is what now divides the market into those that earn from AI and everyone else.