---
title: "37% of firms see any profit from AI, 6% a real one, as agents spread"
summary: "What McKinsey's survey of 1,719 executives, published 25 August 2026, says about AI's return, set against PwC, Deloitte and Nasscom data from January to July 2026."
url: https://www.thetechbag.com/blog/ai-return-few-firms-see-profit
format: Analysis
desk: "AI in the enterprise"
published: 2026-08-26
facts_checked: 2026-08-26
author: TechBag (The TechBag Journal)
tags: ["AI ROI","McKinsey State of AI","AI agents","EBIT","PwC CEO Survey","Deloitte","Nasscom","Microsoft Copilot","AI measurement","software renewals"]
cover: https://www.thetechbag.com/blog/covers/ai-return-few-firms-see-profit.png
---

# 37% of firms see any profit from AI, 6% a real one, as agents spread

*What McKinsey's survey of 1,719 executives, published 25 August 2026, says about AI's return, set against PwC, Deloitte and Nasscom data from January to July 2026.*

Analysis · AI in the enterprise · By TechBag · Published 2026-08-26 · Facts checked 2026-08-26 · Canonical: https://www.thetechbag.com/blog/ai-return-few-firms-see-profit

The short version

- Only 6% of the 1,719 executives in McKinsey’s State of AI survey, published yesterday, work at firms that attribute 5% or more of their operating profit to AI and call the effect significant. That share has not moved in a year. A wider 37% see any EBIT effect at all, about the same as in 2025.[1]
- Agents are spreading faster than the profit: 40% of respondents at firms with more than US$1 billion in revenue say they are scaling AI agents, up from 27% a year ago. Usage is rising while the profit figures stay flat.[1]
- The firms that profit change how the work is done before they measure anything, and they track business results, not usage. A buyer should hold every AI renewal to the same test: a named workflow, a baseline taken before launch, and the full cost, credits included.[3][5]

Four in five people who use AI at work say it makes them more productive. Far fewer of their companies can find that gain in their accounts.[1] McKinsey’s annual survey, released on 25 August, is the best single measure of that gap, and this year it shows the gap holding steady while spending on AI agents grows. The question for anyone who signs off on an AI renewal is practical: what separates the few firms that profit from the many that do not, and how do you know which group your own deployment is in before you pay for another year?

## Most users feel faster; few firms see it in profit

McKinsey’s 2026 survey covered 1,719 professionals and business leaders across regions and industries. 37% of respondents attribute at least some impact on EBIT (earnings before interest and taxes) to AI, about the same share as in the 2025 survey. McKinsey keeps a stricter group, the “high performers”: firms where AI accounts for at least 5% of EBIT and the respondent calls the impact significant. Just 6% of respondents meet both tests, the same as last year.[1]

Productivity is where the survey looks good. 80% of respondents who use AI in their own roles say it has improved their individual productivity, and The Register notes that this has not turned into a broad financial effect.[1] These numbers come from different bases, which is why the chart shows them separately: the first is about people, the other two are about firms.

**Figure: Eight in ten AI users feel more productive; fewer than four in ten firms see any profit from it, and six in a hundred a real one**

*Share of respondents, McKinsey State of AI survey (1,719 respondents), published 25 August 2026*

80%

Say AI improved their own productivity

respondents who use AI in their roles

37%

Attribute any EBIT impact to AI

about the same as 2025

6%

High performers: 5% or more of EBIT from AI, significant impact

flat on 2025

Source: McKinsey, The state of AI in 2026, as reported by The Register (25 Aug 2026) · as of 25 August 2026 · the productivity figure is among respondents who use AI in their roles; the EBIT figures are among all respondents

ReportedWhat the 6% measures

It is a survey answer, not an audited number. Respondents say how much of their EBIT they attribute to AI and whether the effect is significant, so it records what executives believe they can trace, not what an auditor has confirmed.[1]

## Agents rose 13 points in a year; the profit share did not

Agents, AI that carries out multi-step tasks rather than answering a prompt, are where the money is going now. 40% of respondents at organisations with more than US$1 billion in annual revenue say they are scaling AI agents, against 27% a year earlier. Nearly a third of respondents say their organisations decided against buying one or more software products and built the function in-house with agentic coding tools instead.[1]

The costs are now visible: 20% of respondents say AI operating costs have limited how much they use it. Expectations about jobs have moved too. 43% expect little or no change in total employment from AI, but 39% expect their employer to cut jobs because of it in the coming year, up from 32% in 2025. McKinsey adds that last year’s reductions “fell well short of what respondents in last year’s survey had anticipated.”[1]

**Figure: Agent scaling and job-cut expectations rose in a year; the share of firms with real AI profit stayed at 6%**

*Share of respondents, McKinsey State of AI surveys, 2025 against 2026*

|  | 2025 survey | 2026 survey | Change |
| --- | --- | --- | --- |
| Scaling AI agents (firms over US$1bn revenue) | 27 | 40 | 13 pts |
| Expect AI-driven job cuts next year | 32 | 39 | 7 pts |
| High performers (5%+ of EBIT from AI) | 6 | 6 | 0 pts |

Source: McKinsey, The state of AI in 2026, as reported by The Register (25 Aug 2026) · as of 25 August 2026 · agent scaling is among respondents at firms with more than US$1bn revenue; the 2025 high-performer share is reported as flat and shown as 6%

> “Organizations’ conviction in AI is growing faster than the immediate financial returns they can attribute to it.”McKinsey, statement on the 2026 survey, as reported by The Register on 25 August

Michael Chui, a co-author, argues the delay is normal: “It should not be surprising that it has taken time.” He puts the return the high performers do see down to changing the organisation, not to adding tools.[1] That is the useful part for a buyer, and Section 4 takes it up.

## Every 2026 survey finds the same thing: a minority with a return

McKinsey is not alone. PwC’s 2026 Global CEO Survey, of 4,454 chief executives in 95 countries, found 56% saying they have got “nothing out of” their AI investments so far, and 12% reporting that AI both grew revenue and cut costs. Deloitte’s 2026 State of AI in the Enterprise, which surveyed 3,235 leaders from director to C-suite, found 66% reporting productivity gains from AI and 20% reporting revenue growth.[3] In India, a Nasscom-KPMG whitepaper of July 2026 puts the share of organisations reporting ROI across multiple AI use cases at 24%.[4]

**Figure: Whatever the question and whoever is asked, the share reporting a financial return from AI stays a minority**

*Share of respondents reporting each measure of AI return, four surveys published January to August 2026*

McKinsey: any EBIT impact from AI

37%

Nasscom-KPMG: ROI across multiple AI use cases

24%

Deloitte: revenue growth from AI

20%

PwC: AI both grew revenue and cut costs

12%

McKinsey: 5%+ of EBIT from AI, significant

6%

|  | Value (%) |
| --- | --- |
| McKinsey: any EBIT impact from AI | 37 |
| Nasscom-KPMG: ROI across multiple AI use cases | 24 |
| Deloitte: revenue growth from AI | 20 |
| PwC: AI both grew revenue and cut costs | 12 |
| McKinsey: 5%+ of EBIT from AI, significant | 6 |

Source: McKinsey via The Register (25 Aug 2026); PwC and Deloitte via The Next Web (1 May 2026); Nasscom-KPMG (Jul 2026) · as of 25 August 2026 · each survey asks a different question of a different sample; read the bars side by side, not as one series

The gap between how productive people feel and what shows up in the accounts appears in every one of these surveys. Deloitte has 66% for productivity against 20% for revenue; McKinsey has 80% against 37%.[1][3] The likeliest reason is accounting, not technology. An hour saved by one employee is not profit until the firm either stops paying for that hour or uses it to sell more, and most deployments are not designed to do either.

Our view

Our view: the stalled 6% is not proof that AI does not work. It shows that most firms have bought a tool and not changed a process. Surveys that ask whether AI “helps” will keep finding yes; surveys that ask about EBIT will keep finding few firms until the work itself is redesigned.

## What the few do differently: redesign first, then measure in business terms

McKinsey’s November 2025 State of AI reached the same conclusion, as The Next Web summarised it: what set the high performers apart was “not better technology choices but better organisational practices”. They redesigned workflows, scaled faster, built AI into business processes, tracked KPIs for their AI solutions, and had senior leaders visibly committed to the work.[3] A separate McKinsey report covered by the same outlet described most current AI as “tools that accelerate existing work” while they “largely preserve underlying workflows”.[3]

Indian industry bodies say much the same. The Nasscom-KPMG whitepaper argues that workflows and decision rights “must be redesigned from first principles, rather than simply automating broken processes to make them move faster”, and that the advantage goes to firms that reorganise, not to those running the most pilots.[4] Nasscom’s June study of 505 enterprises in more than 12 regions found that its AI-first group mostly tracks AI value with the enterprise KPIs it already uses, on growth, monetisation, resilience and governance, rather than with new AI metrics.[5]

**Figure: The difference is not the model: the few redesign the work and judge AI on the numbers the business already reports**

*Practices of AI high performers against the typical deployment, as described by McKinsey (Nov 2025) and Nasscom (Jun and Jul 2026)*

The typical deployment

- AI added to an existing workflow, which stays as it was
- Success measured by usage, licences issued and how people say they feel
- Many pilots, few scaled
- Owned by IT or an innovation team
- AI spend tracked as a general cost of doing business

The 6% (high performers)

- The workflow is redesigned around AI before scaling
- KPIs tracked for each AI solution, in existing business terms
- Scaled faster, built into business processes
- Senior leaders visibly committed
- Return attributed to organisational change, not to the tool

Source: McKinsey via The Next Web (1 May 2026); McKinsey via The Register (25 Aug 2026); Nasscom (22 Jun 2026); Nasscom-KPMG (Jul 2026) · as of 25 August 2026

## Before you renew: six tests for an AI deployment

Most AI is bought as a subscription, so the moment to apply all this is the renewal. The seat price is the easy part. Microsoft 365 Copilot, for example, is listed on TechBag’s intel page at about US$30 a user a month as an add-on to an existing Microsoft 365 licence.[7] Agents increasingly add usage-based charges on top, which is why one respondent in five already says operating costs constrain their use.[1]

TechBag countTechBag arithmetic: what a renewal has to prove

At about US$30 a seat a month, 200 seats cost US$72,000 a year before tax (200 × 30 × 12), which is US$360 per seat per year that the deployment has to earn back in money saved or revenue won, not in minutes people say they saved. The list price is from our Copilot page; the multiplication is ours, not a quote.[7]

**Figure: An AI renewal should pass six tests, and the first one is set before the tool is switched on**

*Measuring an AI deployment before renewal, TechBag's sequence built from the high-performer practices, August 2026*

Step 1 of 6

1. Name the workflow and its number

One process, one business measure you already report: cost per case, cycle time, error or rework rate, days of working capital. Not “productivity”.

Source: TechBag analysis of McKinsey via The Register (25 Aug 2026) and The Next Web (1 May 2026); Nasscom (22 Jun 2026) · as of 26 August 2026

None of this needs a consultant. It needs the measure chosen before the tool goes live, which is the step most deployments skip. The older pattern of paying for software nobody uses is in [our piece on where enterprise software ROI leaks](https://www.thetechbag.com/blog/maximizing-roi-enterprise-software-best-practices-strategies), and what an AI seat costs a small firm is in [our note on SMB AI seat prices](https://www.thetechbag.com/blog/unleashing-power-ai-smbs-can-compete-big).

If you are buying this in India

The coverage of McKinsey’s 2026 survey that we could read gives no country breakdown, so we have no Indian cut of the 37% or the 6%. The Indian figures come from Nasscom, and they tell the same story. Its January 2026 study of the Indian technology industry found 20% to 40% of work across functions already done through AI, more than 95% of firms using or planning to use AI agents, customer satisfaction and employee productivity as the top KPIs for AI use, and more than 55% citing incomplete or low-quality AI output, so people still have to check the work.[6] The Nasscom-KPMG whitepaper of July 2026 adds that 55% of technology executives struggle to show AI’s value to stakeholders, and that of India’s 9,20,000 AI professionals only 28% have core AI depth.[4]

Those KPIs are the problem the Stepper above addresses: employee productivity is the measure that does not reach EBIT. For an Indian buyer, two practical points apply at renewal. An AI seat billed in US dollars from abroad to a GST-registered business carries 18% GST, paid by the buyer under reverse charge, so the US$72,000 in the example above comes with a further 18% of tax to pay and then claim back as input credit where the business is eligible.[8] And measurement creates records: if the logs you keep to prove value contain customers’ or employees’ personal data, they fall under the DPDP Act, whose Rules were notified on 13 November 2025 with the substantive duties applying from 13 May 2027.[9] Collect the business number you need, not every prompt.

How we did this

Figures as of 26 August 2026

McKinsey’s report page did not load for us, so every McKinsey 2026 figure is as The Register reported it on 25 August 2026, including McKinsey’s statement added that day. McKinsey’s November 2025 high-performer practices, the PwC 2026 Global CEO Survey and Deloitte’s 2026 State of AI in the Enterprise are cited through The Next Web of 1 May 2026; we did not read the PwC or Deloitte reports themselves. The Nasscom figures are from Nasscom’s own report pages; the whitepaper page does not say which survey behind its 24% and 55% figures. The bar chart puts different questions side by side and is not a series. The 2025 McKinsey EBIT share is reported only as “about the same”, so it is not charted. The Copilot price is TechBag’s intel page as it stood in August 2026; the 200-seat sum is our arithmetic. Gartner’s forecasts on agent projects were left out because we could not open the press pages.

Spotted an error? Write to us; corrections are published at the foot of the piece.

Sources

1. [The Register (Brandon Vigliarolo): McKinsey says enterprise AI is finally 'on the road to ROI' (reports McKinsey's State of AI 2026 survey, 1,719 respondents, and McKinsey's statement)](https://www.theregister.com/ai-and-ml/2026/08/25/mckinsey-says-enterprise-ai-is-finally-on-the-road-to-roi/5292388) — 25 August 2026
2. [McKinsey & Company, QuantumBlack: The state of AI in 2026](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai) — 25 August 2026 The page did not load for us; figures are cited through source 1.
3. [The Next Web (Allison Steffens Herrera): McKinsey's new AI report argues the productivity payoff is real but conditional (covers McKinsey's November 2025 State of AI, PwC's 2026 Global CEO Survey and Deloitte's 2026 State of AI in the Enterprise)](https://thenextweb.com/news/mckinsey-ai-productivity-paradox-enterprise-roi-capex) — 1 May 2026
4. [Nasscom and KPMG: Reorganise Or Fall Behind: The Real Race In The AI Decade (whitepaper key findings)](https://nasscom.in/knowledge-center/publications/reorganise-or-fall-behind-real-race-ai-decade) — 30 July 2026
5. [Nasscom: AI-First Enterprises: Trends and Emerging Directions (survey of 505 global enterprises, key findings)](https://nasscom.in/knowledge-center/publications/ai-first-enterprises-trends-and-emerging-directions) — 22 June 2026
6. [Nasscom: WORK REIMAGINED: The Rise of Human-AI Collaboration (Indian technology industry, key findings)](https://nasscom.in/knowledge-center/publications/work-reimagined-rise-human-ai-collaboration) — 13 January 2026
7. [TechBag, Microsoft Copilot intel page: Microsoft 365 Copilot about US$30 a user a month as an add-on](https://www.thetechbag.com/microsoft/microsoft-copilot) — mid-2026
8. [India Briefing: Digital services in India: OIDAR, GST applicability and reverse charge](https://www.india-briefing.com/news/tax-digital-services-oidar-in-india-gst-applicability-and-compliance-22465.html) — updated 29 September 2023
9. [MeitY, Digital Personal Data Protection Act, 2023 and the DPDP Rules, 2025 (notified 13 November 2025; substantive duties from 13 May 2027)](https://www.meity.gov.in/data-protection-framework) — 13 November 2025

Related on TechBag: the [Microsoft Copilot intel page](https://www.thetechbag.com/microsoft/microsoft-copilot) sets out what the add-on includes; [where enterprise software ROI leaks](https://www.thetechbag.com/blog/maximizing-roi-enterprise-software-best-practices-strategies) covers idle seats; [what an SMB’s AI seat costs](https://www.thetechbag.com/blog/unleashing-power-ai-smbs-can-compete-big) compares the bundles.

**Disclosure.** TechBag is a software reseller: it earns a margin when a reader buys through it, including from vendors named here. The analysis is TechBag’s own; no vendor paid for or saw this piece before publication. Facts are labelled reported, TechBag count or our view.

Go deeper on TechBag

- [Microsoft Copilot intel page](https://www.thetechbag.com/microsoft/microsoft-copilot) — what the US$30-a-seat add-on includes and how it is licensed
- [Where enterprise software ROI leaks](https://www.thetechbag.com/blog/maximizing-roi-enterprise-software-best-practices-strategies) — idle seats and the value gap, before AI
- [What an SMB's AI seat costs](https://www.thetechbag.com/blog/unleashing-power-ai-smbs-can-compete-big) — free tiers to US$21 a month against the US$30 enterprise list