Deloitte China and McKinsey have successively updated their respective AI enterprise application reports in recent days. When the two reports are viewed together, the figures are quite glaring.
Deloitte says: 9% of AI projects generate negative returns, and only 4% of enterprises have truly achieved their transformation goals.
McKinsey says: 88% of enterprises have normalized the use of AI in at least one function, but only 6% have become "high-performing enterprises"—that is, those that truly make money from AI, and make a lot of it.
Deloitte looks at China, McKinsey looks at the world, and both reports point to the same conclusion: AI is being used with great fanfare, but the real monetary returns belong only to a very small number of companies.
Deloitte Index: ROI is generally overestimated
The "2026 Enterprise AI Application Index" jointly released by Deloitte China and the University of Hong Kong surveyed over 3,200 enterprise IT leaders from 24 countries, and several figures are quite straightforward:
23% Use AI normally, but business transformation has not yet begun—meaning "used it, but got nothing in return"9% AI projects ultimately generate negative returns due to surging hidden costs and business disruptions—not only didn't make money, but also lost money4% Truly achieved AI transformation goals, with substantial changes in business models49% AI projects lack a clear definition of success, meaning "did it, but don't know if it counts as success"
Deloitte directly pointed out in the report: Enterprises' expectations for the return on investment of AI projects are generally higher than the actual delivery level. Part of this is "performative AI"—deploying a few chatbots, creating a few AI-generated data dashboards, and publicly claiming to have "completed intelligent transformation," yet the contribution rate to EBIT is less than 1%.
Such enterprises account for 23% of Deloitte's sample.
What McKinsey's "6% High-Performance Companies" Look Like
McKinsey's definition of a "high-performance enterprise" is strict: EBIT increases by more than 5% due to AI, and AI has been used to create "significant" value. By this standard, only 6% of enterprises globally meet the criteria.
But these 6% of enterprises do things that are fundamentally different from the remaining 94%:
| Dimension | High-performance enterprises (6%) | Other enterprises (94%) |
|---|---|---|
| AI Disruptive Transformation Plan | more than three times that of other enterprises | Most remain at the pilot stage |
| Fundamental restructuring of the workflow | 3 times that of other institutions | Superimpose AI on the original process without changing the process |
| Large-scale deployment of intelligent agents | more than three times that of other organizations | The vast majority are still in the experimental stage |
| Number of AI use cases | is twice that of other enterprises | Scattered across a few functions, not horizontally expanded |
| Digital budgets are being directed toward AI | Over 1/3 of enterprises allocate over 20% of their budget to AI | Only 1/3 of enterprises have reached this level |
| AI has been/is being deployed at scale | 74% | 33% |
The logic behind this table is: high-performing enterprises do not simply "use" more AI tools; instead, they reorganize business processes to embed AI into the core of value creation, rather than attaching it to the surface of existing workflows.
McKinsey specifically highlighted a detail: high-performance companies more frequently conduct "fundamental restructuring" of their work processes, rather than layering AI capabilities on top of existing processes. In plain language, this means — you cannot simply slap a layer of AI onto outdated processes and expect it to become smart.
Special Position of Chinese Enterprises
McKinsey's report specifically lists data for mainland China, with several figures differing from the global average:
83% of Chinese enterprises regularly use generative AI in at least one function (global average: 78%); 45% of Chinese enterprises have achieved large-scale or full deployment of AI (global average: 38%); 93% of domestic workplace respondents use AI tools (global average: 58%) — from survey data under the comprehensive implementation of "AI+"
Three numbers point in the same direction: Chinese enterprises lead the world in AI penetration, but the proportion of "large-scale deployment" is only 7 percentage points higher than the global average. In other words, many people use it, but not necessarily many have it running smoothly.
This is not contradictory to Deloitte's index that "only 4% have truly succeeded in transformation" — many Chinese companies have adopted AI tools but have not yet completed the restructuring of their business processes. In McKinsey's words, there is still a significant hurdle between "routinely using AI in at least one function" and "achieving large-scale AI deployment."
Typical Symptoms of "Performative AI"
Reading the two reports together, there are several typical symptoms of "performative AI" — see if your company has fallen into any of them:
- Deploying a chatbot counts as digital transformation: Customer service, HR, and IT helpdesk each get a conversational bot, followed by a press release announcing "fully embracing AI." According to Deloitte's data, such enterprises account for a significant proportion.
- AI projects lack a clear definition of success: 49% of enterprises are unclear about what constitutes "AI project success"—is it user engagement? Cost reduction? Or revenue growth? Without a definition, there is no way to measure; without measurement, there is no way to optimize.
- "Pasting" AI onto the original process: McKinsey says 94% of non-high-performing companies are doing this. The right approach is to redesign the process with AI, rather than layering AI tools on top of the old process.
- Hidden costs are not calculated: Deloitte points out that 9% of AI projects ultimately yield negative returns, mainly because hidden costs such as data cleaning, model fine-tuning, personnel training, and system integration are severely underestimated in the early stages of the project.
Two reports share a common conclusion: the biggest bottleneck for AI moving from "pilot" to "scale" is not technological maturity, but organizational execution capability. The technology is ready, but internal processes, talent, and incentive mechanisms within enterprises have yet to catch up.
A one-sentence suggestion for the CIO
If you are planning your AI budget for the next year, data from Deloitte and McKinsey provides a very direct reference framework:
- First ask, "If this AI project succeeds, how much can EBIT increase?" and then choose the technical route;
- If the answer is "uncertain" or "less than 1%," then this project is likely performative in nature;
- The practice of high-performing enterprises is: select 1-2 core business processes, completely reconstruct them with AI, and once they work, expand horizontally;
- Don't launch one AI pilot in each of 10 functions—that's what 94% of companies do, and it's why most businesses' AI transformations stall.
At the end of the day: the endgame of AI transformation is not "we used AI," but "our business model has fundamentally changed because of AI." Deloitte's 4% and McKinsey's 6% are the companies that have reached this point. The remaining 90%+ are still on the way.
Source: Deloitte China × University of Hong Kong "2026 Enterprise AI Application Index" (May 2026) McKinsey "2026 AI Application Status Survey" (Chinese version, updated May 2026) "China Business Journal" (May 23, 2026)
