Only 6% of Companies Are Creating Real Value With AI. Here Is What They Do Differently

AI is now the most discussed topic in corporate boardrooms. In a July 2026 analysis of 4,800 quarterly earnings calls, Boston Consulting Group found that AI was the single most mentioned topic by CEOs globally. A separate Financial Times analysis found that 75% of S&P 500 companies referenced AI on earnings calls over the past year, and 87% of those mentions were entirely positive.

Now here is the number that should stop every executive mid-sentence: according to that same BCG study, only 6% of large US public companies qualify as genuine AI adoption leaders.

Six percent. After three years of foundation models, hundreds of billions in capital expenditure, and thousands of pilot programs, the gap between what companies say about AI and what they actually do with it has never been wider.

I have spent 25 years delivering enterprise IT and automation programs across more than 20 countries. I have watched this exact pattern repeat through every technology cycle: ERP, cloud, RPA, and now AI. The talk always arrives years before the value. What makes this BCG study worth your attention is that it finally puts hard, outside-in numbers on the gap, and it tells us precisely where the value hides.

Why this study is different

Most AI adoption evidence comes from two flawed sources. Corporate communications reflect what leaders think investors want to hear. Surveys reflect what companies say about themselves. Both are theater.

The BCG Institute, working with researcher Cindy Laura Stahl of the University of St. Gallen, built an outside-in adoption score instead. It draws on labor market data from Revelio Labs, IT installation data from HG Insights, and natural language analysis of 10-K filings and earnings calls. No self-reporting. No survey bias.

The score rests on three pillars:

1. AI tech: the tools a company actually uses and the stack that supports bespoke development
2. AI talent: the share of employees in AI-specific roles and the prevalence of AI skills across the broader workforce
3. AI deployment: how many functions have real use cases, and how deep those use cases run, from discussion-only to enterprise-wide production

Applied to more than 600 US public companies above $5 billion in market capitalization, the score sorts firms into four tiers: nascent, emerging, active, and leading. The distribution is brutal. Nascent and emerging firms, the laggards, make up 68% of the sample. Active firms account for 26%. Leaders: 6%.

Finding one: AI value is a step change, not a curve

The leading 6% delivered industry-adjusted total shareholder returns 9.3 percentage points above the sample median over the past three years. Laggards ran 1.7 points below.

But the most important number in the entire study sits in the middle. The active tier, the 26% of companies just below the leaders, companies that have invested, deployed, and built real capability, captured a premium of just 0.6 percentage points. Essentially nothing.

Read that again. Being good at AI pays almost nothing. Being at the very top pays enormously. Value does not accrue progressively along the adoption curve. It arrives as a step change at the summit.

For executives, this reframes the entire investment question. Partial adoption is not a partial win. It is a cost center waiting for the final push. The companies stuck at “active” have absorbed most of the investment pain and captured almost none of the return. The only rational strategies are to cross the final gap or to be honest that you are funding an option, not a result.

Finding two: the returns are fundamentals, not hype

The obvious objection: surely this is investor exuberance. AI-branded stocks get bid up, multiples expand, and the “value” is a narrative premium waiting to deflate.

BCG decomposed the TSR outperformance of leaders versus laggards to test exactly this. The result is the opposite of the hype story. The outperformance comes almost entirely from fundamentals: 10 percentage points from revenue growth and 6 percentage points from net income margin expansion, both industry-adjusted. P/E multiple expansion, the channel through which pure hype would show up, contributes essentially zero.

Leaders even show a negative 5 percentage point TSR contribution from cash effects, consistent with equity issuance and reduced dividends. They are not extracting cash. They are reinvesting it into the machine.

And this is not a shovel-seller effect. The figures are industry-adjusted, meaning technology leaders are measured against the technology median, not against the broader market. Breakout performers exist across industries.

One more detail deserves attention. Underneath the financial results sits a productivity engine: revenue per employee at leaders is growing 4 percentage points faster than at laggards on an industry-adjusted basis, a gap that opened with the release of foundation models in late 2022 and has widened since. This also resolves the economists’ puzzle about AI being invisible in aggregate productivity data. When real gains are concentrated in 6% of firms, averaging them with the other 94% washes the signal out. The productivity revolution is real. It is just not evenly distributed.

Finding three: leaders use AI to grow, not to cut

The dominant public narrative frames AI as a headcount reduction tool, and the recent wave of AI-referencing layoff announcements reinforces it. The data from the leading 6% tells a different story: these firms are growing headcount at a compound annual rate 3 percentage points above the sample median.

BCG maps three distinct value paths among leaders:

Save (10% of leaders). Automate back-office processes and expand margins. IBM’s AskHR agent now handles over 94% of employee requests and contributed to a 40% reduction in HR operating costs over four years. Real value, but note the share: this is the least common path among leaders. Efficiency is an entry point that builds organizational muscle and proves the case. It is not an end state, because doing the same things faster with tools everyone can buy yields no durable advantage.

Scale (59% of leaders). The dominant path. Make each unit of work more productive, then use that capacity to serve more customers, enter adjacent markets, and reach segments that were previously uneconomic. Salesforce reports its AI-assisted service agents spend 20% less time on routine cases, freeing roughly four hours per week for higher-value work, projected to lift upsell revenue by 15%. Margin expansion and revenue growth, simultaneously.

Innovate (21% of leaders). Build offerings that would not exist without AI. Meta’s generative marketing tools tailor creative assets to both product and viewer. Moody’s is converting proprietary credit and risk data into new analytical products. This path demands an appetite for temporary margin trade-offs and revenue streams that do not yet exist.

The common thread: productivity gains are reinvested, not harvested as cuts. This matches BCG’s broader labor market research finding that AI’s main effect on work is augmentation rather than substitution. When productivity rises in roles where demand can expand, output grows, and so does headcount.

The journey: three transitions, and the one that matters most

How do companies actually reach the leading tier? BCG traced how the three pillar scores evolve across tiers and found a consistent sequence.

Transition 1: Expand the toolkit. Move from a handful of general-purpose tools to a broader, specialized ecosystem accessible across functions. Companies here are buying, not building. It is a necessary learning phase, and a dangerous one: many firms get stuck running an ever-expanding portfolio of disconnected pilots.

Transition 2: Go broad and deep on deployment. Convert pilots into production-grade systems in a few functions first, build reusable platforms and playbooks, then scale outward as the marginal cost of each new deployment falls. Across this transition, the share of firms that are simultaneously broad (eight or more functions) and deep (at-scale deployment) jumps from roughly 12% to over 50%. Walmart went deep on demand forecasting first, then fed that foundation downstream into inventory, warehouse robotics, store operations, and last-mile delivery.

Transition 3: Cross the talent gap. This is the finding that should reorganize your 2027 budget. Between the active and leading tiers, tech and deployment scores barely move. The talent score nearly triples.

And “talent” does not mean “hire more AI developers.” Leaders build breadth and depth simultaneously. At leaders, 13% of all employees carry AI-related skills, versus 1% at laggards. AI-specific roles reach 3.5% of the workforce, versus 0.1%. Distributed fluency means teams across the business can spot where AI changes the economics of their function and co-design solutions, generating a richer use case pipeline than any central team could. Dedicated specialists then industrialize the best of them into proprietary, enterprise-grade capabilities.

None of this happens organically. It requires rethinking workflows end-to-end rather than bolting AI onto existing processes, joint ownership between business and IT, and a central team that curates and governs rather than builds everything itself. BCG summarizes it as the 10-20-70 rule: 10% of the effort is technology, 20% is algorithms and data, 70% is people, process, and organizational change.

The strategic logic is simple. Tools have commoditized. Every company can buy the same products from the same vendors. What cannot be bought is the organizational capability to deploy them into the specific economics of your business. That is the moat.

The honest caveat

One finding keeps this study intellectually honest. About 10% of adoption leaders are failing anyway, with declining margins and growth. Closer inspection shows they are trapped by fundamental business model problems AI cannot fix: commoditized core offerings, or legacy models being displaced by digitally native disruptors.

AI amplifies a strong strategy. It does not substitute for one.

What I take from this

There is a final irony in the data. Merely talking about AI on earnings calls buys a small valuation premium, roughly 1 percentage point of P/E-driven TSR, even for laggards. Investors reward the signal. But the real returns, the 9.3 point premium built on revenue, margin, and productivity, belong exclusively to the companies that execute.

The market pays a little for the story and a lot for the outcome. So the question for every leadership team heading into 2027 planning is not whether you are “doing AI.” 26% of large companies are actively using AI and earning nothing from it. The question is whether you are willing to fund the unglamorous 70%: workforce fluency, workflow redesign, and the ownership model between business and IT that turns tools into proprietary capabilities.

Talk is cheap. The 6% are proof that execution is not.

What is keeping your organization in the active tier? I would genuinely like to hear where the friction sits: tools, deployment, or talent.

---

Source: Ulrich Pidun, Adam Job, Tania Wang, Matthew Molloy, Michael Grebe, Marc Roman Franke, and Viacheslav Romanov, “AI Talk Is Cheap. Value Creation Is Rare.” BCG Institute, July 9, 2026.