A BearingPoint study of 1,050 executives across 13 countries does not say AI scaling is hard. It says outcomes have separated โ by nearly eight to one.
BearingPoint is a consultancy that sells AI implementation services, and a study concluding that most companies cannot scale unaided is commercially convenient for it. The figures below are the study’s own.The study does not publish how it defines Leaders and Implementers, or what share of the sample sits in each group. Nothing below claims that maturity causes the outcome. Nothing here is investment advice.
BearingPoint surveyed 1,050 C-suite executives and senior leaders across 13 countries in August 2026. The study, titled Scaling AI for Measurable Impact, was published in September 2026. The headline figure was the 13 per cent: only 13 per cent of companies have scaled their AI initiatives fully in line with the original business case.
That number is real. But the study itself does not treat it as a single finding about a single population. BearingPoint splits respondents by maturity โ calling the top group Leaders and the lower group Implementers. Among Leaders, 47 per cent scale completely as planned. Among Implementers, the figure is 6 per cent. The 13 per cent sits between the two and describes neither.
One disclosure belongs up front. BearingPoint is a consultancy that sells AI implementation services. A study concluding that most companies cannot scale unaided is commercially convenient for it. That does not make the numbers wrong. But it is context worth carrying throughout. On the study’s own numbers, the returns are not in doubt: 74 per cent of organisations with implemented AI report a measurable top-line or bottom-line impact, and around 40 per cent report both revenuegrowth and cost reduction at once.
The key insight: The 13 per cent is an average taken across a split population. An average of that kind describes nobody in it. Reported as a single fact about AI, it invites the wrong conclusion. Reported correctly, it points at the organisations โ not the technology.
The study does not say what share of the sample sits in each group, or how the two groups are defined. Both matter for reading the gap.
The Structural Read
If one group lands roughly half its programmes and another lands one in sixteen, the midpoint is an artefact of the mix. It is not a fact about the technology.
The barriers named in the study reinforce this reading. The depth figure moved slowly too: 11 per cent say AI is deeply integrated into operations with measured impact, up from 7 per cent in 2025. And the workforce figure is the one carrying a date. 62 per cent report AI-induced workforce overcapacity of at least 10 per cent today, and 95 per cent of respondents expect that level by 2030 — an expectation reported by surveyed executives, not a forecast by BearingPoint and not one by this publication.
According to Reuters’ reading of the study, about 40 per cent of respondents cite legal regulation as the main barrier and 34 per cent point to integrating AI with existing IT systems. Neither is a model capability limit. Both are properties of the organisation and its environment.
That pattern maps directly onto what the Business Engineer framework calls Harness Theory. The obstacles respondents name are organisational rather than technical — data infrastructure, regulatory readiness, integration architecture. That is a description of what companies report standing in the way, not a demonstration that clearing those things produces the outcome.
The cost-versus-revenue split from Reuters’ reading sharpens the picture further. About 24 per cent report cost savings of at least 10 per cent. About 4 per cent report revenuegrowth at the same threshold. That is roughly six to one toward cost. It is a comparison of two reported shares at the same threshold โ not a decomposition of where value came from โ but the ratio is consistent with early-phase AI deployment, where efficiency gains arrive before growth effects.
The geographic spread from Reuters’ reading is worth noting without overstating. Comprehensive implementation sits at 20 per cent in China, 18 per cent in the United States, and 8 per cent in Germany. BearingPoint’s expert Frederic Gigant is cited in that context. Country-level sample sizes are not published, so the spread should be treated as directional.
Three Implications
IMPLICATION 1 โ FOR EXECUTIVES The cost-to-revenue ratio of roughly six to one suggests most organisations are still in the efficiency phase. Revenue impact at scale โ the harder half of the value equation โ has not arrived for the majority. The 4 per cent reporting revenuegrowth above 10 per cent is a ceiling to benchmark against, not a floor to assume.
IMPLICATION 2 โ FOR STRATEGY TEAMS The barriers named โ regulation and legacy integration โ are solvable infrastructure problems, not technology gaps. On these responses, the constraint companies describe is not model capability. The 54 per cent who flag high-quality, trusted data as critical to scaling are pointing at the same organisational layer. Clearing it is prior work, not parallel work.
IMPLICATION 3 โ FOR HOW YOU READ THIS STUDY BearingPoint sells AI implementation services. A finding that most companies cannot scale unaided is commercially useful for that business. That commercial alignment does not invalidate the data. But every percentage in this study should be read with the sponsor’s incentives in view โ including the 95 per cent of respondents who expect workforce overcapacity of at least 10 per cent by 2030, which is a respondent expectation, not a forecast.
Business Engineer Framework
Harness Theory
The BearingPoint data is a field test of Harness Theory in practice. The Leaders group is not distinguished by model access in anything the study reports. What separates the two groups, on the study’s own account, is organisational: data, integration, regulatory readiness. Whether clearing those produces the outcome, or whether the grouping already encodes it, the study does not say. Harness Theory describes the gap of eight to one and why closing it is an organisational problem, not a technical one. The Map of AI shows where each layer of that infrastructure sits and which companies are building it.
The BearingPoint study does not show that AI scaling is broadly failing. It shows that outcomes have separated sharply by organisational maturity โ with a gap that is close to eight to one โ and that the most frequently reported limiting factors are regulatory readiness, legacy integration and data quality, not model capability. The 13 per cent headline is an average across a divided population.
It describes the mix. It does not describe the technology, and it does not describe your organisation. Nothing in this article is investment advice.
The study figures above come from BearingPoint’s own page for Scaling AI for Measurable Impact, September 2026, read directly. The cost-saving and revenue-growth shares, the barrier percentages, the country comparison and the Frederic Gigant attribution come from a Reuters write-up of the same study, read here via syndication rather than on Reuters’ own site.BearingPoint is a consultancy that sells AI implementation services. A study finding that most companies cannot scale their AI programmes unaided is commercially convenient for the firm that published it.
That is not a reason to discard the numbers, but it is a reason to read them as vendor-published research rather than as independent measurement.The maturity split is the study’s own and so are the two group names. What the study does not publish is how Leaders and Implementers are defined, or what share of the 1,050 respondents sits in each group. If the grouping is defined partly by delivery, the gap between 47 and 6 per cent is partly definitional.
Nothing above claims that maturity causes the outcome.The comparison of 24 per cent reporting cost savings of at least 10 per cent against 4 per cent reporting revenuegrowth of that scale is a comparison of two reported shares at the same threshold. It is not a decomposition of where the value came from, and the study does not split value that way.The figure that 95 per cent expect workforce overcapacity of at least 10 per cent by 2030 is what surveyed executives say they expect.
It is not a forecast by BearingPoint and not one by this publication. Also absent: per-country sample sizes, sector mix and margin of error. Nothing above predicts anything, and nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
