Adoption Is Not Value

Almost every organization now uses artificial intelligence. Far fewer can show what it earned them. That gap — between adoption and value — is the defining management problem of 2026, and it is not a technology problem.

Two years ago, the question executives asked me was whether artificial intelligence was real. That question is settled. The one that has replaced it is harder and far more useful: why does an organization that has deployed AI across a dozen functions still struggle to point to a line in its accounts that moved because of it?

This article sets out what the most credible 2026 evidence actually shows, why so many initiatives stall short of impact, and what we believe leadership teams should do differently. It is deliberately unglamorous. The organizations getting real returns are not the ones with the most impressive demonstrations; they are the ones that treated AI as an operating-model question rather than a procurement one.

It has become fashionable to argue that progress has stalled. The data does not support it. Stanford’s 2026 AI Index — the closest thing the field has to an independent annual census — records that performance on SWE-bench Verified, a benchmark requiring models to resolve genuine software issues, rose from roughly 60 percent to near 100 percent in a single year. On Humanity’s Last Exam, a deliberately punishing expert-authored benchmark, accuracy climbed from 8.8 percent to 38.3 percent, with leading models passing 50 percent by April 2026.

Capital has followed. Global corporate AI investment reached $581.7 billion, roughly a 130 percent year-on-year increase, with generative AI attracting $170.9 billion. Diffusion has been faster still: generative AI reached 53 percent population-level adoption within three years — quicker than either the personal computer or the internet managed at comparable stages.

Two caveats belong here, and both matter for planning. First, benchmark performance is not operational reliability; a model that excels on curated tasks can still fail in a messy production workflow. Second, the same Index documents a rise in recorded AI incidents and a persistent lag in governance capacity. Capability is compounding faster than the institutions meant to supervise it.

McKinsey’s State of AI survey finds that 88 percent of respondents report their organization regularly using AI in at least one business function, and 72 percent report using generative AI. On the surface, this is a technology that has won.

Then comes the number that should occupy any board’s attention: only 39 percent report any EBIT impact attributable to AI at enterprise level — and among those, most attribute less than 5 percent of EBIT to it. Nearly two-thirds of organizations have not begun scaling beyond isolated use cases. Enthusiasm has been converted into activity, not yet into earnings.

The most-cited counterpart to that finding is MIT’s GenAI Divide study, which concluded that roughly 95 percent of enterprise generative AI pilots produced no measurable profit-and-loss impact. The figure went viral, and it deserves a caveat usually omitted: the sample was modest, and several analysts have questioned whether such a headline rate can be generalised. We cite it not as a precise measurement but because its diagnosis is corroborated elsewhere — pilots fail for organizational reasons, not because the models are inadequate.

The workflow was never redesigned

AI was added to an existing process rather than used to reconsider it. McKinsey identifies fundamental workflow redesign as the single strongest correlate of EBIT impact. A copilot bolted onto an unchanged process produces a faster version of the same result — and rarely a better economic one.

The data foundations were assumed, not verified

Models inherit the quality of what they are given. Fragmented, undocumented or poorly governed data does not merely degrade output; it makes failure difficult to diagnose, which is worse. Most of the difficult work in a successful AI programme happens before any model is selected.

Budget followed visibility rather than return

Spending concentrates in sales and marketing, where results are demonstrable to leadership, while the more promising returns often sit in back-office operations, finance and service delivery — less photogenic, more measurable.

Governance arrived last, if at all

This is the failure we see most often and the one with the longest tail. MIT observed that employees in more than 90 percent of the firms studied use personal AI tools regardless of whether official programmes succeed. Shadow AI is not a discipline problem; it is a signal that sanctioned tools are not doing the job. Left unmanaged, it moves confidential data outside every control the organization has built. Notably, McKinsey’s 2026 trust-maturity work finds that organizations investing seriously in responsible AI are more likely to report EBIT impact above 5 percent — governance correlates with value rather than competing with it.

For any organization operating in Europe, the abstract debate about AI ethics has become a compliance calendar. The EU AI Act entered into force in August 2024. Its implementation proved harder than expected, and in 2026 the European Parliament and Council adopted the Digital Omnibus, which defers the heaviest obligations while leaving the framework intact. The deferral has been widely misread as a reprieve. It is not.

In short: transparency duties are imminent, and the extra runway on high-risk systems is time to prepare, not time to wait. Organizations that begin building AI inventories, documentation and human-oversight mechanisms now will absorb the 2027 obligations as routine. Those that wait will meet them as a crisis.

Discussion of AI is dominated by North America, Europe and China. That framing misses where some of the steepest gains are available. The African Development Bank projects that inclusive AI deployment could add up to $1 trillion to African GDP by 2035 — close to a third of the continent’s current output — with agriculture, wholesale and retail, manufacturing, financial inclusion and health capturing the majority of the gain.

The constraints are structural rather than conceptual: compute access, connectivity, data availability and, above all, skills. The African Union’s Continental AI Strategy, adopted in 2024, is explicit that capability-building must precede large-scale deployment.

One recent example illustrates the whole argument better than any statistic. In early 2026, Ghana’s revenue authority deployed an AI system at Tema Port to analyze import declarations and flag anomalies. On adoption metrics it was an unambiguous success: customs receipts rose sharply within weeks. On governance metrics it was not. Traders could not understand how valuations were reached, and the appeals mechanism was too slow to function — culminating in industrial action by freight associations. The model worked. The system around it did not.

Six moves, in the order we recommend taking them.

The organizations that will look prescient in three years are not those that adopted AI earliest. They are those that were most disciplined about where it was allowed to touch the business, most rigorous about the data beneath it, and most serious about governing it before they were required to. The technology has crossed from novelty into infrastructure. Infrastructure is judged by reliability, accountability and trust — which is precisely the standard we apply to everything else that a business depends on.

At Auvantyx, this is the work: helping organizations translate capability into outcomes they can defend to a board, an auditor and a regulator. If any part of this reflects a question you are currently sitting with, I would welcome the conversation.

About the author

Loick Moukouri

Founder, Auvantyx

Loick Moukouri is the Founder of Auvantyx. His background combines accounting and finance with cybersecurity, cloud and digital technology. He leads the firm’s vision, service offering and specialist network, helping organizations translate business priorities into practical technology, security and risk decisions.

Contact contact@auvantyx.com