Scaling rarely fails because of technology. It fails because of assumptions that were never tested in the pilot.
Part 1 showed that companies invest in AI in the wrong order: too much technology, too little process redesign and capability building. Part 2 answers the follow-up question. Why does a working pilot not hold once it has to grow?
The finding
Between January and April 2026, Gartner surveyed 1,303 leaders at companies with at least USD 50 million in annual revenue. Only 22 percent have successfully scaled AI across multiple business units or adopted an AI-first approach. At the same time, 85 percent plan to increase AI spending in 2026.
The figure needs a careful reading. Gartner combines two groups and measures the breadth of use, not the contribution to earnings. McKinsey shows the other side: about 88 percent of organizations use AI in at least one function, but only about 6 percent qualify as AI high performers. For them, AI contributes at least 5 percent of EBIT, and they report a significant value contribution. Usage is standard. Impact is the exception.
Gap 1: Scaling changes the economics
In a McKinsey survey from May 2026, 93 percent of respondents exceeded their AI budgets. Moving from isolated use cases to broad adoption raises spend by nearly a factor of four. Consumption is hard to plan: for the same task, token usage can vary by up to 30 times. A business case built on pilot data will not carry scale.
Gap 2: Governance is missing
About 11 percent of the organizations in Gartner’s survey do not know what their functions spent on AI in 2025. According to McKinsey, only 20 to 25 percent have mature AI FinOps practices. Gartner describes the successful scalers as follows: they track returns continuously, manage initiatives as a portfolio and stop projects that fall behind plan. Those who do not know what they spend and why cannot decide what to continue.
Gap 3: The operating model stays the same
McKinsey’s high performers are nearly three times as likely to have fundamentally redesigned their workflows as other companies. They are also three times as likely to strongly agree that senior leaders take ownership of AI initiatives. Our assessment: the inverted investment pattern from Part 1 is the cause, and the scaling gap is its consequence.
Why a working use case does not automatically carry over
Introducing AI in a new organization or market does not mean copying a proven model. Three assumptions shift:
- The usage baseline. Volume, user groups and workflows differ, so the cost calculation from the home operation no longer holds.
- Ownership. Where business units buy AI independently, consumption grows without anyone being accountable for the outcome.
- The framework conditions. Governance, data protection and co-determination are part of scaling in every organization. What matters is that they are settled before the investment, not after.
What must hold before the next investment decision
- Cost per outcome, not cost per unit of compute, calculated on a realistic usage baseline.
- A decision rule: who stops or extends an initiative, and based on which metric?
- A named owner for each business unit, and the processes that change there.
The decisive question is not whether an AI use case can be scaled. It is whether its economics, governance and accountability still hold at larger scale.
If you are facing such a scaling decision, we are happy to discuss which assumptions need to be solid before the next investment. Book a strategy call
Sources: Gartner, press release of 1 September 2026 (1,303 respondents, January to April 2026). McKinsey Quarterly, “The cost of intelligence” (20 July 2026), Enterprise AI FinOps Survey, May 2026, 120 participants, 75 qualified respondents. McKinsey, “The state of AI in 2025” (November 2025). AI high performer: at least 5 percent of EBIT attributable to AI and a significant reported value contribution.
