Only 13% of Companies Are Scaling AI as Planned — While Banks Race Ahead Anyway

A study of 1,050 C-suite and senior leaders across 13 countries, published October 1 by consultancy BearingPoint, puts a hard number on a gap enterprises have been dancing around all year: 74% of companies report measurable financial returns from AI, but only 13% say they have scaled it as far as their original business case promised. Less than a third have gotten any AI initiative past the pilot stage at all. The value is real. The scaling mostly isn't.

The study, titled "Scaling AI for Measurable Impact," splits respondents into two tiers that explain most of the gap. Among companies BearingPoint classifies as AI "Leaders," 47% hit their original scaling targets; among "Implementers," just 6% did, pulling the overall average down to 13%. Roughly 40% of all respondents blame legal and regulatory friction for the stall, 34% point to integrating AI with legacy IT systems, and 54% say trustworthy, well-governed data is the real precondition for scaling — not model quality. Geography matters too: 20% of companies in China and 18% in the US report comprehensive AI implementation, versus 8% in Germany.

"AI has crossed an important threshold," said Frederic Gigant, a BearingPoint expert involved in the research, noting that proving AI's value and actually scaling it across an organization are two separate problems entirely.

Two banks, two speeds

What the "Leader" tier looks like in practice showed up in banking within days of the survey's release. On September 24, BNP Paribas signed a five-year deal with Google Cloud to extend agentic AI built on Gemini Enterprise into its Corporate & Institutional Banking arm, which already runs an internal LLM platform, LLM@CIB, for more than 65,000 employees. The first production use case is narrow and deliberate: automating the preparation of corporate credit memos, with research, sales, trading and structuring functions queued up next. BNP Paribas framed it as a multi-cloud, multi-model strategy rather than a single-vendor bet, keeping existing security and governance controls intact rather than rebuilding them around one platform.

"Agentic AI can help our teams handle more complex workflows while maintaining appropriate control," said Charles Holive, Chief AI Officer for BNP Paribas CIB.

M&T Bank offers a slower-building version of the same story. After expanding a Microsoft Copilot pilot from 800 employees to more than 15,000 across 1,800-plus internal applications, the bank reports concrete, if modest, payoffs: roughly six minutes saved per call through automated call-center summarization, and technology outages down more than 80% since 2018 as AI-assisted operations scaled alongside annual system upgrades that have tripled. Both banks spent months on governed data pipelines and narrow use cases before widening scope — exactly the 54%-cited precondition BearingPoint's survey identifies, and exactly what separates the 47% of Leaders from the 6% of Implementers still stuck proving pilots work at all.