A full chapter from The 2026 AI Strategy, published open. The complete ebook is free; this chapter stands on its own.
Follow the money first. Enterprise spending on generative AI more than tripled in a single year, from $11.5 billion in 2024 to $37 billion in 2025, and is set to roughly double again in 2026 (Menlo Ventures; BCG). Roughly seven in ten CEOs call AI a top investment priority, and by industry estimates half believe their own standing depends on getting it right (KPMG; BCG).
Now follow the returns. 56% of CEOs report no financial benefit from AI so far, and only 12% saw both revenue and cost gains (PwC, 2026). An MIT study of enterprise deployments found that roughly 95% of generative AI pilots delivered no measurable P&L return (MIT Project NANDA, 2025). Both numbers are from the same period as the record spend.
McKinsey's 2025 State of AI research contains the explanation, sitting in plain sight. 88% of organizations now use AI, but only 6% clear McKinsey's high-performer threshold: significant value, with at least 5% of EBIT attributable to AI. Another 39% report some enterprise-level impact, usually below that bar. And among the strongest separators of the high performers: they are nearly three times as likely as everyone else to have fundamentally redesigned their workflows around the technology (McKinsey, 2025).
Read those numbers together and the paradox dissolves. The gap is not whether companies use AI; nearly all do. The gap is whether AI has become material to the economics, and the companies where it has are overwhelmingly the ones that redesigned the work. The companies that bought tools got tools.
One caution to carry through everything that follows, because the styleguide of this ebook is honesty: this research base is young, survey-built, and definition-sensitive, so the exact numbers will move. The pattern is already stable, and it is the pattern that matters: adoption is easier than value capture, and value capture depends on redesigned work, trusted data, governance, and ownership.
This matches what I see inside mid-market operations. The copilot makes an analyst faster at one step of a twenty-step process. The other nineteen steps still involve a person pulling data from one system, keying it into another, chasing an approval, and reconciling the result by hand. The individual got faster. The process did not change. The company saw nothing, because the cost of the process was never the typing speed. It was the process.
Adoption is not transformation. That single distinction explains nearly every number above, and it sets the agenda for the rest of this ebook. If the return comes from redesigning how value moves through the business, then the first requirement of an AI strategy is not a tool list. It is a map of how value moves through the business. That map is the next chapter.
One thing this ebook assumes about you: you are already doing AI. Nobody running a mid-market company in 2026 is starting from zero; 91% of middle market firms now use generative AI (RSM Middle Market AI Survey, 2025), and you have the copilots, a pilot or two, and a tools line in the budget to show for it. So the question in front of you is not whether to do AI. It is which of the three available responses you are currently running, because only one of them compounds.
Response one: bolt on features. Buy the copilots, add the chatbot, let each department pick its tools. This is the majority position, and it is where the 56% live. The features ship. The business does not change.
Response two: run isolated pilots. Pick a use case, stand up a proof of concept, evaluate. More serious, and still where, by industry estimates, more than eighty percent of mid-market AI projects end. The pilot demonstrates that AI can do the work. It never puts AI to work.
Response three: rebuild the workflow. Redesign how a piece of the business runs so AI does the operating work and people direct it, measured against a baseline. This response holds because it is not bolted onto anything. It is the new architecture, and it is where the 6% live.
Most readers will recognize themselves in response one or two. That is not a criticism; it is the honest starting position of nearly everyone, and the people who built those pilots genuinely wanted them to work. The rest of this ebook is the path from where you are to response three.