MEMO · TO readers evaluating a workshop or a speaker · RE Strategy

Insights Strategy

The productivity gap: why companies must fix their systems before scaling AI

Most organisations use AI frequently but few have scaled it company-wide, and the gap comes down to architectural debt, not appetite.

Terence Kok

About the author

Terence Kok

Enterprise AI Strategist, Author, Keynote Speaker

Enterprise AI strategist with twenty-five years leading transformation programmes across Asia and the Middle East, specialising in impact assessment, governance and deployment methodology.

Read the full profile ›

Wide AI adoption and meaningful financial return have become two separate facts about the same companies. McKinsey found seventy-eight percent of firms use AI, thirty-three percent have scaled it company-wide, and only 5.5 percent report a profit lift above five percent from it. S&P Global found forty-two percent of enterprises had scrapped most of their AI initiatives by late 2025, up from seventeen percent the year before. The bottleneck is not model quality. It is how AI is built into the company, not how smart the AI is.

Task-level time savings, and Anthropic's own research puts these as high as eighty percent on individual tasks, disappear as slack the moment they meet a workflow that was never redesigned around them. Each shortcut an organisation takes to get an early pilot working sets up the next phase to inherit the same problem, while a competitor doing the harder, slower work of rebuilding how the company actually operates compounds an advantage that becomes difficult to close.

The 5.5 percent of companies getting a real profit lift differ from the other 94.5 percent on six measurable dimensions: they allocate more than twenty percent of their technology budget to AI rather than under ten, govern it enterprise-wide rather than through isolated teams, connect their data rather than leaving it fragmented, and redesign processes fundamentally rather than adding AI on top of them unchanged. Gartner separately found sixty percent of AI projects fail on poor data foundations and over forty percent of agent projects fail on legacy system incompatibility, both detectable before a single dollar is spent on implementation.

The roadmap that follows runs eighteen months in three phases: an audit in the first quarter that eliminates weak projects rather than defending them, a nine-month window that redesigns workflows and connects systems rather than layering AI on top of what already exists, and a final phase that scales what worked under real governance. Delaying any phase does not preserve optionality. It compounds the same architectural debt a competitor is spending that quarter paying down.

Reference

This piece is adapted for Praxora Lab from the original. Originally published at terencekok.com ›

More in Strategy