BCG’s research on what actually determines whether an enterprise AI initiative succeeds landed on a specific split: roughly 10% of the effort is the algorithm itself, licensing, vendor selection, model fine-tuning, 20% is the surrounding technology and data infrastructure, integration, security, monitoring, and the remaining 70% is people and process, change management, role redesign, governance, and manager coaching. The name is just that ratio.
Most enterprise AI budgets get allocated backwards. The majority goes to the model and the platform, and people-and-process work is treated as an afterthought, a training deck circulated once at launch. BCG’s data links exactly that pattern to weak adoption and stalled pilots. The organisations that see real productivity gains are the ones spending the disproportionate 70% deliberately, on workflow redesign, incentive alignment, and sustained manager coaching, not just a rollout announcement.
It’s a useful gut check for any AI business case: if the plan spends more time selecting the model than redesigning the workflow around it, the split has been inverted, and that’s usually where the pilot stalls before it ever reaches scale.