At the Swiss School of Management, the Research Center continuously scans the horizon for developments that genuinely change how organizations work and how leaders should prepare. This summer, a group of researchers at MIT released The GenAI Divide: State of AI in Business 2025, a rigorous study that cuts through the hype and clarifies where generative AI is, and is not, creating measurable value. The findings matter for executives, for practitioners, and for the next generation of doctoral research we cultivate at SSM.
What the research finds
The report describes high adoption but low transformation: consumer tools like ChatGPT and Copilot see broad use, yet enterprise-grade or custom systems rarely reach production—only ~5% of task-specific pilots scale with durable P&L impact. Seven of nine sectors show minimal structural change, with meaningful disruption concentrated in Technology and Media & Telecom. The bottleneck isn’t infrastructure, regulation, or even talent—it’s learning: most systems fail to retain feedback, accumulate context, and improve over time, so users revert to flexible consumer interfaces for quick tasks and abandon brittle enterprise tools for mission-critical work. A “shadow AI economy” has emerged in which employees privately use personal LLM accounts daily, often generating more practical ROI than sanctioned pilots; successful enterprises study this shadow usage to guide procurement. Investment is also misallocated: roughly half (and often more) of AI budgets flow to front-office sales/marketing because results are easier to attribute, while back-office areas (finance, procurement, operations) quietly deliver faster and clearer savings, like reduced BPO and agency spend, when solutions truly learn and fit the workflow. Finally, organizations that buy and co-develop with external partners report about 2× higher deployment success than those building entirely in-house, and the leaders are moving toward agentic systems with memory and feedback loops as the window to “cross the divide” narrows.
What this means for business leaders
For leaders, the message is pragmatic: treat GenAI like an operational transformation, not a feature. Start where learning matters: pick narrow, high-value workflows, demand persistent memory, require deep integration with existing systems, and measure vendors on business outcomes (time-to-value, error reduction, external spend avoided), not model benchmarks or flashy demos. Translate “shadow AI” into sanctioned, bottom-up pilots led by power users, then scale what works. Prioritize back-office ROI (document flows, AP/AR, customer service routing) before chasing headline front-office use cases; expect selective workforce impacts (customer support, software engineering, admin) driven more by outsourcing reductions than broad layoffs. And when choosing partners, buy before you build, insist on clear data boundaries, and evaluate the vendor’s ability to learn and adapt over time—because enterprises are already locking in learning-capable platforms and the window is closing.
Preparing Leaders to Cross the Divide
The key message is clear: organizations still have much to learn about effective, profitable, and responsible AI. The leaders who will bridge this divide combine advanced education with critical thinking, operational fluency, and a rigorous focus on outcomes. This is precisely the mission we pursue at the Swiss School of Management: developing knowledge, cultivating leaders, and preparing managers to navigate AI with clarity and purpose.
Read the full article here: The GenAI Divide: State of AI in Business 2025