Why Is It so Hard to Get ROI From AI? Because Building From First Principles Isn’t Easy

Why It Matters

Without first‑principles preparation, AI projects waste capital and fail to generate the productivity gains enterprises expect, slowing digital transformation across regulated sectors.

Key Takeaways

• Foundation work—data, governance, workflow mapping—precedes AI ROI.
• Rushing AI without strategic goals leads to “weaponized” inefficiencies.
• Redesigning processes for AI-native execution drives measurable productivity gains.
• AI agents are maturing; firms must prepare for autonomous software development.
• Measuring AI value includes intangible benefits like enhanced customer relationships.

Pulse Analysis

The promise of artificial intelligence has become a buzzword, but the reality on the shop floor tells a different story. Companies that treat AI as a plug‑and‑play add‑on often stumble because they lack the foundational data pipelines, governance frameworks, and process maps that enable reliable model deployment. As State Street’s CTO Manoj Bohra analogized, building AI is more akin to constructing a bridge than flipping a switch; the initial investment in data quality and regulatory compliance can span years before any return materializes. This perspective resonates across regulated industries where missteps can trigger compliance breaches as well as sunk costs.

A second, equally critical lesson is the need to redesign business workflows for an AI‑native environment. Deloitte’s Bill Briggs warned that simply overlaying AI on legacy processes merely scales existing inefficiencies, a phenomenon he likened to the early days of electrification when firms swapped steam for electricity without rethinking plant layouts. ReviveHealth’s Kathy Pham illustrated how misaligned objectives—such as using AI to read bedtime stories when the goal is parent‑child bonding—can erode value. In contrast, firms like Genspark are targeting quick wins, such as AI‑generated video prototypes for ad pitches, which cut production costs and improve win rates, demonstrating that purposeful process redesign can unlock tangible revenue.

Looking ahead, the AI landscape is shifting from experimental pilots to production‑grade autonomous agents. Lambda’s Stephen Balaban noted that AI agents can now write software, foreshadowing a wave of self‑service AI across functions. This evolution forces service firms to move toward outcome‑based pricing, charging for results rather than headcount. Simultaneously, concerns over AI sovereignty and efficiency are prompting investments in heterogeneous data‑center architectures that blend GPUs, CPUs, and specialized inference chips. Companies that combine solid foundational work with strategic process re‑engineering will be best positioned to capture the long‑term ROI that AI promises.