Why Organizational Redesign Determines AI Productivity
Most AI programs do not fail because the models are weak. They fail because the business keeps the old workflow and adds AI on top.
That was also one of the clearest lessons of 1990s IT productivity. Companies did not get the biggest returns from computers, databases, and enterprise software simply by buying the technology. They got them when they redesigned processes, shifted decision-making, retrained teams, and changed how work moved across the business. For senior executives investing heavily in AI today, that historical parallel matters: AI productivity depends as much on organizational redesign as on the tools themselves.Brynjolfsson and Hitt’s firm-level research [1][McKinsey’s *How IT Enables Productivity Growth* [2]
AI Productivity Comes From Organizational Redesign, Not AI Adoption Alone
Brynjolfsson and Hitt’s work helped explain why the so-called productivity paradox was incomplete. At the firm level, IT investments often did create value, but much of that value showed up only over multi-year periods because companies had to build complementary organizational capital around the systems.[1][3][4]
That is the executive takeaway for AI strategy. If a company adds copilots, models, or automation tools but keeps the same approval layers, reporting lines, handoffs, and incentives, the gains will usually stay local. Task speed may improve, yet company-level productivity remains hard to see.
By contrast, when leaders redesign end-to-end workflows, the economics change. AI can eliminate rework, compress cycle times, improve exception handling, and push decisions closer to the point of action. In practice, that means treating AI as an operating model change, not a software deployment.[2][5]
Process Redesign Was the Real Engine of 1990s IT Productivity
McKinsey’s research on 1990s productivity growth makes the same point from another angle: IT enabled productivity when it was tailored to sector-specific processes and combined with managerial innovation.[2][6] The technology mattered, but the bigger gains came from changing how core functions actually ran.
Consider retail. McKinsey found that U.S. retail productivity outpaced the broader economy in the 1990s, but not because every retailer spent on technology equally well.[6] The strongest performers used information systems to redesign replenishment, distribution, store operations, and supplier coordination.
That pattern should feel familiar to any executive thinking about AI. The biggest upside is rarely in isolated use cases. It is in redesigning a process such as customer service, underwriting, procurement, claims handling, or software delivery so that AI changes the flow of work across the entire system.
Company Examples Show How Redesign Unlocks Technology ROI
Walmart is one of the clearest historical examples. Its gains came not from IT spending in isolation, but from combining technologies such as barcode scanning, satellite communications, and EDI with cross-docking logistics and tighter supplier coordination.[7] In other words, the value came from redesigning the operating model around faster information flow.
Dell followed a similar logic. Its build-to-order model linked customer demand directly to production, reducing inventory and improving responsiveness.[7] Again, the breakthrough was not just digitization. It was process redesign enabled by technology.
Even in service businesses, the lesson held. Massachusetts Mutual Life redesigned underwriting by introducing a case-manager role supported by workstation systems, allowing specialists to support the process only when needed. That redesign reportedly gave management confidence in a 40% productivity goal within months.[8]
The common thread is simple: technology created visibility and speed, but organizational redesign turned those capabilities into business impact.
What This Means for AI Strategy Today
For AI, the implication is straightforward. Start with high-friction workflows, not with the tool. Ask where delays, handoffs, escalation loops, and fragmented knowledge are constraining performance. Then redesign the process so AI can remove bottlenecks instead of merely assisting people inside them.
That usually requires a broader set of changes than many AI programs plan for: new roles, new performance metrics, cleaner data ownership, narrower approval chains, and clearer decision rights. Research on IT and workplace organization also suggests that decentralized decision-making and new work practices often complemented technology investment especially well.[5]
The 1990s IT lesson for AI productivity is not that returns are automatic. It is that the biggest gains go to companies willing to redesign how work gets done. Executives who treat AI as a business transformation effort, not just a technology rollout, are far more likely to capture meaningful productivity growth and durable competitive advantage.