The Productivity Paradox Is Still Alive

The first time I read Brynjolfsson's 1993 paper, what struck me was not that productivity failed to show up in the numbers. It was how obvious the four explanations were once someone actually named them. Organizations poured money into IT through the 1970s and 1980s, and macroeconomic productivity statistics barely moved. The puzzle felt like a paradox only because we kept assuming the relationship between spending and output was simple. It never was.

Brynjolfsson gave the productivity paradox four explanations, and I think he was right to keep them separate because each one demands a different response. Mismeasurement: the productivity statistics did not capture quality, variety, customization, or convenience improvements. Time lags: IT benefits require organizational restructuring and learning, which take years. Redistribution: productive firms pass gains to customers as lower prices rather than capturing them as higher profits. Mismanagement: firms invest without aligning processes, skills, or governance, and produce waste instead of productivity. The last one is where I keep seeing the same pattern repeat in 2026. Organizations buy generative AI licenses and announce transformation initiatives, but the workflow stays the same and the dashboard goes unused. The technology arrives. The complementary changes do not.

Brynjolfsson and Hitt (1996) resolved the paradox at the firm level. IT capital stock has a significantly higher marginal product than non-IT capital, especially when paired with decentralized organizational structures and worker empowerment. The paradox was partly an artifact of aggregation. When you measure at the economy level, the firm-level signal gets lost. But Brynjolfsson and Hitt also showed something that I think gets missed too often in the victory lap: value and profitability are not the same thing. IT can make firms more productive while competitive markets push the gains to consumers as lower prices. A firm can create real value and capture none of it as profit. When I hear companies say "we invested in AI and our productivity went up," I want to ask: productivity for whom? Your employees, your shareholders, or your customers?

Then comes Carr (2003), who took the logic to its provocative endpoint. As IT becomes ubiquitous and standardized, it becomes commodity infrastructure, necessary like electricity or railroads but incapable of generating sustained competitive advantage. The argument turns on scarcity. Sustained advantage requires that some firms possess a resource others do not. When every company can access the same cloud platform, the same large language models, the same vendor ecosystem, no one holds a scarce resource. I think Carr is right about infrastructure and wrong about the slide from infrastructure to capability. Bharadwaj (2000) demonstrated that firms with high IT capability (defined as the organizational ability to mobilize IT infrastructure, human IT resources, and IT-enabled intangibles) achieve significantly higher profit ratios and lower cost ratios than matched controls. The mechanism is social complexity, path dependence, and causal ambiguity. The hardware is the same. The capability to deploy it well is not.

This is where the Resource-Based View does real work. Barney (1991) established that sustained advantage requires resources that are valuable, rare, inimitable, and non-substitutable. VRIN characteristics, not types of resources. I stress that distinction because the wrong question is "is IT VRIN?" The right question is whether a specific IT-related resource bundle has VRIN characteristics in a specific competitive context. IT spending alone is rarely VRIN because competitors can buy the same technology. IT capability can be VRIN because it bundles infrastructure, human skill, and organizational intangibles in ways that are hard to replicate. Every company can rent GPU time. Not every company can reconfigure how decisions get made around what those GPUs produce.

But RBV has a limit that matters more now than it did in 1991. VRIN resources explain advantage in stable environments. In turbulent markets, the question shifts from "do you have the right resources?" to "can you reconfigure your resources fast enough?" That is where dynamic capabilities enter. Teece et al. (1997) defined them as the organizational ability to sense opportunities and threats, seize them by mobilizing resources, and transform by reconfiguring routines. Eisenhardt and Martin (2000) sharpened the construct by insisting that dynamic capabilities are not vague abilities. They are specific, identifiable processes like product development, strategic decision-making, and alliancing. They also made a point I think is underappreciated: dynamic capabilities are necessary but not sufficient for advantage. The advantage resides in the resource configurations these capabilities produce.

The distinction between ordinary and dynamic capabilities is where Carr's argument collapses. Using a generative AI tool to draft documents is an ordinary capability. Embedding AI into organizational decision routines and restructuring workflows around it is potentially a dynamic capability. Detecting cyber intrusions is ordinary. Reconfiguring your detection methods after a new threat class emerges is dynamic. The commodity argument wins on hardware and standard software. It loses on the organizational ability to reconfigure, which stays heterogeneously distributed across firms because it is embedded in routines that are socially complex and path-dependent.

Melville et al. (2004) gave us the model that refuses the direct shortcut. IT resources interact with complementary organizational resources, which improve business process performance, which then drives organizational performance, all moderated by the competitive environment. IT does not improve performance directly. Torres, Sidorova, and Jones (2018) showed this concretely for business intelligence and analytics. BI&A reaches firm performance through business process change capabilities and functional performance, not through analytical output by itself. Dashboards do not create value. Acting on what they show does.

I keep coming back to the mismanagement explanation from Brynjolfsson's original four. Not because it was the only explanation, but because it is the one that keeps recycling. Torres, Sidorova, and Jones made it empirical for analytics. The productivity paradox was never just a measurement problem that better data could fix. It was a theoretical lesson: the naive assumption that spending produces performance is wrong. You need complementary resources. You need process change. You need organizational restructuring. When I read about companies investing in AI and then measuring productivity gains by counting how many employees use the tool, I hear the same mistake Brynjolfsson named in 1993. The measure is wrong. The mechanism is missing. The paradox is not back. It never left.