Organizations have long viewed artificial intelligence as a way to achieve productivity gains. However, recent research into AI adoption in US manufacturing companies reveals a more nuanced reality. AI often sees measurable but temporary declines in performance, followed by increased growth output, revenue and employment.
This phenomenon following the trajectory of “J-Curve” helps explain why the economic impact of AI is overwhelmingly overwhelming despite its potential for transformation.
“AI is not plug and play,” says Professor Christina McKehelan, a digital fellow at the University of Toronto's Digital Economic Initiative and one of the lead authors of the new paper, “The rise of industrial AI in America: the microfound of productivity J-Curve.” “It requires systematic change, and the process introduces friction, especially in established companies.”
Professor Mu-Jeung Yang, University of Colorado, Boulder. Zachary Kroff was previously an analytics specialist for the Analytics Group at the US Census Bureau. Professor Eric Brinjolfson, PhD, from Stanford University, co-authored the report.
Using data from two US Census Bureau surveys covering tens of thousands of manufacturers in 2017 and 2021, researchers found that AI adoption J-Curve differed between companies that adopted AI technology in industrial applications. Short-term losses were greater in older and more established companies. Evidence for young companies showed that losses can be mitigated by specific business strategies. And despite early losses, early AI adopters showed stronger growth over time.
Let's take a look at what this study shows about AI adoption and application, and the types of companies that are better than others when using new technologies.
1. Adopting AI will initially reduce productivity.
This study shows that AI adoption tends to hinder productivity in the short term, and that companies are experiencing measurable declines in productivity after they begin using AI technology.
Even after controlling for size, age, capital stock, IT infrastructure, and other factors, researchers found that organizations that adopted AI for business functions saw a 1.33 percentage points of productivity decline. When correcting selection bias – organizations where higher returns are likely to become AI adopters earlier – the short-term negative impact was significantly greater at around 60% points, the researchers wrote.
This decline is not just about increasing pain. It refers to a deeper inconsistency between new digital tools and legacy operational processes, researchers found. AI systems used for predictive maintenance, quality control, or demand forecasting often also require investments in data infrastructure, staff training, and workflow redesign. Without these complementary parts in place, even the most advanced technologies can dive or create new bottlenecks.
“As companies work through adjustment costs, they tend to experience stronger growth,” McElheran said. “But that first dip – the downward slope of the j-curve – is very realistic.”
Leading an AI-led organization
Directly on MIT Sloan
Sign up now
2. Short-term losses precede long-term profits.
Despite early losses in the firm, the study found clear patterns of recovery and ultimate improvement. Over a longer period, there was a four-year gap in research data – manufacturers that adopted AI tended to outperform their non-employment peers in both productivity and market share. This recovery followed the initial adjustment period in which companies tweaked processes, expanded their digital tools and leveraged data generated by AI systems.
However, the upward direction was not evenly distributed. Companies that saw the strongest profits tended to be already digitally mature companies before adopting AI.
“People who have already undergone digital transformation or have been digital from Get-go have a much easier ride, as historical data can be a good predictor of future outcomes,” McElheran said. The size also helps. “If we solve these adjustment costs, if we can expand our profits with more production volumes, more markets and more customers, we will be making J-Curve's rise faster,” she said.
Researchers have been crucial for this recovery, as companies gradually shift towards more AI-compatible operations, often investing in automation technologies such as industrial robots.
3. Older businesses are increasing their short-term losses.
Short-term losses do not feel even in all companies, the study found. The negative impact of AI adoption was most prominent among established companies. These organizations usually have long-term routines, layered hierarchies, and legacy systems that are difficult to relax.
These companies struggle to adapt due to institutional inertia and the complexity of their operations. “We see that older businesses in particular struggle to maintain key production management practices, such as monitoring key performance indicators and production targets,” the researcher wrote.
“Old companies have seen a decrease in the use of structured management practices after adopting AI,” McElheran said. “And that alone accounts for almost a third of the productivity losses.”
In contrast, younger and more flexible companies appear to be better at quickly and consolidating AI technology and consolidating it less and more disruptively. You may also need to learn less, making the transition to AI-enabled workflows more seamless.
“Together, our findings highlight the dual role of AI as a transformative technology and catalyst. They note that results demonstrate the importance of complementary practices and strategies that mitigate adjustments, and also show that “flats the J-Curve Dip and boosts long-term returns to achieve long-term productivity for AI at scale.”
