Restaurants and construction are two industries that sit at opposite ends of the same economy. One spent 30 years going nowhere, then found a new gear almost overnight. The other has spent literally half a century getting worse. The diverging fortunes of these industries sit at the centre of a question that has puzzled economists for years: why does productivity growth stall in some parts of the economy while surging in others, even when the businesses involved face similar costs, regulations and technology?

This is a question that has long intrigued Chad Syverson, the George C. Tiao Distinguished Service Professor of Economics at the University of Chicago Booth School of Business. In a recent conversation with Elvira Sojli, Associate Professor and Scientia Fellow Alumni in the School of Banking and Finance at UNSW Business School, Prof. Syverson set out what separates industries that can adapt their productivity from those that cannot.

In the discussion, Prof. Syverson, whose research focuses on the interactions of firm structure, market structure, and productivity, discussed the evidence for productivity: from the individual manager running a single store, to whether AI can lift an entire economy, to what governments can and cannot buy with an industrial policy budget.

How the pandemic forced restaurants to innovate

Every industry Prof. Syverson has studied contains a mix of high performers and laggards operating side by side, often selling to the same customers under the same regulatory settings: “That different trajectory and path for productivity across industries is not just true across industries, it’s also true within industries across different companies,” he told A/Prof. Sojli. “It doesn’t matter how thin you slice the part of the market you’re looking at, you’re going to find really high productivity producers and really low productivity producers.”

Understanding what pushed restaurants forward while construction fell further behind offers a working model of the forces that separate a productive business from an unproductive one, and a caution to anyone who assumes that spending money in an industry automatically changes how it operates.

Learn more: Why innovation isn’t translating into productivity growth

Prof. Syverson has conducted research with Dr Austan Goolsbee, now president of the Federal Reserve Bank of Chicago. One study covers an industry that customers visit several times a week; the other, an industry most people deal with once or twice in a lifetime. The underlying question behind both papers stayed the same: what changed, or failed to change, in how the business gets done.

Restaurant productivity sat still for three decades: “They had flat productivity for 30 plus years, I mean, just absolutely didn’t change,” he said. “Then COVID hit. Productivity fell sharply over a couple of months, then rose as the pandemic receded. But it went past its old level.” Output per unit of input across the industry settled about fifteen per cent above where it had sat before the pandemic – and stayed there.

He traced the reasoning for this productivity jump to the takeaway and delivery services. Full-service restaurants had rarely bothered with either before 2020, according to Prof. Syverson, because table service already covered the bulk of revenue and delivery was seen as a distraction. Once government restrictions removed table service as an option, operators had no choice but to work out how to get food out the door, and many discovered that delivery, once organised properly, moved meals to customers efficiently, even after accounting for the extra effort of not eating in-house.

The capability to run a delivery operation had existed well before COVID-19, but restaurants had no pressing reason to build it until the pandemic removed the alternative. Not every operator survived the transition: businesses that could not rework their operations around takeaway and delivery closed, while those that did carried a new operating model into the recovery. The 15% gain measured across the industry reflects the composition of who remained as much as the improvement made by any individual restaurant. The pattern raises a question for any business currently treating an operating constraint as pure cost rather than as a prompt to redesign how it reaches customers.

Fifty years of decline in construction

Construction offered no equivalent turnaround. Prof. Syverson and Dr Goolsbee examined decades of data and reached what Prof. Syverson called “the strange and awful performance of the industry”.

Set against the 30-year stagnation he described in restaurants, construction’s record was worse still: “it had 50 years of decline at least, and so you’ve got an industry that’s actually getting less and less efficient every year at doing what it does. It takes more and more inputs to make what it used to make,” he said. The number of workers needed to build a single-family home in the United States has not fallen in 50 years, and he noted that adjusting for the quality of the finished house does not change the picture.

Learn more: As tech advances ramp up, can policy really spur innovation?

Prof. Syverson and Dr Goolsbee were able to rule out several common explanations. The decline was not driven by a lack of investment in physical capital or standard technology, nor by rising market power among builders, nor by growing difficulty in finding workers. A/Prof. Sojli asked whether improved safety on building sites accounts for part of the fall, given the added supervision and compliance that today’s building sites carry compared with fifty years ago.

Prof. Syverson confirmed that the researchers assigned a dollar value to statistical deaths and injuries avoided or treated and treated the reduction as additional output the industry generated but never counted in the official numbers. On that basis, the 50% decline over the period narrowed to around 46%, a move in the expected direction but nowhere near enough to account for the bulk of the fall.

What remained, according to Prof. Syverson, was more speculative, since the underlying data did not exist to test it directly. Zoning restrictions and the sheer number of parties able to block or delay a project, on major infrastructure and on ordinary housing alike, add friction that shows up nowhere in productivity statistics until a project stalls.

"You can’t teach an old dog new tricks. It’s kind of true of companies, too"

CHAD SYVERSON

Why individual managers move the needle

Prof. Syverson has also examined the impact of management on productivity and conducted research on two large retail chains, each operating hundreds of stores. Because the chains tracked exactly who managed each store and when managers moved between them, Prof. Syverson could separate a store’s performance into two parts: the store itself, which remained fixed regardless of who ran it, and the manager’s own effect, which varied when the manager did. Pricing, product range and HR policy were already set at the head office in both companies, yet performance still varied widely from store to store, even with those decisions held constant. “We found out managers have a big role in predicting the productivity of stores, even inside companies,” Prof. Syverson said.

What made a manager good remained largely a mystery. Practices such as running feedback loops on performance and correcting shortfalls quickly are known from other research to improve results, but Prof. Syverson wanted to know whether any manager could apply them with equal success, or whether something about the individual mattered as well.

The one measurable trait shared by high performers, the ratio of full-time to part-time staff, worked in opposite directions at the two chains studied: raising it helped at one company, cutting it helped at the other, leaving it more a curiosity than an explanation. “We know there’s something special about the person themselves. What we don’t know is what that is,” Prof. Syverson said, adding that a more comprehensive answer would likely require the kind of personality data companies rarely collect.

Whether AI is already lifting productivity

A/Prof. Sojli turned the conversation to the impact of AI on productivity and asked whether a productivity lift was coming or whether AI would disappoint, as past technology waves have. “If you squint, you might convince yourself that we’re already seeing some movement in productivity statistics related to AI. But if you’re honest, you can’t say that we know for sure,” Prof. Syverson said.

US productivity growth had accelerated since 2022 and 2023 after two decades of weak performance, and industries reporting heavier AI adoption had shown faster gains than others, though the pattern remained too noisy to prove cause and effect, and Prof. Syverson cautioned that a change of this kind usually needs five or six years of data before it can be confirmed.

Subscribe to BusinessThink for the latest research, analysis and insights from UNSW Business School

He linked the uncertainty to the idea of a productivity J-curve: new general-purpose technologies tend to be undermeasured early on because the intangible investment required to use them is counted as an expense rather than an investment, and then overmeasured later once that investment pays off. If AI turns out to fit that pattern, he said, even the acceleration already visible in the data may still understate the technology’s true effect on productivity.

The scale of investment required raised a further question about who benefits. Larger firms have the resources to adopt AI quickly, which favours incumbents, but established companies often struggle to change how they operate, even when trying to do the same thing in a fundamentally different way, which can favour new entrants instead. “You can’t teach an old dog new tricks. It’s kind of true of companies, too,” Prof. Syverson observed.

He pointed to the AI industry itself as an illustration: new entrants sit alongside an established technology giant at the frontier of the field, and the same mix of old and new firms could plausibly play out again among AI’s users, some large incumbents succeeding on the strength of resources they already have, some new firms succeeding by applying the technology to old markets in ways incumbents find hard to match. Which force wins out is likely to vary by market, and Prof. Syverson said he had no strong sense yet of which kinds of markets would tilt which way.

The debate over market power

Prof. Syverson’s research also extends into whether firms across the economy have gained pricing power over time, a debate he reviewed for a forthcoming Annual Review article that draws together the accumulated evidence on the topic. Fellow University of Chicago economist Professor Ufuk Akcigit has separately argued that rising market power has also weakened knowledge spillovers between firms, as part of the broader debate about weakening competition.

“My view is, yes, they probably have gone up. I doubt they’ve gone up as much as the largest estimates are saying. I think it’s somewhere in the middle,” Prof. Syverson said, adding that the pattern varies widely across industries, with some markets showing large increases in pricing power and others becoming more competitive over the same period. He estimates part of the increase comes from input markets, including labour and supplier relationships, where a company can hold power over what it pays rather than only over what it charges; the two sources need not move together.

A further possible cause carries no implication of a policy failure. If production increasingly involves higher fixed costs relative to the cost of each additional unit sold, he said firms need a bigger markup, a larger sales volume, or both, simply to cover those costs – a shift in the underlying economics of production rather than weaker competition enforcement.