Is Business Scaling About To End AI's Productivity Paradox?

Complaints about AI remain legion. Despite the vast sums invested, the returns to date at most businesses are generally characterized as disappointing. 

But business and technology have been here before. From the steam engine to electricity to computers, it has taken organizations time to figure out how to most effectively apply new technologies. Rather than a fundamental criticism of AI, the latest productivity paradox - massive investments failing to generate significant productivity increases - may simply reflect the early stage experimentation that eventually lead to the promised - and hoped for - gains. JL

Steven Rosenbush reports in the Wall Street Journal:

As of last year, 78% of companies said they used artificial intelligence in at least one function. From these efforts, companies typically find cost savings of less than 10% and revenue increases of less than 5%. The numbers reflect a “productivity paradox,” in which massive improvements in AI capabilities haven’t led to a corresponding surge in national-level productivity. “Many companies are disappointed with their AI projects.” Only 1% of U.S. companies that have invested in AI have scaled their investment, while 43% are still in the pilot stage. “One cannot expect productivity gains at the pilot level or even at the company unit level. Significant productivity improvements require achieving scale.”

AI adoption among companies is stunningly high, but most of them are struggling to put it to good use. They intuit that AI is essential to their future. Yet intuition alone won’t unlock the promise of AI, and it isn’t clear to them which key will do the trick.

As of last year, 78% of companies said they used artificial intelligence in at least one function, up from 55% in 2023, according to global management consulting firm McKinsey’s State of AI survey, released in March. From these efforts, companies claimed to typically find cost savings of less than 10% and revenue increases of less than 5%.

While the measurable financial return is limited, business is nonetheless all-in on AI, according to the 2025 AI Index report released in April by the Stanford Institute for Human-Centered Artificial Intelligence. Last year, private generative AI investment alone hit $33.9 billion globally, up 18.7% from 2023.

The numbers reflect a “productivity paradox,” in which massive improvements in AI capabilities haven’t led to a corresponding surge in national-level productivity, according to Stanford University economist and professor Erik Brynjolfsson, who worked on the AI Index. While some specific projects have been enormously productive, “many companies are disappointed with their AI projects.”

Resolving the productivity paradox

For companies to get the most out of their AI efforts, Brynjolfsson advocates for a task-based analysis, in which a company is broken down into fine-grained tasks or “atomic units of work” that are evaluated for potential AI assistance. As AI is applied, the results are measured against key performance indicators, or KPIs. He co-founded a startup, Workhelix, that applies those principles.

Companies should take care to target an outcome first, and then find the model that helps them achieve it, says Scott Hallworth, chief data and analytics officer and head of digital solutions at HP.

Orchestrating and scaling AI

A separate report from McKinsey issued in January helps explain why AI adoption is racing ahead of associated productivity gains, according to Lareina Yee, senior partner and director at the McKinsey Global Institute. Only 1% of U.S. companies that have invested in AI report that they have scaled their investment, while 43% report that they are still in the pilot stage. “One cannot expect significant productivity gains at the pilot level or even at the company unit level. Significant productivity improvements require achieving scale,” she said.

 

The critical question then, is how companies can best scale their AI efforts.

Ryan Teeples, chief technology officer of 1-800Accountant, agrees that “breaking work into AI-enabled tasks and aligning them to KPIs not only drives measurable ROI, it also creates a better customer experience by surfacing critical information faster than a human ever could.”

The privately held company based in New York provides tax, booking and payroll services to 50,000 active clients, with a focus on small businesses. The company isn’t a Workhelix customer.

 

Additionally, he says, companies should look beyond individualized AI usage, in which employees use GenAI chatbots or AI-equipped productivity tools to enhance their work. “True enterprise adoption…involves orchestration and scaling across the organization. Very few organizations have truly reached this level, and even those are only scratching the surface,” he said.

The use of AI at 1-800Accountant begins with an assessment of whether the technology improves the client experience. If the AI provides customers with answers that are as good, better or faster than a human, it’s a good use case, according to Teeples. In the past, the company scheduled hourlong appointments with advisers who answered simple client questions, such as the status of their tax return.

Now, the company uses an AI agent connected to curated data sources to address 65% of customer inquiries, with 30% arranging a call with a human. (The remaining 5% drop out of the inquiry process for various reasons.) The company uses Salesforce’s Agentforce to handle customer inquiries and its Einstein platform for orchestration across 1-800Accountant’s back end.

Teeples said the company is saving money on the cost of human advisers. “The ROI in this case was abundantly clear,” he said.

Orchestrating AI across the enterprise requires the right infrastructure, especially when it comes to data, according to Gabrielle Tao, senior vice president for data cloud at Salesforce. It is important, she said, to harmonize data, for example, by creating a consistent way to refer to business concepts such as “orders” and “transactions,” regardless of the underlying data source.

AI deployments should target tasks that are both frequent and generalizable, according to Walter Sun, global head of artificial intelligence at SAP. Infrequent, highly specific tasks such as a marketing campaign for a single event might benefit from AI, but applying AI to regularly occurring tasks will achieve a more consistent ROI, he said.

We’ve been here before

Historically, it has taken years for the world to figure out what to do with revolutionary general-purpose technologies including the steam engine and electricity, according to Brynjolfsson. It isn’t unusual for general-purpose models to follow a “J-curve,” in which there’s a dip in initial productivity, as businesses figure things out, followed by a ramp-up in productivity.

He says companies are beginning to turn the corner of the AI J-curve.

The transformation may occur faster than in the past, because businesses—under no small amount of pressure from investors—are working to quickly justify the massive amount of capital pouring into AI.