Your AI Strategy Isn't Failing. It's in the Dip.

Why the productivity crash you're seeing right now is the precondition for the payoff, not evidence there won't be one.

| Written by Preethy Suresh

PwC's 29^th Global CEO Survey, released at Davos in January 2026, asked 4,454 chief executives across 95 countries a direct question. Has AI produced increased revenue or reduced costs at your company over the last 12 months? 56% said neither. Only 12% could point to both. 

This isn't researchers judging pilots from the outside. This is CEOs - the people who signed off on the spend, defended it to boards, put their credibility behind it. They are now admitting that most of what's been invested has produced no measurable return.

The number lands harder when you look at the context.

The same survey found CEO confidence in short-term revenue growth at a five-year low, at 30%, down from 38% in 2025 and 56% in 2022. AI budgets keep climbing. Boards keep pushing. And the executives being asked to justify all that spend can't yet demonstrate what it's produced. 

The reactions I'm seeing fall into two camps.

The first is I knew it was hype. This is the AI-skeptic camp, quietly enjoying the fact that the technology is finally getting its retribution.

The second is we need to push harder. More licenses, more mandates, more training drives. Two opposite reactions. Both wrong. Both making the same assumption. That the failure means the technology isn't working.

What if these numbers aren't a signal that AI has been overhyped, or that we're deploying it wrong? What if they're a signal that we are, collectively and globally, sitting inside a pattern that has repeated every single time a general-purpose technology has arrived in human history?

Because we are. And it has a name.

The Pattern Has a Name

The economist Erik Brynjolfsson and his team, wrote a paper for the National Bureau of Economic Research in 2018 called "The Productivity J-Curve: How Intangibles Complement General Purpose Technologies." Their argument is straightforward.

When a general-purpose technology like AI arrives, productivity doesn't climb. It dips. Sometimes for years. This happens because the technology requires enormous complementary investments before the gains materialise. Process redesign. New business models. New human capability. New organisational structures.

The shape on the chart is a J. The dip is the bottom of it. The climb on the other side is steep.

In a follow-up paper published with the US Census Bureau in November 2024, Brynjolfsson and his co-authors did something more useful. They confirmed the J-curve is happening right now, at the firm level, in AI adoption specifically. Short-run losses from organisational disruption. Medium-term performance improvements once the redesign catches up. Documented. Measurable. Predictable.

We've Seen This Movie Before

Factories electrified in the 1890s. Productivity stayed flat for roughly thirty years. Why? Because factories were still laid out for steam power, with one massive central engine driving everything through belts and shafts. Electricity's real power was distribution. A motor on every machine, modular floor plans, completely different workflows. The technology arrived in a decade. The operating model took a generation.

Fast-forward to 1987. Nobel laureate Robert Solow wrote a review in the New York Times Book Review called "We'd Better Watch Out", where he made the observation that later became famous. "You can see the computer age everywhere but in the productivity statistics." Companies had bought computers for everyone. Productivity didn't move. It took until the late 1990s, when business processes were finally redesigned around digital workflows rather than digitised versions of paper ones, for the gains to show up.

Every general-purpose technology in modern history has done this. Electricity. Computers. The internet. AI is not the exception. AI is the rule.

The pilots stalling. The ROI missing. The 56% of CEOs who can't demonstrate a return. That isn't the end of the story. It's the middle. We are sitting in the dip, and every quarter spent there generates a fresh round of claims that the technology has failed. Almost none of those claims recognise the shape of the curve they're standing on.

What Actually Happens in the Dip

Two wrong reactions dominate right now.

Pull back. AI was overhyped. Let's slow down. Let the dust settle. This is the reaction that lost companies the internet. Lost them mobile. Lost them cloud. The dip is the worst possible moment to disengage. Every quarter spent on the sidelines is a quarter competitors are spending on the redesign.

Push harder on the tool. More training. More licenses. Mandate it. Force it. This is the reaction that produced the 56% number in the first place. Pushing harder on a tool while the operating model stays the same doesn't get you up the curve. It just makes the dip deeper.

Both reactions have the same root cause. They treat AI as a technology problem. It isn't. It's an operating model problem.

The dip isn't empty time. The dip is where the work happens. And the work has a name. Workforce architecture. The small percentage of organisations that will come out of the dip on the upswing are doing the same things. They're mapping work at the task level, not the job level. They're sorting tasks into Automate, Enhance, and Elevate. They're redesigning teams around the new work, not retrofitting new tools into old org charts.

They're rebuilding performance management, career paths, hiring profiles, and learning systems, because every one of those was designed for a pre-AI work model.

Every productivity revolution in modern history started with a period that looked exactly like failure.

Every single one.

The companies that won weren't the ones who saw the future most clearly. They were the ones who did the unglamorous redesign work while everyone else was panicking, hedging, or pretending the dip wasn't real.

We are in the dip right now. The next two to three years will decide which companies make it up the other side, and which ones spend the rest of the decade explaining why their AI investment didn't deliver.