The 60% AI Productivity Gain Is Real. What the Teams Achieving It Have in Common

The 60% number gets cited in decks, dropped in all-hands meetings, and repeated by vendors who had nothing to do with producing it. Almost nobody reads what actually happened in the teams that generated it. That gap between the headline and the practice is where most AI rollouts quietly fail.

It's worth being honest about the baseline. A 2025 NBER study of 6,000 executives across four countries found that over 80% of firms report zero productivity gains from AI. Meanwhile, Foxit's 2026 research found that executives who feel more productive from AI actually gain just 16 minutes per week once verification time is accounted for. Their end users lose 14 minutes. The gains are concentrated. The question is what separates the teams in the top decile from everyone else.

Narrow beats broad, every time

The clearest pattern among teams that achieve substantial gains is specificity. They don't adopt "AI for productivity." They adopt a specific tool for a specific task with a specific success metric. GitHub's own research on Copilot found developers completed coding tasks 55% faster when using the tool, but that figure applies to coding tasks, not to all developer work. The moment you zoom out to "does this make our engineering team more productive overall," the number gets murky.

Goldman Sachs documented a 30% productivity boost from AI, but limited it explicitly to coding and customer service, and found no meaningful economy-wide relationship between AI adoption and output. The gains are real where the application is surgical.

Zipify's customer service team is a useful example. Rather than rolling out an AI chatbot as a general support layer, they built an internal AI assistant paired with an analytics dashboard aimed specifically at response speed and knowledge retrieval. The result was a 65% reduction in response times, tickets resolved twice as fast, and a 24% rise in customer satisfaction. The outcome traces directly to the specificity of the implementation, not the sophistication of the technology.

The work changes, not just the tools

The second differentiator is structural. Teams that gain nothing from AI tend to add the tool on top of existing workflows. Teams that gain a lot redesign the workflow around the tool's actual strengths. This sounds obvious. It almost never happens in practice.

The "Navigating the Jagged Technological Frontier" study from INFORMS Organisation Science makes this point structurally: workers who were skilled at using AI got substantial quality and productivity benefits, while workers who weren't got little or none. Skill here doesn't mean prompt engineering. It means understanding which tasks are inside the AI's capability frontier and routing work accordingly. That's a workflow redesign problem, not a training problem.

Boston Consulting Group found in 2026 that using four or more AI tools correlates with plummeted productivity, a phenomenon they called "AI brain fry," where the cognitive overhead of managing and verifying multiple AI outputs outweighs the time saved. More tools, worse outcomes. The teams that gain are the ones that go deeper on fewer tools, not wider across many.

If you're evaluating which tools deserve that deeper investment, explore which AI tools drive the biggest gains by use case rather than category. The distinction matters: "AI writing tools" is a category. "AI tools that cut first-draft time for customer-facing copy by 40%" is a use case. Only one of those helps you make a decision.

Measurement determines behaviour

The third differentiator is how teams define success. Teams that report strong gains measure output: tickets resolved, code shipped, documents reviewed, time to hire. Teams that report weak or no gains tend to measure adoption: seats activated, logins per week, satisfaction scores. The McKinsey 2025 Superagency report found that 48% of workers cite training as a key barrier to AI productivity, but training without output measurement just produces more confident non-results.

This matters for team dynamics specifically. When a team is measured on output, every member has an incentive to find where the tool genuinely helps. When they're measured on adoption, they have an incentive to look like they're using the tool, regardless of whether it's doing anything. Foxit's finding that executives report feeling more productive while gaining 16 minutes weekly is the clearest illustration of this: the subjective experience of using AI looks like productivity, even when actual output hasn't moved.

The practical fix is a pre-mortem before any AI rollout. Ask: what would the output numbers look like after 90 days if this tool is actually working? Write down the number. Then measure it. If your team can't agree on what number would prove the tool is working, the rollout isn't ready.

What the 60% figure actually requires

Significant AI productivity gains aren't a function of which tools you buy. They're a function of how narrowly you deploy them, how thoroughly you redesign the surrounding work, and how honestly you measure what changes. Most teams do none of these things. Some do one. The teams in the top decile do all three.

The team dynamics piece is underrated here. High-performing teams have someone whose job it is to own the AI workflow, identify where the tool's frontier sits, and flag when cognitive overhead is creeping back up. It's not a full-time role in most organisations, but it needs to be someone's explicit responsibility. Otherwise the tool gradually gets used for everything, verified for nothing, and measured by vibes.

Start this week by picking one task your team does repeatedly, identifying one tool that demonstrably accelerates that specific task, and setting a 90-day output target before you roll it out. One task. One tool. One number. That's what the teams hitting real gains actually did first.