> ## Content Index > Fetch the complete content index at: https://www.trentcotton.com/llms.txt > Use this file to discover other available public pages before exploring further. # Where Did AI's Productivity Gains Go? 3 Answers CHROs Need in 2026 - URL: https://www.trentcotton.com/where-did-ais-productivity-gains-go-3-answers-chros-need-in-2026/ - Published: 2026-09-30T21:33:55.000Z - Updated: 2026-09-30T21:33:54.000Z - Author: Trent Cotton - Tags: Podcast Recaps, AI in HR ## **TL;DR:** **If AI made our people more productive, why isn't it showing up in profit?** **Short answer:** The gain is real, and most companies can't see where it went. Eight in ten respondents in McKinsey's 2026 State of AI survey say AI improved their productivity, yet only 37% can trace any of it to EBIT. My read is that the missing productivity went one of three places: it got spent on work that never mattered, the employer kept it through slower wage growth, or it's sitting undocumented on someone's desk. **Key stats you need to know:** - 80% of McKinsey's 1,719 respondents across 97 countries say AI improved their individual productivity, while only 37% attribute any EBIT impact to AI, a share essentially unchanged from 2025 ([McKinsey State of AI 2026](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai?ref=trentcotton.com)). - Workers in the occupations most exposed to AI saw real wage growth run about 6.7 percentage points slower than low exposure occupations after 2023, with no statistically significant change in employment ([Apollo Global Management](https://www.apollo.com/content/dam/apolloaem/pdf/daily-spark/2026/jul/30/Whitepaper-Impact%20of%20AI%20on%20U.S.%20Labor%20Market-2026-R2%201.pdf?ref=trentcotton.com)). - 72% of AI power users say their workload is reasonable, compared with 69% of people who don't use AI at all, a gap of three points ([Culture Amp AI at Work Benchmark](https://www.cultureamp.com/company/announcements/ai-at-work-benchmark-2026?ref=trentcotton.com)). **The leadership takeaway:** Over the next 12 to 24 months, any restructuring built on "AI made us more productive" needs a map of where those hours went, or it's a bet placed with someone else's money. 0:00 /1:00 1× **Watch the full breakdown on The Human Capitalist:** [Where Did AI's Productivity Go? on YouTube ](https://youtu.be/ba%5FInNECOCk?ref=trentcotton.com) **Listen on Spotify:** [The Human Capitalist Podcast ](https://open.spotify.com/show/57xtYAUsIYrg7gQ0a6RFwc?ref=trentcotton.com) --- I came across a number this summer that made me stop. McKinsey surveyed 1,719 people in 97 countries for its latest State of AI report. Eight in ten said AI made them more productive. Only 37% could say it reached the P&L, and that figure hasn't moved in a year ([McKinsey State of AI 2026](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai?ref=trentcotton.com)). That gap matters because companies aren't waiting for the answer. They're redesigning roles, freezing hiring, and cutting staff based on an assumption about what AI is doing to the business. Nobody can point to where the productivity went. I pulled three pieces of research that each give a different answer, and I think every CHRO needs to know which one describes their company. I call it the Spent, Kept, or Hidden test. ## Where Did the Productivity Gains From AI Go? The productivity gains from AI went somewhere, and the three most credible explanations are that the hours got spent on low value work, the employer kept the gain through wage compression, or the gain exists but nobody documented it. Each answer comes from a different body of research, and each one points to a different fix. Start with the adoption data. Nearly nine in ten organizations in McKinsey's 2026 survey now use AI regularly in at least one business function, and 2026 was supposed to be the year of scaling and agents. Profit sat still. The standard explanation is that ROI on any new technology has a long tail. That would be fine if workforce decisions were on hold too. They aren't. McKinsey found 39% of respondents expect AI to reduce their organization's total employment over the coming year, and 14% say it already did in the past year ([McKinsey State of AI 2026](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai?ref=trentcotton.com)). I've written about how often AI gets used as cover for cuts that were coming anyway in [The Alibi Layoff](https://www.trentcotton.com/the-alibi-layoff-is-ai-really-behind-the-cuts/). In the restructures I've watched, the step that gets skipped is mapping which work AI absorbed and which work still needs to be uniquely human. I know that mistake because I made it. In a prior role, we looked at an operations heavy function and decided we could cut the staff by 75%, add automation, and be fine. We weren't. A large part of what that group did was talk to clients. When we cut 75% of the team, we cut the chance for a client to reach a real person, and we saw it in the P&L. The labor data has started to show the same blind spot. Researchers at Glean surveyed 6,000 digital employees and found AI saves them about 11 hours a week, then 6.4 of those hours go to what they call "bot-sitting," supervising and fixing the output ([Culture Amp](https://www.cultureamp.com/blog/employees-stuck-in-technical-challenge?ref=trentcotton.com)). Saved hours have a way of getting quietly reabsorbed. ## What Is the Spent, Kept, or Hidden Test? The Spent, Kept, or Hidden test is a diagnostic I use at The Human Capitalist to trace where AI productivity went when it never reached the P&L. Each answer has its own evidence and its own mindset from my book, The FutHRist. ### Spent: the hours went to work that never mattered I use the Value of Work Matrix here, a tool I built with two axes: value to the organization and the human focus a task requires ([The Value of Work Matrix Toolkit](https://www.trentcotton.com/the-value-of-work-matrix-toolkit-map-your-teams-actual-work-against-actual-hours/)). AI empties the bottom left quadrant first, the low value, low focus work. It's easy to hand over and it wasn't producing value to begin with. The task disappears, the hours come back, and if nobody redeploys those hours, the P&L can't move. I walk through each quadrant in [my Value of Work Matrix breakdown on YouTube](https://youtu.be/%5FumOtBmp%5FBY?ref=trentcotton.com). The CEO mindset asks one question over and over: where is human capital deployed, and is that the best use of it relative to enterprise value? In a boardroom, I'd say it this way: "We bought hours back with AI. Show me the line item where we reinvested them." ### Kept: the employer captured the gain through pay Apollo's economists compared BLS wage data against observed AI usage across 321 occupations. In the occupations where AI does the largest share of the work, real wage growth after 2023 ran about 6.7 percentage points slower. Apollo found no statistically significant employment effect and concluded firms are capturing the gain through wage compression rather than job cuts ([Apollo Global Management](https://www.apollo.com/content/dam/apolloaem/pdf/daily-spark/2026/jul/30/Whitepaper-Impact%20of%20AI%20on%20U.S.%20Labor%20Market-2026-R2%201.pdf?ref=trentcotton.com)). Before anyone runs with a doom headline, put the numbers in scale. The study covers 5.8 million workers, roughly the population of Wisconsin and about 3.7% of the U.S. labor force, or one worker in 27\. Line those workers up by pay and the gap widens at almost every step, with the bottom wage quartile taking a 10.7% hit. Someone kept the gain, and the evidence points to employers. ### Hidden: the gain exists and nobody wrote it down Culture Amp's first AI at Work benchmark covers about 112,000 employees at 292 organizations, and 71% say AI makes them more productive ([Culture Amp AI at Work Benchmark](https://www.cultureamp.com/company/announcements/ai-at-work-benchmark-2026?ref=trentcotton.com)). The workload numbers barely move: 72% of power users call their workload reasonable, compared with 69% of non users. Power users are 15 points more likely to say their organization motivates them to go beyond what they'd do elsewhere, while awareness of internal career opportunities fell 10 points since July 2025, the largest single year change in the benchmark. Culture Amp doesn't tie that drop to AI, and I won't either. Still, people got faster, they're willing to do more, their workload feels the same, and fewer of them know where to go next. This calls for the Coach mindset, a tougher version than most people picture, which I laid out in [CHRO as Coach](https://www.trentcotton.com/chro-as-coach-the-hard-truth-hr-leaders-need/). Sit down with the leaders pushing AI the hardest and ask three questions. Where is the ROI at each step between the tool and the P&L? What work did you restructure, as opposed to what work did you just make faster? Name one person whose pay or career path changed this year because AI made them measurably better. If a leader can't answer the third question honestly, the first two don't matter. ## How Should C-Suite Leaders Respond? C-suite leaders and CHROs should respond by tracing AI's gains from the individual to the P&L one team at a time, starting with the handoffs where those gains quietly disappear. Here's the playbook I'd run starting Monday. 1. **Audit one handoff.** Use the Engineer mindset, which obsesses over handoffs and gaps because that's where problems pile up quietly. Pick one team and ask: if someone here got measurably better at their job because of AI, where is that written down, and how did it carry through the rest of the process? I dig into this pattern in [Your HR Problem Is Hiding at the Handoff](https://www.trentcotton.com/your-hr-problem-is-hiding-at-the-handoff/). 2. **Tie pay to the gain.** If a role now produces measurably more, the comp conversation belongs in the same meeting. Apollo's data shows the bottom wage quartile absorbing a 10.7% slowdown in real wage growth. A flat band on a role that just got more productive is a retention problem waiting to happen. 3. **Map hours at the task level.** Build a Value of Work Matrix for the same team using task level hours. Ask which quadrant AI pulled time out of this year. If it came out of the low value, low focus box, you have your answer for why the P&L didn't move. The capacity math behind this lives in [Is Headcount Planning Broken?](https://www.trentcotton.com/is-headcount-planning-broken-the-new-fte-math/). 4. **Redeploy before you reduce.** Before any headcount decision, confirm the hours AI freed up have a destination tied to revenue, retention, or client experience. My 75% cut taught me what happens when they don't. 5. **Make the career path visible.** With internal career awareness down 10 points in a year, tell people what the time they saved buys them: new work, new skills, or a new role. People who can't see the next step start looking elsewhere for it. ## What This Means for You as a Leader AI is rewarding the individuals using it and exposing the organizations that can't account for what those individuals produce. The gap between 80% and 37% starts as a measurement problem, and measurement is squarely HR's job. Leaders who trace the hours will make workforce decisions they can defend in front of a board. Leaders who skip the tracing will keep restructuring around a productivity number nobody can find. I go deeper on the CEO, Coach, and Engineer mindsets, along with the other two, in my book "The FutHRist." If you want to see where you stand across all five, the free assessment takes about ten minutes at [trentcotton.com/futhrist-start-here](https://www.trentcotton.com/futhrist-start-here/). Watch the full episode on [The Human Capitalist](https://youtu.be/ba%5FInNECOCk?ref=trentcotton.com) or listen on [Spotify](https://open.spotify.com/show/57xtYAUsIYrg7gQ0a6RFwc?ref=trentcotton.com). And remember to invest in yourself, because no one's going to do it quite like you do. ### Resources for HR Leaders If you found this useful, here's what I've built for HR executives dealing with the same challenges. The FutHRist Planner Free A planning tool for HR leaders navigating AI-driven transformation. [Download →](https://cottontrent.gumroad.com/l/futhrist-productivity-planner?ref=trentcotton.com) AI Ethics in Recruiting Free A framework to evaluate AI risk in your hiring stack. [Download →](https://cottontrent.gumroad.com/l/AIEthicsinRecruiting?ref=trentcotton.com) FutHRist Team Assessment Workbook Diagnose your team's readiness for the future of work. [Get it →](https://cottontrent.gumroad.com/l/futhrist-team-workbook?ref=trentcotton.com) The Value of Work Matrix Toolkit Quantify workforce decisions and defend your seat at the table. [Get it →](https://cottontrent.gumroad.com/l/value-of-work-matrix-toolkit?ref=trentcotton.com) Sprint Recruiting Bootcamp Cut time-to-fill without cutting quality. [Get it →](https://cottontrent.gumroad.com/l/sprint-recruiting?ref=trentcotton.com) FutHRist Team Workshop Kit Run a future-of-work strategy session with your HR team. [Get it →](https://cottontrent.gumroad.com/l/futhrist-team-kit?ref=trentcotton.com) --- ## FAQ ## Is AI improving productivity or profit? Both show up in the data, at very different rates. McKinsey's 2026 survey found 80% of respondents say AI improved their individual productivity, but only 37% attribute any EBIT impact to AI, and that share hasn't changed since 2025. ## Is AI creating or destroying jobs? found real wage growth in highly AI exposed occupations ran about 6.7 percentage points slower after 2023, with no statistically significant employment effect, though McKinsey respondents still expect cuts: 39% anticipate AI related workforce reductions next year. ## Why are workers in AI exposed jobs seeing slower wage growth? Apollo's economists concluded firms are capturing AI productivity gains through wage compression instead of layoffs. The effect is steepest at the bottom, where the lowest wage quartile saw a 10.7% decline in real wage growth relative to low exposure occupations. ## Why don't employees feel less busy if AI saves them time? The saved time gets reabsorbed. Culture Amp found 72% of AI power users call their workload reasonable versus 69% of non users, and Glean's research found 6.4 of the 11 hours AI saves each week go to supervising and fixing its output. ## What is the Spent, Kept, or Hidden test in simple terms? It's a diagnostic from Trent Cotton and The Human Capitalist for tracing missing AI productivity. Spent means the hours went to low value work, Kept means the employer captured the gain through slower wage growth, and Hidden means the gain exists but was never documented. ## What should CHROs do first? Pick one team and trace a single AI gain from the individual to the P&L. Document where it went, check whether pay or career paths changed as a result, and map the team's hours on a Value of Work Matrix to see which quadrant AI pulled time out of.