AI cultural debt: the workforce intelligence blindspot

Deloitte has put a number on a suspicion many people analytics leaders have carried for a while: the technology is the easy part, and the part everyone measures is the part that matters least.

In its 2026 Global Human Capital Trends, built on a survey of more than 9,000 leaders across 89 countries, Deloitte reports that 59% of organisations take a tech-focused approach to AI, and that those organisations are 1.6 times more likely to fall short of above-expectation returns than their human-centric peers.¹ The headline figure comes from supplemental research with 100 C-suite leaders rather than the full sample, and that caveat is worth keeping; the direction it points is consistent with everything else in the report. Technology is now ubiquitous and replicable. The human response to it is neither, which is where the advantage, and the risk, actually sit.

Deloitte's name for the risk is AI cultural debt: the cost an organisation accumulates by deploying AI while neglecting its effect on people, "similar to how financial debt accrues interest". Forty-two per cent of workers say their organisation rarely evaluates AI's impact on people, leaving them to settle unanswered questions on their own. Is it cheating to use AI for this? Who is to blame when it is wrong? The norms that fill that silence are the debt, and only 5% of organisations report great progress on the problem even as 65% say their culture needs to change significantly.²

Counting prompts is the new counting keystrokes

The reason adoption dashboards mislead is that usage and value have come apart. Deloitte found that 80% of leaders, managers and workers are concerned colleagues are using AI to appear more productive than they are.² If four in five people suspect the usage is partly performance, then logins and prompt volumes are measuring the theatre, not the work. Counting prompts is the new counting keystrokes: a precise tally of activity that says nothing about whether anything got better.

The evidence on AI and productivity explains why the same usage number can hide opposite outcomes. A randomised controlled trial of more than 5,000 support agents found a 35% throughput gain for the least experienced quartile and almost none for veterans.³ In Nature Human Behaviour, a meta-analysis of 106 experiments found that human-AI combinations, on average, performed worse than whichever of the two did better alone.³ A separate synthesis pooling 371 estimates found no consistent relationship between AI adoption and aggregate productivity once methodological noise was controlled.³ Two employees can show identical adoption on a dashboard and deliver completely different value. The dividend is conditional; the dashboard treats it as uniform.

This is the Solow paradox in modern dress: the AI shows up everywhere except in the measured output. Almost every company now invests in AI, yet only 1% of leaders believe their organisation has reached maturity,⁴ and 83% of firms report low workforce analytics maturity, so most lack the instrument that would tell them which side of the divide they are on.⁵

What the floor measures, and what it can't

The fairest objection is that adoption metrics are not a trap but a sensible first step. You cannot judge the value of a tool nobody uses, and Wharton's three-wave study of enterprise AI describes firms maturing exactly as good capital allocators should, benchmarking usage and then holding AI to the same return standards as any major investment.⁶ That discipline is real progress, and the point is well taken. Usage is a legitimate leading indicator and the rational floor of measurement. The failure is stopping at the floor. Wharton's own conclusion, that people rather than tools now set the pace, points past adoption to the harder layer above it.⁶

That layer is human, and three signals sit in it. The first is decision quality: who reviews or overrides AI output, and how often the rework quietly cancels the headline speed. The second is flow-state preservation, since the same assistant can protect deep focus or fragment it into prompt-management and notification load, a new source of technostress the meta-analytic evidence flags directly.³ The third is skill-shift velocity: who is genuinely getting better at working with AI, which is the difference between an organisation that compounds its capability and one that plateaus. Trust belongs beside them, because unevaluated AI corrodes it in both directions, and trust is itself a measurable leading indicator.²

A sceptic will note that cultural debt rests largely on perception surveys rather than hard outcomes, and that is a fair hit.¹³ But it cuts the other way. The absence of direct, continuous measurement of AI's human effects is precisely the blindspot the term is naming. If organisations instrumented decision quality, focus and cognitive load the way they instrument logins, claims about AI's impact on people would not have to rest on an annual survey at all. The shift the analytics profession is already making, from volume metrics to impact metrics, is the same move at the level of the whole function.⁵

Visier frames the destination as "the business case for humans": equipping people with AI built to amplify what they do, judged by outcomes rather than adoption.⁷ Deloitte's data makes the case less sentimental than it sounds, since the human-centric approach is the one statistically associated with better returns.¹ So the question for any leader reviewing an AI dashboard this quarter is uncomfortably simple: does it tell you that people are using the tools, or that the tools are making the work better? The gap between those two questions is where the debt is accruing.

Footnotes

1. Poynton, S., Flynn, J., Scoble-Williams, N., Reyes, V., Mallon, D., & Cantrell, S. (2026, March 4). 2026 Global Human Capital Trends. Deloitte Insights. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends.html ↩ ↩² ↩³

2. Flynn, J., Van Durme, Y., Harrington, S., & Reichheld, A. (2026). Dealing with AI's cultural debt. In 2026 Global Human Capital Trends. Deloitte Insights. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2026/ai-cultural-debt.html ↩ ↩² ↩³

3. Gruda, D., & Aeon, B. (2025, October). Seven myths about AI and productivity: What the evidence really says. California Management Review Insights. https://cmr.berkeley.edu/2025/10/seven-myths-about-ai-and-productivity-what-the-evidence-really-says ↩ ↩² ↩³ ↩⁴ ↩⁵

4. McKinsey & Company. (2025). Superagency in the workplace: Empowering people to unlock AI's full potential at work. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work ↩

5. Mallampalli, P. (2026). 10 workforce analytics trends shaping HR in 2026. AIHR. https://www.aihr.com/blog/workforce-analytics-trends/ ↩ ↩²

6. Korst, J., Puntoni, S., & Tambe, P. (2025, October 28). Accountable acceleration: Gen AI fast-tracks into the enterprise (2025 AI Adoption Report). Knowledge at Wharton / GBK Collective. https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report ↩ ↩²

7. Visier. (2025, November 13). The Visier 2026 Trends Report reveals how AI will redefine leadership, workforce strategy, and the human element of business [Press release]. https://www.visier.com/company/news/the-visier-2026-trends-report-reveals-how-ai-will-redefine-leadership/ ↩