How Can Employee Listening Help Us Understand the Value of AI Work?

Employees have adopted generative AI faster than the systems that evaluate their work have adapted. Global spending on AI software is projected to reach $452 billion in 2026 (Gartner), and EY's Work Reimagined survey of 15,000 employees across 29 countries found that 88% use AI at work while only 5% use it in ways that fundamentally change how work gets done. MIT's Project NANDA review of more than 300 AI initiatives found that 95% of generative AI deployments produce no measurable return.

Part of that gap results from a problem organizations rarely name. Professional work has always been priced, evaluated, and rewarded partly on the effort visible in it. A document that took three days to prepare reads as more valuable than the same document produced — generated, one might even say, after noting an em dash or two! — in twenty minutes. The payroll structures that compensate people for that work now encode such assumptions directly. Generative AI breaks the connection between time spent and value delivered, and it breaks it faster than most compensation structures and quality judgments can adjust.

Why Does Work Produced Quickly Get Valued Less?

One of Perceptyx’s cofounders built a proposal generator years before generative AI reached the workplace. It ran on Excel formulas and the mail merge function in Microsoft Office. He would take a prospect call, capture the details in a structured form, and use concatenation to assemble personalized lines of text that mail merge dropped into the right sections of a template. In a matter of minutes, he could produce a 30-page proposal customized to that specific prospect.

We argued about it for a long time. His position was that speed wins deals: respond while the opportunity is live, maintain momentum, and never leave a prospect waiting. The opposing position was that a 30-page proposal arriving five minutes after the call would be read as boilerplate. The recipient would assume nothing in it was written for them, would skim rather than read, and would miss the customization entirely because the delivery time told them not to look for it. The identical document sent two days later would read as evidence that somebody had worked through the night on their behalf.

The quality of the output was not in question in that argument, only the signal attached to it. Effort is a heuristic people apply because it is usually cheap and usually accurate. However, it stops being accurate the moment production cost falls without quality falling with it.

What Happens When Audiences Start Screening for AI?

Recipients have now learned to look for markers of machine-generated text, and they look for them constantly. Automated LinkedIn comments that restate the post above them, sales emails that simulate familiarity with a recipient nobody researched, confident claims with no evidence behind them, and recognizable sentence rhythms have trained a large audience to run a detection pass before an evaluation pass.

That detection habit does not discriminate between careless output and careful output. Work built on real expertise and a genuinely useful perspective can trip the same filters if it was drafted or edited with AI assistance. The reader discounts it before finding the thinking inside it. In other words, poor uses of the technology have raised the cost of good uses.

Underneath lurks a question neither the producer nor the audience can currently resolve: what was the ratio? A finished document might represent a professional who supplied the analysis and the judgment but used a model to tighten the prose. It might equally represent a generic prompt and no review at all. 

This has a direct effect on the people producing it. In the EY data, 38% of employees worry that using AI will erode their own skills and expertise. An employee who suspects their contribution has become invisible to the people evaluating it has a rational reason to limit how much they use the tool, or to hide that they used it. Trust rather than technical skill predicts adoption, and this is one of the mechanisms behind that finding. 

Organizations need standards that separate AI assistance from professional abdication. Disclosure alone will not do it, because disclosure answers whether a tool was used rather than what the person contributed. What employees need is a way to make expertise and accountability visible in the output itself.

What Breaks When Billable Hours Meet AI?

The perception problem has a structural twin inside professional services, consulting, agencies, research, and any function that bills or measures by time. Consider an attorney, consultant, analyst, or researcher who can now complete in one hour a task that previously took ten. Several questions should follow immediately. Most firms have not answered any of them:

• Does the client receive the benefit of the reduced time, or does the firm?
• Does the employee lose billable hours they are measured on?
• Does the firm respond by raising utilization expectations?
• Is the person rewarded for faster delivery or penalized for lower recorded hours?
• Does the efficiency gain create pressure to absorb ten times the caseload?
• How does the firm price expertise once time no longer approximates it?

For the individual, the answer determines whether AI improves their working life or degrades it. A tool that lets someone finish faster while their utilization target holds steady has handed them a problem. They now face a number they cannot reach while doing the work the way the tool allows.

Two distinct issues need separating here. The first is low-quality or unsafe use, of the kind visible in court filings that cited fabricated cases because nobody verified the model's output. The second is high-quality supported work, where a professional uses AI to accelerate research, structure an argument, test assumptions, or sharpen clarity while retaining full accountability for what goes out the door. Both reduce time, yet they produce opposite value. A metric that tracks hours cannot tell them apart.

Which Performance Measures Stop Working First?

Any measure that treats time or volume as a proxy for contribution loses validity as AI compresses production. The exposed set includes:

• Billable hours
• Utilization rates
• Time to completion
• Volume of output or documents produced
• Responsiveness and turnaround speed
• Individual productivity targets calibrated to pre-AI task duration

This mirrors a failure pattern already well documented in learning and development, where organizations measure course completions because completions are easy to count. A completion record proves attendance, not capability, and Perceptyx research finds content utilization below 50% points to a delivery problem rather than a demand problem. Counting the activity was never the same as measuring the result. AI takes that long-standing weakness in activity metrics and makes it acute across far more of the business.

Post-acquisition organizations run into a comparable misalignment when legacy goals and incentive structures survive into a new operating model, and employees receive one message about strategy while their compensation rewards behavior from the old company. Employees resolve that conflict by following the incentive. The stated strategy loses.

What Should Replace Time as a Signal of Value?

Measurement has to move to the parts of the work that AI does not supply. In practice that means weighting outcomes achieved, quality of judgment, client impact, risk identified and managed, and the value created rather than the hours consumed.

Making that shift requires evidence of capability rather than assertions about it. Perceptyx Develop generates that evidence by teaching through adaptive conversation and scoring comprehension in real time against defined learning objectives, which produces a record of what a person demonstrably understands and can apply. Activate reinforces the behaviors that follow in the tools where work already happens. The 2024 randomized controlled trial by Haunstrup and Jensen, run with 226 managers and 4,442 employees, found that training paired with just-in-time nudges produced behavior change still measurable at eight months, where training alone did not.

Human contributions also have to be named before they can be rewarded. The work that remains distinctly human in an AI-supported workflow includes framing the problem correctly, applying context a model does not have, evaluating evidence, challenging weak assumptions, exercising ethical judgment, building trust with the people involved, and accepting accountability for the result. Those contributions belong in performance reviews, career paths, and how work gets assigned. Where they go unrecognized, the organization is asking employees to develop capabilities it declines to reward.

How Do You Tell If Your Measures Have Stopped Working?

Useful employee survey items to field include:

• Our performance measures reflect the way work now gets done.
• I am recognized for the judgment and expertise I bring, not only for volume.
• Using AI to work faster does not put my performance metrics at risk.
• The organization recognizes the human expertise behind AI-assisted work.
• I can see a positive future for myself in the organization.
• I understand what responsible AI use looks like in my specific role.

Run those items alongside the readiness and confidence measures in a standard AI adoption baseline, then segment by function and by whether the group is measured on time. The pattern to look for is a population that reports strong AI capability and weak confidence that the organization will value what they produce with it. That combination predicts quiet non-adoption, and no license report will show it.

Redesigning measurement before scaling AI is cheaper than reversing an incentive structure employees have already learned to work around. The organizations that get this right will be the ones that decided what they were paying for before the technology answered the question for them.

Ready to Find Out Whether Your Measures Still Measure Value?

Schedule a meeting with our team to see how employee listening can identify where your performance systems have fallen out of step with how work now gets done.

For the full business case on measuring capability instead of activity, including the evidence base behind validated comprehension, read Beyond Course Completions: From Fragmented HR to an Intelligent Development Engine. For global data on how employees are experiencing AI at work, read Beyond the Hype: Global Employee Perspectives on Generative AI.