AI is Breaking Time-Based Pay. Here's What the Evidence Says Comes Next.
Time is the default unit of measurement for work, but AI is disrupting this model: it lets professionals produce more in less billed time, and it lets salaried employees finish contracted hours of work in a fraction of the time. But sadly, neither group is being rewarded for that. Instead, research shows people are hiding their productivity from clients, from managers, and from each other, because the systems built around visible time haven't caught up with what AI can actually do. Outcome-based alternatives, in pricing and in how work itself is structured, are becoming the credible solution, as the evidence for them is growing.
This is the second piece in a two-part series on time as the unit of work; part one looked at where the billable hour and the fixed working week came from, and why measuring time instead of output has always disadvantaged women specifically.
What is AI doing to the time-based pay model, right now?
AI attacks the hours-as-work-proxy model from two directions simultaneously, and the emerging evidence shows people responding by hiding the gains for fear of being penalised for working less, instead of being rewarded for them.
The optimistic case is now coming from the top. Jamie Dimon has predicted AI could bring a three-and-a-half or four-day workweek within decades, and Bill Gates has floated a two-day week powered by AI abundance. But this is not the first time a productivity leap has come with that promise. In 1930, John Maynard Keynes predicted a 15-hour working week within a century; a hundred years of electrification, computing, and the internet moved the average American workweek from 38 hours to 34 (Schawbel, 2026). Workforce researcher Dan Schawbel, who has tracked four-day-week pilots for seven years, finds the same pattern already repeating with AI: his own research with GoTo found employees are recovering more than two hours a day through AI tools, over 10 hours a week, but argues efficiency gains under current incentive structures flow to output, not to workers' calendars (Schawbel, 2026).
The scale of what AI actually changes can now be rigorously measured. The most cited field experiment on this, a pre-registered randomised trial led by Harvard's Fabrizio Dell'Acqua with 758 Boston Consulting Group consultants (roughly 7% of BCG's individual-contributor workforce), found that on tasks within AI's capability, consultants using GPT-4 completed 12.2% more tasks, worked 25.1% faster, and produced work rated over 40% higher in quality by human graders. AI also acted as a skill leveller: consultants who scored in the bottom half on a pre-study baseline assessment saw the largest performance gains, more than double the improvement seen among top-half performers (Dell'Acqua et al., 2025). That is the underlying scale of time compression professional services pricing now has to account for.
In billed professional work, clients aren't reliably capturing that saving: a 2025 survey of roughly 650 in-house counsel (ACC–Everlaw) found nearly 60% saw "no noticeable savings yet" from outside counsel's AI use (Bloomberg Law, 2025). Separately, research on "shadow AI" in mid-sized firms found nearly 60% of fee-earners admit to using unapproved AI tools, while 68% of firm leaders wrongly believe they have full visibility into it (Legal Futures, 2026).
In salaried work, the same concealment behaviour shows up at much larger scale, and it long predates AI. Microsoft and LinkedIn's 2024 Work Trend Index, one of the largest annual surveys of its kind, found 78% of AI users globally now bring their own, unsanctioned AI tools to work rather than using company-sanctioned ones. The bring-your-own-AI ("BYOAI") behaviour is a strong indicator that AI adoption is already running ahead of what formal measurement systems can see (Microsoft/LinkedIn, 2024).
Why do AI productivity gains stay hidden?
Layered on top of this is a broader, well-documented pattern of "performative work". In a 2026 survey of 1,003 US professionals, 66% admitted to staying visibly active after their actual work was finished, and 73% of managers admitted doing the same for their own bosses. Workers cited "workplace culture" and "managers who measure presence over output" as the leading causes - far more than any specific fear of AI - and 71% said they would log off immediately if there were no consequences for doing so (Software Finder, 2026).
Future of Work Strategist Nirit Cohen, writing in Forbes on that same survey data, draws out the sharper point for employers: AI is freeing up employees' time, and companies are giving them every reason to hide it. Her analysis argues the deeper failure isn't the concealment itself but what organisations do with freed-up time once AI creates it: rather than redirecting it toward learning, higher-value work, or genuine capacity, most workplaces simply fill it with more busywork, so employees have learned that revealing efficiency gains gets them a heavier workload, not recognition. She also raises a sharper, unresolved fairness question: a 2026 lawsuit filed by 26 Meta employees alleges the company used AI-token-usage dashboards and activity data, alongside other performance signals, in selecting people for layoffs, disadvantaging employees on protected medical, parental, or family leave who couldn't accumulate the same visible activity signals. Meta disputes this characterisation and says workforce decisions used documented, neutral criteria; the claim has not been established in court. But the case illustrates exactly the risk this article is pointing at: once "visible activity" becomes a proxy metric that feeds into real decisions, it inherits every bias that visibility-based measurement has always carried, including against people who cannot be "seen" working at all (Cohen, 2026).
There is a second, self-defeating loop layered on top of this. Mark Ma, professor of business administration at the University of Pittsburgh, analysed millions of employee reviews alongside hundreds of AI investment and layoff announcements by US public companies over five years, and found that as AI investment announcements rise, so do AI-linked job cuts, with employee sentiment toward AI turning sharply negative specifically after those layoff announcements. That negative sentiment is statistically associated with lower, not higher, firm productivity: employees who fear AI will replace them resist adopting it, undermining the efficiency gains the layoffs were meant to capture. Ma also found that stock market reactions to these layoff announcements averaged close to zero, suggesting investors don't reliably believe the strategy works either, and that roughly 90% of executives (per a related Atlanta Federal Reserve study Ma cites) believe AI has not yet measurably boosted productivity at their companies despite heavy investment (Ma, 2026).
Put together: whether the incentive is protecting a billable target or protecting a job, the pattern is consistent. People conceal exactly the productivity gain that AI investment is supposed to surface, because the measurement system still rewards time and visibility, not outcomes.
What does the evidence say about outcome-based alternatives?
The alternative to measuring time is measuring output, and this is no longer a theoretical proposal; it has substantial, credible data behind it in both pricing and work design.
In professional services pricing: flat-fee and value-based billing adoption is measurably rising. Clio's Legal Trends Report (survey base of 1,000+ legal professionals annually) found flat-fee use up 34 percentage points compared to 2016, and 59% of firms now use flat fees exclusively or alongside hourly rates. Interestingly, firms using flat fees are getting paid roughly twice as fast and closing matters nearly three times faster than hourly-billed ones (Clio, 2024).
In work design, the strongest evidence comes from reduced-hours, outcome-focused trials. The largest published to date is a July 2025 study by Boston College sociologists Wen Fan and Juliet Schor, covering 2,896 employees across 141 organisations in six countries against a 12-company control group. It found improvements in burnout, job satisfaction, and mental and physical health, with gains driven by better work ability, reduced fatigue, and fewer sleep problems. Individual-level reductions in hours tracked more closely with well-being gains than company-wide reductions did, suggesting the mechanism is genuinely about people getting more control over their own time, not simply a blanket policy change (Fan et al., 2025). The UK's 2022 pilot, run with researchers from Cambridge and Boston College and funded by the ESRC's Digit research centre, found no loss of productivity across 61 organisations and 89% of participating organisations still running the policy a year later (UKRI, 2024).
A related, longer-running body of research looks specifically at Results-Only Work Environments (ROWE), where employees are evaluated purely on output rather than hours or location. A study of ROWE's rollout at a Fortune 500 corporate headquarters found significant reductions in employee turnover, regardless of employees' gender, age, or family life stage, alongside reduced negative work-home spillover (Moen, Kelly & Hill, 2011). This is the most direct evidence available that replacing a time-based measure with an outcome-based one doesn't just fail to hurt productivity: it specifically closes some of the gap created by flexibility stigma, because there is no longer a visibility penalty to overcome.
It must be said that neither model is a simple swap. Value-based pricing shifts delivery risk onto the provider and is harder to scope accurately. Outcome-based work models require managers to define and evaluate actual output, which is a genuinely harder management skill than watching who's logged in. But the direction of the evidence is now consistent: measuring what people produce, not how long they're visibly present, does not cost productivity, and in most of the published trials, it improves it.
What this means in practice
For organisations navigating AI adoption, and for anyone pricing or evaluating their own work, three conclusions follow directly from the evidence in this series:
Time was always a proxy, not a measure of value, whether the context is a billed hour or a contracted working week. It was adopted because it was easy to observe, not because it was accurate, and the research on flexibility stigma and the "ideal worker" norm shows it never measured commitment either.
AI has removed the last justification for that proxy. When AI compresses the time a task takes without compressing its value, systems that still reward visible time actively punish efficiency and predictably produce concealment, whether that's a professional hiding AI use to protect a billing target, or an employee staying "active" to protect their job.
The organisations capturing AI's actual value are the ones redesigning how work is measured and paid for, not just adding AI on top of an hours-based system and hoping the numbers work out. The data on outcome-based alternatives, in both pricing and work design, is now robust enough to act on.
From AI time savings to fair pricing: where should organisations start?
The evidence above points to specific, testable actions rather than a single overhaul, and the same research already cited in this piece offers a rough sequence for how to begin.
▪️ Audit what's already being hidden. With nearly 60% of fee-earners at mid-sized firms admitting to unapproved AI use, and 78% of employees globally already bringing their own AI tools to work without sanction (Microsoft/LinkedIn, 2024), assuming your organisation has no "shadow AI" problem is itself a risk. The lowest-cost first step is finding out how much time AI is already saving inside the business, not assuming the answer is zero.
▪️ Decide what freed-up time is for, before you free it up. Cohen's analysis is as much a warning as a diagnosis: filling reclaimed hours with more busywork is exactly what teaches employees to conceal their gains in the first place. Naming, in advance, whether saved time goes toward learning, deep work, or genuinely reduced hours removes the incentive to hide it.
▪️ Pilot outcome measurement at small scale before scaling it. The UK's 2022 four-day week pilot and the ROWE research both point the same way: individual-level control over time use tracks with the largest wellbeing gains, and productivity didn't fall in either case. A contained pilot, in one team or one billing model, carries less risk than an organisation-wide switch and gives you your own data to make the case internally.
▪️ Revisit pricing and performance measurement together, not separately. Clio's data shows flat-fee billing already outpacing hourly rates in law, but a pricing change without a matching change to how staff are evaluated internally just relocates the same incentive problem. The organisations capturing genuine value from AI are treating pricing and performance measurement as one decision, not two.
None of this requires waiting for a definitive future-of-work model to emerge. It requires treating the evidence already available, on concealment, on outcome-based trials, on pricing shifts, as reason enough to test something now, rather than waiting for time-based measurement to fail more visibly than it already has.
This is part two of a two-part series. Part one looked at where the billable hour and the fixed working week came from, the "ideal worker norm," and why time-based measurement has always disadvantaged women specifically.
Sources
Schawbel, D. (2026): AI Will Not Give You Fridays Off, Fortune
Dell'Acqua, F. et al. (2025): Navigating the Jagged Technological Frontier, Organization Science
Bloomberg Law (2025): AI Does Little to Reduce Law Firm Billable Hours, Survey Shows
Legal Futures (2026): Containing the spread of unapproved AI
Microsoft/LinkedIn (2024): Work Trend Index
Software Finder (2026): The Productivity Performance Trap: Why Employees Pretend To Work After Finishing Early (survey of 1,003 full-time US professionals via CloudResearch Connect, May 2026, ±3.6% margin of error)
Cohen, N. (2026): AI Productivity: Why 66% Of Employees Hide Time Savings, Forbes
Ma, M. (2026): Layoffs tied to AI hurt worker productivity, and the reason may surprise managers, The Conversation
Clio (2024): Clio's latest Legal Trends Report reveals major shift
Fan, W., Schor, J. B., Kelly, O. & Gu, G. (2025): Work time reduction via a 4-day workweek finds improvements in workers' well-being, Nature Human Behaviour 9, 2153-2168 (DOI: 10.1038/s41562-025-02259-6)
UKRI (2024): Making the case for a four-day working week (Cambridge/Salford/Digit research)
Moen, P., Kelly, E. L. & Hill, R. (2011): Does Enhancing Work-Time Control and Flexibility Reduce Turnover? A Naturally Occurring Experiment, Social Problems 58(1), 69-98
Comments