AI ROI Pressure: When AI Theater Replaces the Work (2026)

AI ROI pressure is the demand, flowing from boards to executives to employees, to show measurable returns on AI spending now. In 2026 that pressure has produced something stranger than failure: companies performing AI adoption instead of practicing it.

The evidence comes from the executives themselves. In WRITER’s 2026 enterprise AI survey, run with the research firm Workplace Intelligence across 1,200 C-suite executives and 1,200 employees, 75% of executives admitted their company’s AI strategy is more for show than for actual internal guidance.

Read that again slowly. Three out of four of the people responsible for AI strategy say, in private, that the strategy is theater.

I wrote in my hub post on gamification and AI that when intelligence becomes cheap, human motivation becomes the scarce input. AI theater is what happens when a company spends on the intelligence and then measures the humans on how convincingly they appear to use it.

This post is my field manual for that trap. Through the Octalysis Framework, I will show you why good people perform adoption they do not believe in, why the usage dashboards keep rising while the value does not, and how to redesign the incentives so the proving stops and the producing starts.

Speed Run Notes

• The theater is confessed. WRITER’s 2026 survey found 75% of executives admit their AI strategy is more for show than actual guidance, while 59% of companies pour at least $1 million a year into AI.
• The returns are missing. MIT’s GenAI Divide report found 95% of enterprise AI pilots deliver no measurable P&L impact, and S&P Global measured 42% of companies abandoning most AI initiatives in 2025.
• Workers are performing too. AI-polished “workslop” hits 40% of US desk workers monthly at a cost of roughly $186 per employee, while 48% hide real AI use from their managers.
• Octalysis explains the spiral. Mandates run on Core Drive 8 fear and Core Drive 5 conformity, which reward visible usage; Core Drive 1 purpose, which rewards real value, gets crowded out.
• The fix is measurement design. Score shipped outcomes instead of usage counts, make honest non-use safe to admit, and let teams own the problems AI solves rather than the tools it showcases.

Author Credibility: Yu-kai Chou

Yu-kai Chou created the Octalysis Framework after studying gamification since 2003 — years before the term entered mainstream vocabulary. As a Human-Systems Architect & Behavioral Designer, his framework has been applied by LEGO, Microsoft, Porsche, Coca-Cola, Salesforce, and MrBeast, impacting over 1.5 Billion Users.

Chou has taught the Octalysis methodology at Harvard, Stanford, Yale, Tesla, Google, BCG, and IDEO.

His work has been cited by Harvard, Stanford, MIT, Forbes, Wall Street Journal, Wired, US Department of Energy, NIST, NSF, NCBI, US Department of Education, ClinicalTrials.gov, and Google Scholar — with 3,700+ more academic publications. Explore his books here.

The Confession Hiding in the Survey Data

Start with the money. WRITER’s survey, fielded between December 2025 and January 2026 across the US and Europe, found 59% of companies investing at least $1 million a year in AI.

Then look at what the same executives say the money produces. Only 29% report significant returns from generative AI, and only 23% from AI agents.

Nearly half of the executives surveyed, 48%, went further and described their company’s AI adoption as a massive disappointment. And 39% admitted their company has no formal strategy for generating revenue from AI tools at all.

Hold those numbers next to the confession I opened with. A company that spends a million dollars a year, sees no significant return, and has no revenue strategy still needs its AI story to look good, because the board is asking, the market is asking, and every earnings call now contains the two letters.

So the strategy becomes a costume. The 75% who call their own strategy a show are describing the gap between what the company performs and what it practices.

The independent research says the gap is real. MIT Project NANDA’s 2025 GenAI Divide report, built on 52 executive interviews, 153 senior-leader surveys, and more than 300 public AI initiatives, found that despite $30 to $40 billion in enterprise spending, 95% of AI pilots deliver no measurable P&L impact.

And the abandonment data shows the pattern compounding. S&P Global Market Intelligence surveyed more than 1,000 organizations across North America and Europe in 2025 and found the share abandoning most of their AI initiatives had jumped to 42%, up from 17% a year earlier, with the average organization scrapping 46% of its AI proofs-of-concept before production.

Billions in, almost nothing measurable out, nearly half the projects in the bin. That is the stage on which AI theater plays.

What Is AI ROI Pressure?

AI ROI pressure is the organizational force created when leadership must justify large AI investments on short timelines, and the easiest thing to show is activity rather than outcomes. Under it, visible adoption becomes the product: usage dashboards, mandated tool rollouts, AI mentions in strategy decks, and employees demonstrating that they use AI whether or not it helps.

Notice what the pressure selects for. Real returns are slow, lumpy, and hard to attribute, while usage metrics are instant, clean, and chartable.

A leader squeezed for proof reaches for the chartable thing. The moment usage becomes the proof, everyone below that leader starts optimizing usage, and the theater begins.

The pressure arrives from three directions at once. Markets and boards expect every company to have an AI transformation story, and punish the stock or the CEO who lacks one.

Vendors supply the second push, selling transformation on timelines no integration has ever met, so the gap between the purchase announcement and the value has to be filled with something. Visible activity fills that gap right on schedule.

The third direction is peer benchmarking. When a rival announces an AI initiative, your board asks why you have none, regardless of what the rival’s initiative actually returns.

None of these audiences can easily observe value, but all of them can observe performance. An organization optimizes for the audience it has.

In my work on motivation design, I call this pattern a feedback system pointed at the wrong behavior. The gamification lens makes it legible: every workplace is already a game, with points, visible scoreboards, and win conditions, and the only question is whether the game rewards what the organization actually needs.

AI ROI pressure builds a game where the win condition is being seen adopting. People are winning that game right now, in companies everywhere, and the 95% pilot-failure number is the score.

Workslop: The Costume Department of AI Theater

What does performed adoption look like at the level of a single document? Researchers at BetterUp Labs and the Stanford Social Media Lab gave it a name in September 2025: workslop, AI-generated content that looks polished and complete but lacks the substance to move the work forward.

Their survey of 1,004 full-time US desk workers found that 40% had received workslop in the past month. Each piece cost the recipient an average of one hour and 51 minutes to untangle, because the burden of making sense of hollow work transfers downstream to whoever receives it.

The researchers priced the tax at roughly $186 per employee per month. For an organization of 10,000 people, that compounds to over $9 million a year in lost productivity.

The relational damage ran deeper than the hours. About half of recipients viewed the workslop sender as less creative, capable, and reliable than before, 42% saw them as less trustworthy, and over a third saw them as less intelligent.

Workslop is what an employee produces when the incentive says “show you used AI today” and the work itself never gets a vote. The sender wins the adoption game, the receiver pays two hours, and the trust between them quietly erodes.

Notice who absorbs the most of it: managers. In the same study, 54% of managers reported receiving workslop, against 38.5% of individual contributors, which means the people responsible for judging whether AI is working are the ones most buried in evidence manufactured to look like it is.

That detail should worry every executive reading an adoption report. The layer of the organization that feeds your dashboard is the layer drowning in the performance.

I covered the individual version of this illusion in my post on AI coding assistant productivity, where developers felt dramatically faster while measurably slowing down. Workslop is the same perception gap, weaponized into deliverables and passed to a colleague.

Core Drive 8: Adoption as a Loyalty Test

Why would a reasonable employee produce hollow work on purpose? Because the alternative feels dangerous, and the danger is explicit.

In the same WRITER survey, 60% of executives said they plan to lay off employees who can’t or won’t use AI. That is the sharpest possible activation of what I call Core Drive 8 (CD8): Loss & Avoidance, the motivational drive to act out of fear of losing what you have.

CD8 sits in the Black Hat region of the Octalysis octagon, alongside Scarcity and Unpredictability. As I explain in my guide to White Hat versus Black Hat gamification, Black Hat drives are powerful and fast-acting, and they make people feel anxious, controlled, and eager to escape.

When keeping your job depends on being seen using AI, the rational move is to be seen using AI. Whether the usage helps anyone is a separate question the incentive never asks.

The behavioral economics underneath is well mapped. Kahneman and Tversky’s Prospect Theory established that losses weigh roughly twice as heavily as equivalent gains, so “adopt or be laid off” moves people far harder than “adopt and maybe improve.”

Fear-driven adoption does move the dashboard. It also produces exactly the behaviors the dashboard cannot see: 29% of employees in WRITER’s survey, including 44% of Gen Z respondents, admitted to actively sabotaging their company’s AI strategy, through moves like refusing tools, feeding data into unapproved apps, or quietly undermining rollouts. Among executives, 76% called that sabotage a serious threat to their company’s future.

A workforce performing compliance on the surface while resisting underneath is the signature output of an all-Black-Hat design. I made the structural version of this argument about the overseers themselves in agentic AI governance: when a role runs on pure CD8 liability, the people you most need in it are the first to route around it.

Core Drive 5: Everyone Is Watching Everyone Perform

Fear explains the floor of AI theater. Conformity explains its ceiling, and conformity belongs to Core Drive 5 (CD5): Social Influence & Relatedness, the drive that moves us to act based on what other people do, expect, and appear to value.

CD5 contains mentorship, companionship, and healthy competition. It also contains conformity anchors: the quiet pull to match whatever behavior the group visibly rewards.

In an AI-theater workplace, the visible behavior is enthusiastic adoption. Slide decks mention copilots, standups feature AI anecdotes, and the super-users get celebrated in all-hands meetings.

So people match the display they are shown. The strangest evidence for this is that the performance runs in both directions at once.

Slack’s Workforce Index, fielded in August 2024 across 17,372 desk workers in 15 countries, found that 48% would be uncomfortable admitting to their manager that they used AI for common workplace tasks. Among the uncomfortable, 47% felt that using AI was cheating, 46% feared being seen as less competent, and 46% feared being seen as lazy.

Put the two surveys side by side and the absurdity of the theater comes into focus. Some employees fake using AI to look loyal, while others hide real AI use to look diligent, and both groups are managing impressions instead of outcomes.

MIT’s GenAI Divide data shows where the hidden half goes: only 40% of companies had purchased an official LLM subscription, yet workers from over 90% of companies reported regularly using personal AI tools for work. The real adoption happens in the shadows, invisible to the very dashboards the theater was built to feed.

When a company’s measured AI usage and its actual AI usage are two different populations, no executive can steer. That measurement fog, far more than any model limitation, is what AI ROI pressure buys with its mandates.

Middle managers live at the point where the two performances collide. From above comes the mandate to show team adoption; from below comes a stream of polished deliverables they are too pressed to verify.

So the manager becomes both audience and cast. They forward the team’s AI numbers upward, present the wins in the monthly review, and privately spend their evenings untangling the workslop the numbers rode in on.

WRITER’s data catches the organizational version of this strain: 78% of companies report tension between IT and other lines of business over AI. When the official channel and the actual work disagree, every layer in between absorbs the friction.

Even the celebrations feed the loop. Spotlighting super-users is deserved recognition, and the survey shows those super-users are roughly three times more likely to have earned a promotion and a raise.

But in a theater culture, a spotlight is also an instruction. What gets rewarded in public becomes what gets performed in private, and colleagues who watched the ceremony learn that the fastest route to the stage is to look like the person on it.

Core Drive 1: What the Theater Crowds Out

The saddest casualty of AI theater is the drive that should power the whole transition: Core Drive 1 (CD1): Epic Meaning & Calling, the motivation that comes from being part of something bigger than yourself.

Here is what makes the loss so unnecessary: the purpose case for AI at work is easy to make honestly. A nurse who spends less time on documentation and more with patients, a teacher freed from grading mechanics, an engineer who ships the fix twice as fast. Each is a CD1 story where the technology serves the mission.

Employees can tell the difference between that story and its costume. When the stated purpose is “transform how we serve customers” but the measured behavior is “log your copilot sessions,” people believe the measurement, because the measurement is what pays.

Self-Determination Theory, which I map against Octalysis in my guide to Ryan and Deci’s motivation framework, predicts exactly this crowding-out: when controlling external pressures dominate, intrinsic motivation and autonomous engagement shrink.

I documented the retention consequence in the first spoke of this series, on AI and employee motivation: the employees most fluent with AI are often the first to leave, because fluent people have options, and purposeless pressure is easiest to escape when you have options.

AI theater accelerates that exodus. Your best people did not join to win a performance game, and they are precisely the ones who can see it is a performance game.

There is a second-order cost that compounds quietly: cynicism. Once an employee has watched one purpose claim turn out to be a costume, they discount the next one, and Core Drive 1 (CD1): Epic Meaning & Calling is uniquely hard to re-ignite after it has been faked at someone.

This is why I tell leadership teams that an honest “we are experimenting and most experiments will die” beats an inflated “we are transforming.” The modest claim survives contact with reality, and surviving claims are the only ones that build CD1 over time.

Goodhart’s Law Comes for the Adoption Dashboard

Everything above is a modern staging of a very old finding. In 1975, the economist Charles Goodhart observed that any statistical regularity tends to collapse once pressure is placed on it for control purposes.

The compressed phrasing everyone now cites as Goodhart’s Law spread through the anthropologist Marilyn Strathern’s 1997 paper on university audit culture: once a measure becomes a target, it stops being a good measure.

AI usage metrics were decent measures, once. In early 2023, a team spontaneously using AI probably had found real value, so usage correlated with value.

Then the boards started asking, the mandates arrived, and usage became a target. The correlation died on contact, exactly as Goodhart predicted, and now a rising adoption curve can mean value creation, workslop production, fear compliance, or résumé theater, with the dashboard unable to say which.

Here is the practical test I give executives. If your AI metrics can be improved by behavior that creates zero value, some of your measured improvement is that behavior, and under layoff-grade pressure, much of it is.

Run your current dashboard through that test honestly. Logins can be gamed by opening a tab, prompts-per-day can be gamed by trivial prompts, AI-assisted documents can be gamed by workslop, and training completions can be gamed by clicking through modules on a second monitor.

A metric that survives the test, like a verified drop in cycle time on a named workflow, is rarer and slower. It is also the only kind that tells you anything once the mandates start.

Every row has the same shape. A Black Hat pressure meets a measurable proxy, the proxy gets optimized, and the mission pays the bill.

The Octalysis Redesign: From Proving to Producing

Diagnosis without redesign is just better-informed despair, so here is the build. The goal is to move AI motivation from the Black Hat bottom of the octagon to the White Hat top, where Core Drive 1 (CD1): Epic Meaning & Calling, Core Drive 2 (CD2): Development & Accomplishment, and Core Drive 3 (CD3): Empowerment of Creativity & Feedback live.

1. Retire usage metrics as success metrics

Keep usage data for capacity planning if you like, but never present it as achievement. The moment a usage number appears in a performance review or an all-hands slide, you have re-armed the theater.

Replace it with outcome metrics owned by each team: cycle time on real deliverables, error rates, customer resolution speed, revenue per workflow. These are harder to chart and nearly impossible to fake, which is precisely their value.

A support organization I would design this way does not report “copilot sessions per agent.” It reports median resolution time and reopen rate, and lets any agent abandon the copilot the day those numbers say it is losing.

2. Make honest non-use safe, and honest use safer

Declare an explicit amnesty in both directions. An employee who says “AI makes this task worse” should be rewarded for the measurement, and an employee who automated half their job should be celebrated rather than quietly given more work at the same pay.

Remember the Slack finding: 48% of workers hide AI use because it feels like cheating or looks lazy. You cannot govern what people hide from you, and people hide what they expect to be punished for.

The amnesty has to be structural, with the gains shared. If automating a workflow earns the automator a piece of the saved time, say one reclaimed afternoon a week for learning or deep work, disclosure becomes the winning move.

If automation earns a silent increase in quota, concealment stays rational forever. Employees run this math instantly, and they are rarely wrong about it.

3. Let teams pick the problem, then the tool

Core Drive 3 (CD3): Empowerment of Creativity & Feedback activates when people choose their approach and see the results of their choices quickly. Invert the usual rollout: instead of assigning a tool and demanding usage, assign a problem worth solving and let the team decide where AI fits.

My Strategy Dashboard process forces the discipline here: define the business metrics, the users, and the desired actions before a single feature or tool is named. Teams that run it stop asking “how do we prove adoption” and start asking “which desired action does this tool amplify.”

4. Tell the CD1 story with receipts

Purpose claims need evidence or they curdle into more theater. If the mission is “AI frees us for higher-value work,” show the freed hours landing somewhere meaningful: the support engineer now doing proactive outreach, the analyst now testing hypotheses instead of formatting them.

One true, specific, named story per quarter beats fifty adoption charts. People extend belief to what they can verify.

The receipts requirement also disciplines the storytellers. A communications team that must find a real, checkable example each quarter will push the program toward producing one, which quietly converts the marketing function into a quality gate.

5. Build a CD2 ladder for real capability

Core Drive 2 (CD2): Development & Accomplishment needs a progression system with real rungs. WRITER’s data shows the raw material exists: the survey’s AI super-users reported saving around nine hours a week versus two for the least engaged, and were roughly three times more likely to have earned both a promotion and a raise in the past year.

Make the ladder about demonstrated outcomes, with certification tied to shipped improvements rather than completed modules. A badge for finishing training is decoration; a rank earned by cutting a process from five days to one is accomplishment.

6. Give executives a stage worth standing on

The 75% confession exists because boards reward AI optics. Executives need a replacement script: report the portfolio honestly, including the 46% of pilots that die, framed as cheap experiments that bought knowledge.

The companies on the right side of MIT’s divide, the 5% extracting real value, earn the only AI story that compounds. Honest small wins now beat impressive vapor later, because vapor gets audited eventually, by S&P if not by your own CFO.

What to Measure Instead of Usage

If you strip the adoption dashboard, what goes in its place? Instrument the outcomes and the honesty, in roughly this order.

First, task-level before-and-after: pick ten high-volume workflows, baseline them, and measure completion time and quality after AI enters, with the team’s agreement on the definition of quality. This is slow and unglamorous, which is how you know it is measurement rather than theater.

Second, downstream burden: survey recipients, the way the workslop researchers did, about how much arriving work needs rework. A falling rework number is among the cleanest signals that AI output is real.

Third, the honesty gap itself: compare officially sanctioned AI usage against anonymously self-reported usage. The two converging means your people no longer need to hide, and that convergence is a leading indicator for every other number you care about.

Fourth, retention of your AI-fluent people, the canary metric from the first spoke of this series. The most capable adopters leaving while the usage dashboard rises is the theater announcing itself.

Fifth, track the kill rate and report it with pride. A portfolio where nothing ever gets cancelled is a portfolio nobody is measuring, so count how many experiments died, what each one cost, and what it taught.

The S&P data makes the point for you: the average organization already scraps 46% of its AI proofs-of-concept. The difference between a healthy lab and a theater is whether those deaths are cheap, early, and instructive, or expensive, late, and hidden from the board.

None of these metrics fit neatly on an earnings-call slide. All of them predict whether the million-dollar annual spend becomes the 29% that sees returns or the 42% that becomes abandonware.

And one warning as you instrument: measurement itself can become theater. A company that spends six months building the perfect AI-ROI measurement framework, complete with steering committee and vendor, has found one more way to perform rigor instead of practicing it, so start with two workflows and a spreadsheet this quarter.

FAQ: AI ROI Pressure

What is AI ROI pressure?

AI ROI pressure is the demand on companies and their employees to show measurable returns on AI investment quickly, which in practice pushes organizations to optimize visible adoption signals like usage dashboards and mandated tool rollouts. Because real returns are slow and hard to attribute while usage is instant and chartable, the pressure frequently produces performative adoption, what this post calls AI theater, rather than value.

Why would executives admit their AI strategy is for show?

In WRITER’s 2026 survey of 1,200 C-suite executives, conducted with Workplace Intelligence, 75% said their AI strategy is more for show than for actual internal guidance, and 48% called their AI adoption a massive disappointment. The candor reflects a squeeze: boards and markets expect an AI story, only 29% see significant generative AI returns, so the story gets performed while the strategy catches up.

What is workslop and how is it connected?

Workslop, a term from BetterUp Labs and Stanford Social Media Lab researchers, is AI-generated work that looks polished but lacks the substance to advance the task, forcing the receiver to redo or decode it. Their September 2025 survey found 40% of US desk workers received workslop in the prior month, at roughly $186 per employee in lost productivity. It is the individual-level symptom of AI theater: output produced to display adoption rather than to move work forward.

How should companies measure AI ROI instead of usage?

Baseline specific workflows and measure completion time and quality after AI enters them, track how much received work needs rework, compare sanctioned usage against anonymous self-reports to find the honesty gap, and watch retention of AI-fluent employees. Usage counts fail as success metrics because they can be inflated by valueless behavior, which is Goodhart’s Law applied to adoption dashboards.

Does mandating AI use work?

Mandates reliably move usage numbers and just as reliably trigger resistance: WRITER’s survey found 60% of executives plan layoffs tied to AI adoption, while 29% of employees admit to sabotaging their company’s AI strategy. Fear-based Core Drive 8 designs produce fast, anxious, short-lived compliance, so a mandate can open the door to AI, but only ownership, capability-building, and honest purpose keep people walking through it.

Strike the Set, Keep the Work

The theater metaphor matters because theaters are expensive. Sets, costumes, rehearsals, and an audience whose time gets spent watching, with the WRITER, MIT, and S&P numbers as the box-office receipts: millions spent performing adoption that returns little.

My position after two decades of motivation design is that the fix costs less than the show. Companies do not need more AI enthusiasm; they need measurement that cannot be gamed, safety for honest signals, ownership of real problems, and a purpose story backed by receipts.

And the timing favors the honest. With 42% of companies abandoning most of their AI initiatives, the next budget cycle will be unforgiving to programs whose only evidence is a usage curve, while a team holding before-and-after numbers on ten real workflows will sail through the same review.

The strange comfort in the 75% statistic is that the executives already know. Nobody has to convince them the strategy is a costume; they said so themselves, and the ones who act on that admission first will inherit the talent, the trust, and the returns the theater kept promising.

If you want the full system for designing motivation that runs on purpose and accomplishment instead of fear and optics, my books lay out the complete framework, from the 8 Core Drives to the design process I use with client organizations. Explore my books here and build something worth dropping the act for.

Related Reading

• AI Companion Apps, Loneliness, and the Way Back to Humans: the most personal arena where designed motivation either serves people or performs for them.
• Agentic Commerce: Designing for Buyers Who Aren’t Human: what motivation design becomes when the AI is the customer.
• AI in Education Motivation: What Should Count as Learning: the same evidence crisis, staged in classrooms instead of boardrooms.
• AI Morale Overload and Agent Burnout: what the always-on AI workplace does to the humans inside it.
• Buridan’s AI: Pick a Main to Escape AI Tool Paralysis: why tool sprawl fragments the attention your AI strategy depends on.

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?