A 10,000-person organisation is quietly losing £6 million a year to a problem most executives do not know exists. Workers call it “workslop” — AI-generated output that looks polished on the surface but is so flawed that colleagues spend more time correcting it than they would have spent doing the work themselves. And the workers producing it are often doing exactly what their leaders told them to do.
The productivity illusion at the top
Something strange is happening inside the companies that have bet hardest on generative AI. A recent survey of 5,000 white-collar workers, reported by The Guardian this week, captured the disconnect starkly: 92% of senior executives say AI makes them more productive, whilst 40% of non-managers report that AI saves them no time at all.
That gap is not a perception problem. It is an operations problem working its way through teams, and it is costing real money.
Strategic Reality: The executive-worker AI productivity gap is not about adoption resistance. Workers are using the tools as instructed. The problem is that the tools are being deployed without the guidance, training, or process redesign needed to capture actual productivity gains.
The numbers leaders need to see
Stanford researcher Jeff Hancock and colleagues surveyed 1,150 US desk workers to understand what happens downstream when AI output gets passed between colleagues. Their findings, summarised below, should reshape how boards think about AI return on investment.
| Metric | Finding |
|---|---|
| Workers encountering workslop monthly | 40% |
| Average time spent correcting workslop | 3.4 hours per worker per month |
| Estimated annual productivity loss | £6.4m for a 10,000-person organisation |
| Executives reporting AI productivity gains | 92% |
| Non-managers reporting AI time savings | 60% (40% report none) |
| Firms reporting no return on AI investment | 95% (MIT report) |
The MIT figure deserves attention. After several years and billions in enterprise AI spend, the overwhelming majority of firms have no measurable return to show. SAP and Deloitte assessments are marginally more optimistic but still describe the ROI-positive cohort as a minority, with most organisations expected to wait two to four years for returns to materialise — a long runway for a technology investment.
What workslop actually looks like
The mechanism is simple and, once you see it, obvious. A worker is told to use AI. They generate a draft in seconds. It looks coherent. It reads professionally. They pass it on. The next person in the workflow — often a more senior colleague, a client-facing manager, or a specialist — discovers that the output contains factual errors, misinterpretations of the original brief, or subtle logical inconsistencies that only become apparent on careful reading.
That second person now has two bad options. They can send it back, which creates friction, or they can fix it themselves, which absorbs the “saved” time plus interest. Either way, the apparent gain at the point of generation becomes a real loss further down the chain.
Hidden Cost: Workslop shifts effort from drafters to reviewers. The productivity metric most organisations track — “time to produce a first draft” — looks great. The metric they do not track — “total cycle time including rework and review” — has often got worse.
The Guardian piece quotes a freelance product designer describing colleagues pasting chatbot output directly into emails and, when challenged on unclear content, responding with “I’m not sure what AI meant by that.” Judgment itself is being outsourced. That is the point at which productivity tooling becomes a liability, because the organisation loses the quality-control layer that made professional work professional.
A University of Michigan researcher embedded in primary care clinics found the same pattern in a setting with much higher stakes. Clinicians encouraged to use AI to draft patient email replies ended up doing more editing, not less, and worried about sending AI-generated errors to patients. Because the tool was optional, many simply stopped using it once the novelty wore off — a rational response that is unavailable to workers whose AI use is mandated.
Why the mandate model is the root cause
The temptation, on seeing these numbers, is to blame the technology. That is the wrong conclusion. The tools can produce genuine productivity gains in the right hands, for the right tasks, with the right guardrails. What is breaking down is the management model around the tools.
Leaders who have personally experimented with AI — usually on low-stakes tasks where they are the only reviewer — extrapolate their own experience to everyone else’s work. They mandate AI use, often alongside layoffs that cite AI-driven efficiency, and then pressure remaining staff to produce more. Workers use the tools as instructed, because refusing is not career-safe. The output looks productive. Nobody has measured the downstream cost.
Competitive Reality: The firms currently experiencing workslop have given themselves a disguised cost disadvantage. Competitors who resisted the pressure to mandate AI, or who invested in proper deployment, now have cleaner internal workflows whilst paying similar licence fees.
The stakeholder impact nobody models
Different roles in an organisation experience AI mandates very differently. The lived reality varies in ways that do not show up in any dashboard most executives see.
| Stakeholder | Experience under mandated AI | What gets measured | What gets missed |
|---|---|---|---|
| Senior executives | Personal use on summaries, emails | Personal time savings | Team-level cycle time |
| Mid-level managers | Absorbing rework from team output | Team delivery pressure | Review time increases |
| Specialist staff | Editing AI output back to accuracy | Output volume | Judgment decay over time |
| Junior staff | Producing first drafts via AI | Output per hour at drafting | Skill development slowing |
| Customers and patients | Receiving AI-assisted communications | Response time | Content quality and trust |
The people reporting 92% productivity gains are looking at their own desks. The people reporting zero gains are looking at the whole pipeline.
A better deployment model
The alternative to the mandate is not to ban AI. It is to deploy it the way any other serious business tool gets deployed: with clear use cases, measurable outcomes, and honest assessment of where it adds value and where it subtracts it.
Implementation Note: The highest-performing AI deployments share three characteristics — narrow task definition, integrated quality review, and opt-in usage for tasks where the tool demonstrably helps. The worst-performing share one characteristic: blanket mandates without use-case definition.
Priority actions by organisational maturity
The right first move depends on where a company sits on the adoption curve. Below is a rough framework, sequenced so that earlier stages build the foundations later stages need.
Early stage (exploring AI deployment):
- Resist the pressure to mandate tools before use cases are defined.
- Pilot narrow, measurable applications (draft generation for specific document types, meeting summaries, first-pass research).
- Track total cycle time, not drafting time.
- Train reviewers specifically on evaluating AI output for factual drift.
Mid stage (tools deployed, unclear ROI):
- Audit current AI usage to identify where workslop is accumulating.
- Interview reviewers and specialists — not just AI users — to surface rework burden.
- Retire mandates for tasks where ROI is not demonstrable; keep AI as an option.
- Invest in prompt discipline and output evaluation skills, not just tool access.
Mature stage (AI embedded, seeking optimisation):
- Redesign workflows around AI strengths rather than bolting AI onto existing processes.
- Negotiate clearer AI mandates with unionised staff where relevant — increasingly a live issue, per the Communications Workers of America.
- Build quality-control gates into AI-generated content pipelines.
- Measure judgment and skill retention over time, not just short-term output.
Success Factor: Organisations that outperform peers on AI ROI typically restrict AI to tasks where the human remains the author and the AI is an accelerator. Organisations that underperform typically ask AI to be the author whilst humans become editors — which is where workslop breeds.
Hidden challenges leaders underestimate
Beyond the visible rework cost, four second-order problems deserve attention. None of them show up in the first year. All of them compound.
Judgment decay. When junior staff use AI to generate work they would previously have drafted themselves, they skip the reasoning step that built their professional judgment over time. A five-year cohort of professionals who have never personally wrestled with a complex draft is a very different workforce from the one that preceded them. The cost is invisible until the senior generation retires.
Accountability fragmentation. When a client receives flawed output, the question “who wrote this?” no longer has a clean answer. The person who ran the prompt, the person who reviewed it, and the model itself are all partial authors. Organisations have not yet adapted their accountability frameworks, and regulators are beginning to notice — particularly in regulated sectors such as healthcare, financial services, and law.
Morale erosion. The Guardian piece quotes one copywriter describing morale collapse after an AI mandate paired with layoffs. This is not incidental. Workers who are told the tool will save them time, and then find it consumes more time, lose trust in leadership judgment. That loss is hard to reverse and shows up in retention figures before it shows up anywhere else.
Security drift. AI-generated content often includes plausible-sounding but fabricated detail. In contexts where accuracy matters — patient communications, legal correspondence, financial analysis — this creates a new class of error that traditional quality processes were not designed to catch. Data protection implications compound when workers paste sensitive context into third-party tools to get better output.
Warning: ⚠️ Each of these four challenges strengthens the case for deliberate, use-case-driven deployment rather than blanket mandates. An organisation that waits until year three to address them will find remediation much more expensive than prevention.
The strategic takeaway
AI can deliver real productivity gains. The evidence that most current deployments are not doing so is now substantial enough that treating it as a temporary adjustment phase is no longer credible. The gap between the 92% of executives reporting gains and the 40% of workers reporting none is a gap between two different measurement frames — and the workers are the ones looking at the whole system.
Three factors separate the organisations capturing real AI value from those producing workslop:
- Use-case specificity: AI is deployed against defined tasks, not rolled out as a general-purpose mandate.
- Quality infrastructure: Review, evaluation, and feedback loops are designed in from the start, not retrofitted after errors emerge.
- Honest measurement: Total cycle time, not drafting time. Quality of output, not volume. Morale and retention, not just productivity dashboards.
Next steps for your organisation
- Audit your current AI deployment against the three success factors above.
- Interview your specialist and reviewer staff — not just AI users — about rework burden.
- Quantify the gap between your executive productivity perception and your worker productivity reality.
- Identify one mandated AI use case to retire and one narrow use case to strengthen.
- Build a quarterly review of AI ROI that tracks cycle time, quality, and morale alongside output volume.
The companies that will emerge with genuine AI advantage over the next two to four years are not the ones with the most aggressive mandates. They are the ones that measured carefully, deployed narrowly, and refused to let the appearance of productivity substitute for the reality of it.
For further analysis of AI adoption patterns and their strategic implications, see our insights archive. To discuss your organisation’s AI deployment approach, get in touch.
Analysis based on reporting by The Guardian, 14 April 2026 (original article), drawing on survey research by Jeff Hancock and colleagues at Stanford and BetterUp, the MIT report on enterprise AI returns, and assessments from SAP and Deloitte.
Resultsense is a UK publication making sense of AI for professionals and businesses. This article represents independent editorial analysis.