Invest1 publisher2 min readPublished
AI workslop cleanup runs to 96,000 hours a year at a 10,000-person company
A BetterUp and Stanford Social Media Lab survey of 1,150 US desk workers prices AI-generated busywork at $186 a month per affected worker. The $9m annual total that follows is a gross cleanup cost, counted before any credit for the drafting it replaced.
The Investor · Invest desk
What happened
- BetterUp Labs and the Stanford Social Media Lab define workslop as AI-generated content that looks good but lacks substance, leaving colleagues to do the thinking and the clean-up.
- The figures come from an online survey of 1,150 full-time US desk workers run in September 2025 by BetterUp in partnership with the Stanford Social Media Lab.
- Forty per cent of the US desk workers surveyed said they had received workslop in the previous month.
- Each incident takes an average of two hours to resolve, according to the research.
- BetterUp prices the damage at $186 a month per employee and $9m a year for a 10,000-person company.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- exposure The two hours come out of the recipient's own project time, so the cleanup is absorbed by the receiving team's payroll while the budget that bought the AI tool records no charge.
- constraint A gross cleanup figure cannot be subtracted from an AI programme's return until someone measures what the sender saved, so the $9m does not resolve an ROI question on its own.
- contradiction The organisation that priced the problem also sells the remedy, promising that its platform turns AI investments into real performance gains. That is a reason to want the two-hour finding replicated by a party with nothing to sell.
- decision An executive quoting this number has to choose between buying governance and slowing deployment, and the research itself prescribes guardrails, modelling thoughtful AI use and cultivating pilot mindsets.
Spread $186 a month across all 10,000 employees and the annual bill is $22.3m [3], more than twice the figure BetterUp published [6]. Apply it only to the 40% who said they received workslop in the past month [3], which is 4,000 people [1], and twelve months at $186 comes to $8.93m [2]. So the $186 is a cost per affected worker, grossed up. Spread the $9m across the whole payroll and it is $900 a head a year [4].
The hours are easier to check than the dollars. Two hours an incident [4], once a month, for 4,000 workers, is 96,000 hours a year [5]; at 2,000 hours to a full-time year, 48 people [6].
BetterUp did not publish the wage rate or the number of incidents behind the $186. One two-hour incident a month implies a fully loaded $93 an hour [7]. At $50 an hour, $186 buys 3.7 hours, closer to two incidents [8].
Every dollar in the study sits on the receiving end. BetterUp and the Stanford Social Media Lab asked 1,150 full-time US desk workers what landed on them and how long it took to fix [2]. A return on an AI programme also needs the hours the sender saved producing the thing. The $9m is a gross cost. The researchers also report that organizations are misled by false productivity and experience stalled AI adoption [9], neither of which is in the $9m.
Who can push back matters too. One respondent quoted in the research said: "Since it was provided by my supervisor, I felt uncomfortable confronting her about its poor quality and requesting she redo it." [8]
The two hours is the durable finding here, in my view: it is the one quantity a respondent could plausibly observe. The dollar figures are that number multiplied by assumptions. The counter-argument is that precision barely matters at this size: discount the 40% by two thirds and a 10,000-person company is still losing about 32,000 hours a year [9]. What would settle it is a measurement of the sender's side. If the colleague generating the slides saved three hours and cost a recipient two, workslop is a transfer between cost centres and the company nets ahead. If they saved twenty minutes, the loss is real and BetterUp has underpriced it, because the survey counts only the incidents the recipient noticed [7].
What to watch
- BetterUp and Stanford are recruiting for a larger workslop study. If it publishes the wage rate and the incident frequency behind the $186, someone outside can check the $9m.
- An employer publishing cleanup hours from its own timesheets, rather than monthly recall, would test the two-hour average against something observed.
- Whether a repeat survey after a year of corporate AI guardrail policies still finds 40% incidence.