Enterprise technology leaders are facing an uncomfortable truth about their AI investments: the very workers they equip with AI tools are spending hours each week monitoring and correcting those tools, a phenomenon the workplace AI company Glean calls 'botsitting'. According to Glean's report, based on a survey of UK digital workers, the average worker saves around 12 hours per week through automation, but roughly half of that – 6.3 hours – is consumed by supervising AI outputs.
The Hidden Tax on AI Productivity
The report found that nine in 10 UK digital workers now use AI at work, putting British workers ahead of their US counterparts (84%). Yet only 42% describe their workplace as AI-first, suggesting significant friction remains. On a personal level, four in five (78%) admit AI makes them more productive, but only one in five (18%) believe this has any significant effect on overall organisational performance.
Glean's data indicates that employees typically spend more time botsitting (38%) than actively prompting AI (36%). "The hidden labour is becoming a quality-control problem," Glean declared. More than a third of AI sessions fail altogether, and 77% of UK workers have had to correct or re-do AI-generated work in the past month; 26% did so in the past week.
Rethinking AI ROI Metrics
Dr Rebecca Hinds, Head of the Work AI Institute, warned that companies are treating AI adoption as a vanity metric. "Too many companies are treating AI adoption like a vanity metric: more seats, more prompts, more usage," she said, implying that reported productivity gains are largely overstated.
For CTOs and technology procurement leaders, the implication is clear: traditional metrics – total time saved, number of prompts, number of seats – do not capture the hidden cost of oversight. The report argues that companies must reframe AI's actual impacts by measuring error correction, prompt refinement, and output validation to understand where AI truly delivers results.
What This Means for Enterprise Buyers
While the Glean research focuses on general office work, the 'botsitting' problem is acutely relevant for supply chain and logistics technology managers. In areas like trade documentation digitalisation or customs technology, where accuracy is critical, the cost of AI failures could be even higher. If a third of AI sessions fail in routine office tasks, the failure rate in complex compliance or regulatory contexts may be steeper, demanding even more human oversight.
| Metric | Value |
|---|---|
| Workers using AI at work (UK) | 90% |
| Workers using AI at work (US) | 84% |
| Workers describing workplace as AI-first | 42% |
| Workers who say AI makes them more productive | 78% |
| Workers who believe AI significantly improves organisational performance | 18% |
| Weekly hours saved through automation | 12 |
| Weekly hours spent botsitting | 6.3 |
| Time spent botsitting vs prompting | 38% vs 36% |
| AI sessions that fail | >33% |
| Workers correcting AI work in past month | 77% |
| Workers correcting AI work in past week | 26% |
The Path Forward
Glean calls this growing overhead the "cleanup bill". The report suggests that companies need better visibility into the full lifecycle of AI-assisted work – from prompt to output validation – to prevent productivity leakage. For enterprise software buyers evaluating AI platforms, this means demanding transparency in error rates, validation workflows, and mechanisms to reduce the burden on human supervisors.
The research serves as a counterpoint to the narrative that AI eliminates work. Instead, it shifts the nature of work toward supervision. As Dr Hinds noted, productivity gains are overstated if they don't account for the time spent policing the machine. The challenge for IT teams under pressure to deliver ROI is to build metrics that capture both the time saved and the time spent supervising – and to drive tools that reduce the latter.
Sources:
