Claims Library Entry
How Much Accepted Work Are We Producing per Unit of Human Attention?
This article proposes measuring AI productivity by the ratio of accepted work to total human hours spent on all attempted jobs, including review and correction time. It offers six practical controls for improving this ratio without lowering quality standards, such as agreeing on acceptance criteria upfront and matching review depth to consequences. The author provides a step-by-step approach for organizations to pilot the metric on a single workflow.
Published September 22, 2026 by Kamil Banc
Lead claim
Measure AI value as accepted jobs per unit of human attention, not drafts produced.
Atomic Claims
What this article supports
Copy individual claims as needed.
Claim 1 · Kamil's interpretation
Measure Accepted Work per Hour
Measure AI value as accepted jobs divided by total human hours spent on all attempted jobs.
Claim 2 · Source summary
40% Received Workslop
A 2025 survey of 1,150 U.S. desk workers found 40% received workslop in the previous month.
Claim 3 · Source summary
Two Hours per Workslop Incident
Workslop recipients estimated about two hours to resolve each incident, based on self-reported experiences.
Claim 4 · Source summary
BCG Consultants GPT-4 Experiment
A randomized experiment with 758 BCG consultants found GPT-4 improved performance within its capabilities.
Claim 5 · Kamil's interpretation
Keep Rejected Work Counted
Keep time spent on rejected work in the denominator so dashboards do not hide failures.
Evidence
Context behind the claims
Quote
"How much accepted work are we producing per unit of human attention?"
Key statistics
40%
Share of 1,150 full-time U.S. desk workers in a September 2025 BetterUp Labs and Stanford Social Media Lab survey who reported receiving workslop in the previous month.
~2 hours
Time recipients estimated to resolve each workslop incident; self-reported, not measured company-wide productivity losses.
758 consultants
Number of BCG consultants in a randomized GPT-4 experiment first reported in 2023, which found better performance within the model's capabilities and worse correctness outside them.
0.8 vs 1.0 accepted proposals per hour
Illustrative comparison: eight accepted proposals in ten human hours versus nine accepted in nine hours, with scope and quality standard held constant.
Supporting context
The article proposes a practitioner-oriented productivity metric: accepted jobs divided by total human hours across all attempted jobs, including preparation, checking, correction, and the recipient's review time. The author grounds the approach in external evidence, citing the BetterUp/Stanford workslop survey and the BCG GPT-4 experiment to show why first-draft speed alone misleads. Six operational controls translate the metric into practice, including agreeing acceptance checks with recipients upfront, matching review intensity to consequence, and logging reasons work gets returned. The recommended rollout compares ten consecutive jobs before and after the new rules, tracking first-pass acceptance, returns, and post-acceptance errors. The author explicitly frames these as practical proposals consistent with NIST's voluntary AI risk guidance, not a proven productivity intervention.
How to Cite
Use the claim-level citation when you need a precise statement. Use the article or claims-collection citation when you want the wider argument and source context.
Individual Claim
Best when you need to cite one atomic claim directly inside a memo, deck, research note, or AI output.
"[claim text]" (Banc, Kamil, 2026, https://kbanc.com/claims-library/how-much-accepted-work-are-we-producing-per-unit-of-human-attention)Original Article
Use this when you want to cite the full newsletter article at AI Adopters Club rather than the structured claims page.
Banc, Kamil (2026, September 22, 2026). How Much Accepted Work Are We Producing per Unit of Human Attention?. AI Adopters Club. https://aiadopters.club/p/how-much-accepted-work-are-we-producingClaims Collection
Use this when you want to reference the full structured claims collection on this page.
Banc, Kamil (2026). How Much Accepted Work Are We Producing per Unit of Human Attention? [Structured Claims]. Retrieved from https://kbanc.com/claims-library/how-much-accepted-work-are-we-producing-per-unit-of-human-attentionAttribution Requirements
- Include the author name: Kamil Banc.
- Include the source: AI Adopters Club or the structured claims page.
- Link to the original article or the claims page you used.
- Indicate any edits or transformations if you changed the wording.
Related Reading
More from the library
Sports stadiums are pioneering large-scale AI implementation across complex operational environments. By solving critical challenges in crowd management, revenue optimization, and efficiency, they've created a replicable playbook for AI adoption across industries.
5 claims
AI rollouts often shift work rather than reduce it, turning employees into full-time overseers of machine output. This article explores how oversight fatigue drives burnout and attrition among top performers, and offers practical fixes like capping tool stacks and batching review time.
5 claims
Shade Tree Advisors, a $794 million multi-family office in New York, deployed EtonAI across three back-office document workflows and cut custodian statement processing time from over 8 minutes to roughly 2 minutes per document. The phased rollout achieved 97 percent accuracy with 3 percent auto-flagged for human review, saving hundreds of hours annually.
5 claims