The 80% Problem: What Factory Audit Data Reveals About Supply Chain Labor and the Case for Automation
New audit data shows most factories can't meet working hour standards, and it's raising uncomfortable questions about whether robots are part of the solution or just a convenient excuse.
画像クレジット: Image via Supply Chain Dive. Used under fair use for news commentary. · source
Picture a factory floor at 9 PM on a Thursday. The shift was supposed to end two hours ago, but the order deadline is tomorrow. Workers are tired, the supervisor is stressed, and somewhere in a corporate office thousands of miles away, a compliance officer is updating a spreadsheet that will eventually flag this as a violation. This scene, it turns out, is playing out in roughly 80% of factories audited under one of the world's largest social compliance programs.
New data from amfori BSCI (Business Social Compliance Initiative) reveals that in 2025, four out of five audited factories failed to meet standards on working hours. The number is striking, but what interests me more is what it suggests about the structural impossibility of current manufacturing models, and why this data point keeps appearing in conversations about industrial automation.
To be precise, the amfori BSCI framework audits factories on a range of social compliance metrics, with working hours being one of the most commonly violated categories. The 80% figure refers to factories that fell short on working hour requirements, which typically means excessive overtime, insufficient rest periods, or both.
It's worth noting that amfori BSCI is one of the largest compliance initiatives globally, covering supply chains for major retailers and brands. This isn't a small sample of problem factories; this is a systematic finding across a broad audit program. The implications for reputational and financial exposure are significant, as the original reporting notes, but I think the robotics angle here is underexplored.
関連記事
More in Industrial
Everyone bought robots. Now nobody's robots talk to each other. Welcome to the multi-vendor mess that's quietly strangling warehouse automation.
Mark Kowalski · 12 hours ago · 7 min
Google DeepMind is funding research into what happens when millions of AI agents interact unsupervised. The industrial automation world should be paying close attention.
Robert "Bob" Macintosh · 20 hours ago · 4 min
Rising inflation erodes more than paychecks. For warehouse operators already squeezing margins, it changes the math on automation investment in ways that aren't obvious at first glance.
Here's where I need to be careful, because there are two very different ways to interpret this data, and the robotics industry has a tendency to cherry-pick the convenient one.
The first interpretation: excessive working hours are a symptom of labor shortages and production demands that humans physically cannot meet sustainably. If factories are consistently requiring overtime that violates international standards, perhaps the work itself needs to be restructured. Automation, in this framing, becomes a way to reduce the burden on human workers by handling repetitive, physically demanding, or time-sensitive tasks.
The second interpretation: automation becomes a convenient justification for displacing workers while claiming humanitarian motives. "We're protecting workers from excessive overtime" sounds better than "we're reducing labor costs," even when the latter is the primary driver.
I don't know which interpretation is more accurate in aggregate. Probably both are true in different contexts. But the 80% figure does suggest something structurally broken about current production models that goes beyond individual factory management failures.
This is where I get a bit pedantic, but I think it matters. The relationship between industrial automation and labor conditions is not as straightforward as either automation advocates or critics suggest.
A 2023 paper from researchers at MIT (Autor et al., if you want to look it up) found that automation's effects on labor are highly context-dependent. In some cases, automation of dangerous or repetitive tasks improved working conditions for remaining workers. In other cases, it simply eliminated jobs without meaningful improvements for anyone. The paper emphasized that policy frameworks and implementation approaches mattered more than the technology itself.
There's also work from the ILO (International Labour Organization) suggesting that in supply chains with strong compliance oversight, automation can reduce pressure on working hours by increasing productivity without requiring overtime. But, and this is crucial, this assumes the productivity gains are used to reduce hours rather than increase output targets. That's a management and policy question, not a technology question.
I haven't found rigorous studies specifically examining whether factories that adopt more automation have better working hour compliance. This seems like an obvious research gap. If anyone has data on this, I'd genuinely like to see it.
The 80% figure itself isn't entirely new. Working hour violations have been a persistent issue in global supply chains for decades. What's arguably new is the scale of the data and the willingness of compliance organizations to publicize unflattering numbers.
From a robotics perspective, what's changed in the last few years is the feasibility of automation for tasks that previously required human flexibility. Collaborative robots (cobots) and AI-driven quality control systems are now deployable in factories that couldn't afford or couldn't accommodate traditional industrial automation. This is incremental over earlier cobot deployments, but the cost curves have shifted enough that it's becoming relevant for the types of factories that appear in these audit reports.
The genuine novelty, if there is one, is in the convergence of compliance pressure, reputational risk, and automation feasibility. Brands are increasingly accountable for supply chain conditions (see: various ESG reporting requirements and consumer activism). Factories are struggling to meet standards with current labor models. And automation technology has reached a point where it's a plausible response to some of these pressures.
Whether it's a good response is, well, that's the question.
I'll be honest: this data raises more questions than it answers, and I'm skeptical of anyone who claims to have definitive conclusions.
First, we don't know what happens to workers when factories automate in response to compliance pressure. Do they get reassigned to less physically demanding roles? Do they lose jobs entirely? Do working conditions improve for those who remain? The research here is thin, and most of it comes from developed economies that may not generalize to the supply chain contexts where these audits occur.
Second, there's a selection effect problem. Factories that adopt automation may be systematically different from those that don't in ways that affect both their likelihood of automation and their compliance outcomes. Correlation studies won't tell us much without careful controls.
Third, and I know I'm being picky here, but the 80% figure needs context. What percentage of factories failed on working hours five years ago? Ten years ago? Is this getting better, worse, or staying the same? The trend matters more than the point estimate.
If I were advising a research team on this, here's what I'd want:
Longitudinal data on factories that have adopted automation, tracking compliance outcomes before and after implementation. Matched controls would be essential.
Qualitative research on how factory managers actually make decisions about automation versus overtime. The stated reasons and the actual reasons are often different.
Analysis of whether compliance pressure (reputational risk, audit findings) actually drives automation investment, or whether cost reduction remains the primary motivator with compliance as a post-hoc justification.
Worker-level data on outcomes. Not just employment numbers, but working conditions, wages, and job quality for those who remain employed.
None of this exists in a comprehensive form, as far as I can tell. We're making policy and investment decisions based on assumptions that haven't been rigorously tested.
I started this piece intending to write something more definitive, but the data doesn't support definitive conclusions. The 80% figure is real and concerning. Automation is increasingly feasible as a response. But whether automation actually improves labor conditions, or simply shifts the problem elsewhere, remains genuinely unclear.
What I can say is that the framing matters. If we treat automation as a humanitarian intervention ("protecting workers from excessive overtime"), we risk ignoring the displacement effects and the ways automation can be used to avoid addressing root causes like unrealistic production demands and pricing pressure from brands. If we treat automation as purely exploitative ("replacing workers with machines"), we risk ignoring cases where it genuinely improves conditions for remaining workers.
The reality is probably messier than either narrative. Factories are complex systems, supply chains are complex systems, and labor markets are complex systems. Anyone claiming simple answers is probably selling something.
(I realize I've written 1,400 words without offering a clear policy recommendation. That's intentional. I don't think the evidence supports confident prescriptions yet. What it supports is more research, more transparency in audit data, and more skepticism toward both automation evangelists and automation critics who claim certainty they don't have.)
The 80% figure should be a wake-up call. But a wake-up call to what, exactly, remains to be determined.
Strong earnings, lukewarm reaction. Bob Macintosh thinks the Oracle situation tells you something useful about where AI infrastructure hype meets hard numbers.
