The evidence behind the shift

Drawn from the ILLÆSA literature review, version 5 · September 2026

A case against audits is not by itself a case for anything. This page sets out both halves: why an audit cannot serve as proof, and what the evidence says works instead — including where it is thin and where it cuts the other way.

1. An audit is not proof

• One audit firm compared its own worker surveys with its own audits. In Bangladesh, 30 percent of surveyed workers had witnessed or experienced sexual harassment; the firm’s audits in the same period recorded inhumane treatment, a broader category, in 0.18 percent of cases (Human Rights Watch, 2022).
• Across 16,795 audits of 5,819 factories in 66 countries, auditors reported fewer violations when the supplier paid for the audit, when they had audited the factory before and when teams were less trained (Short, Toffel & Hugill, 2016).
• Twenty-eight social audits in two years failed to resolve passport retention at a single Malaysian supplier (Transparentem, 2021).
• When US Customs banned palm oil imports from Sime Darby Plantation in December 2020, the Roundtable on Sustainable Palm Oil said its earlier review of audit findings had raised no red flags (The Edge, reporting the RSPO statement, 2020).
• On Indian tea estates and in Ghanaian cocoa communities, workers on ethically certified sites reported similar patterns of coercion and underpayment to workers elsewhere (LeBaron, 2021).

Audits are not useless. Pre-announced audits by well-trained auditors improve health and safety, though not concealed violations such as child labour (Short, Toffel & Hugill, 2020), and a systematic review of 33 studies finds that codes and the audits behind them often fail to improve conditions, while eight studies show gains in safety (Vandenbroucke, 2024). The problem is not inspection. It is inspection used as proof.

2. What the alternative has delivered

• Continuous engagement. Under the ILO and IFC Better Work programme, the probability of workers in Jordan being concerned about sexual harassment fell by 18 percentage points by the sixth year; supervisory training raised productivity by 22 percent; and across factories tracked in Vietnam the revenue-to-cost ratio rose by 25 percent after four years (ILO & IFC, 2016).
• Binding obligations. The Bangladesh Accord recorded a remediation rate above 93 percent across more than 1,600 factories covering more than two million workers, attributed to binding obligations and collective purchasing power rather than inspection alone (Reinecke & Donaghey, 2025).
• Capability at the supplier. At Nike’s suppliers, monitoring alone produced limited results; combined with help to schedule work and improve quality it was associated with considerably better conditions (Locke, Qin & Brause, 2007). Adopting lean manufacturing was associated with a 15 percentage point fall in non-compliance on wages and hours (Distelhorst, Hainmueller & Locke, 2017).
• Worker participation. In a randomised trial across 84 Bangladeshi garment factories, strengthening worker–management safety committees improved safety without reducing pay or productivity, in factories that were already well managed (Boudreau, 2024).

3. What it costs, and who carries it

• In a survey of 525 facilities in 32 countries, 46 percent had between three and five social audits in 2022, 68 percent paid the full audit fees themselves, and 76 percent had at least one brand insisting on its own proprietary audit (WRAP, 2023).
• The Social & Labor Convergence Program reports 10,700 converged assessments in 2025 covering 7.5 million workers, each shared on average 2.7 times; the scheme estimates the audit spending that sharing could redirect at 35 million dollars. These figures are the scheme’s own (SLCP, Impact Report 2025).

4. What traceability and continuous signal can carry

• The gap. In 2021 just under half of surveyed companies knew where their tier-one suppliers were, and 2 percent could say the same at tier three and beyond (McKinsey, 2021). By 2025, visibility beyond tier one had reached 42 percent of a small sample, driven by tariff compliance (McKinsey, 2025).
• Verification that tests a claim. Isotopic tests by US Customs found 13 of 86 sampled garments consistent with cotton from Xinjiang; the agency’s own staff call the method “not a silver bullet” (Reuters, 2023). In the Brazilian Amazon, satellite alerts reduced deforestation because they triggered enforcement (Assunção, Gandour & Rocha, 2023).
• Product passports. The EU Digital Product Passport will carry provenance, material and environmental data; it is infrastructure a due diligence system can use, not evidence of labour conditions (Regulation (EU) 2024/1781).
• The limits of the ledger. Of 271 supply chain blockchain projects reviewed in 2021, 23 percent were market-ready (Vadgama & Tasca, 2021); data on a blockchain “may simply be immutable garbage” (Powell et al., 2022).
• Worker voice technology. In a randomised trial in two Indian garment factories, an anonymous channel was used by 5 percent of workers, yet turnover and absenteeism fell (Adhvaryu et al., Economic Journal). Tools oriented to due diligence rarely identified modern slavery where workers did not trust them (Rende Taylor & Shih, 2019).

Data about a chain becomes evidence only when something happens because of it.

Where the evidence is thin

Most audit-failure evidence comes from apparel, agriculture, electronics and mining; logistics, construction and services are largely untested. No study yet costs a full supplier-run system for a small or medium-sized supplier, and the savings claimed for shared schemes are self-reported. No independent study compares the detection rate of anonymous worker surveys with that of audits. And some findings cut the other way: in one buyer’s chain, orders did not rise when labour standards improved (Amengual, Distelhorst & Tobin, 2020). We state these limits because an argument that hides them would not survive the people it is meant to persuade.

The full literature review, with notes and page references, is available on request. Start a conversation.