Worker Rights Training at Scale: Measuring Knowledge Gain and Behavioural Intent

Apple upholds strict labor and human rights standards across its global supply chain. Its Supplier Employee Development Fund (SEDF) — a $50 million commitment to world-class worker rights training, worker voice platforms, and education programmes — engages strategic partners to support millions of workers worldwide. This case study reports findings from an impact assessment of one SEDF-supported programme.

In Fall 2025, an impact assessment was conducted on a digital labour-rights training programme delivered through SEDF — across 33 facilities to more than 13,000 workers. The assessment was designed to measure three outcomes: knowledge change before and after training, behavioural intent in realistic workplace scenarios, and contextual factors that influence whether workers act on training content.

The central finding: training-driven knowledge gain is associated with workers’ selection of the appropriate response in realistic workplace scenarios — at scale, in operationally integrated form, across diverse facilities.

Findings:

• Workers’ overall accuracy on knowledge questions rose from 74% to 83% (an increase of 9 percentage points).
• The largest gains occurred on topics where pre-training accuracy was lowest, with some topics showing improvements above 20 percentage points.
• Across multiple workplace scenarios, more than 80% of workers selected the appropriate response to non-compliant requests.
• Workers who demonstrated post-training knowledge of a topic were significantly more likely to select appropriate scenario responses
• Peer-support, perceived consequences, and empowerment indicators each independently associated with workers’ likelihood of declining a non-compliant request.

The assessment was administered within the same session as the training itself, with no additional time burden on the worker or the facility. Statistical analysis controlled for age, gender, education level, and employment experience. Limitations are noted in the methodology section.

The training programme is a digital labour-rights training delivered through the Apple Supplier Employee Development Fund (SEDF) directly to supplier employees on their own mobile devices. Through its global community of leading strategic partners, SEDF has provided opportunities for millions of people around the world to learn, grow, and explore pathways to advance in their jobs. This programme was developed to “test and scale a new digital rights-training game delivered directly to supplier employees’ mobile devices,” co-designed with supplier employees, integrating their feedback through testing, focus groups, and one-on-one interviews. By the end of 2025, the programme had reached more than 1.6M supplier employees.

The programme has been operating for several years and has previously been measured primarily through completion, accuracy and recall — measures useful for assessing workforce-training programmes operating at this scale.

The 2025 assessment was undertaken to extend the existing measurement framework with additional outcome layers — change in knowledge attributable to the training, and likelihood that learned content would translate into action. The objective was operational: produce evidence the programme team could use to direct future content development, identify topics where additional reinforcement is needed, and surface contextual factors at the facility level that may shape whether workers act on what they learn. Implementation was integrated with the existing training delivery system.

Population and scope

Instruments

• Knowledge battery — multiple-choice questions addressing labour-rights content covered in the training. Pre-training and post-training items were matched on each topic, so each worker’s post-training answer could be compared directly with their own pre-training answer rather than only across group averages.
• Scenario battery — vignette-based items presenting realistic workplace situations involving non-compliant requests. Workers selected from response options.
• Contextual-factors battery — items measuring perceived peer support for refusing non-compliant requests, perceived consequences of refusing, and three empowerment indicators (confidence in supervisor communication, trust in grievance channels, perceived access to accurate information).

Analysis

Knowledge change was estimated using regression analysis controlling for age, gender, education level, and years of employment experience. Significance testing was conducted at the p < 0.01 level. Within-subject change was reported as the share of workers who improved on each topic. Scenario response was modelled as a function of post-training knowledge and contextual-factors variables, controlling for the same demographic covariates.

The measurement design used here, embedded in the Goodbit platform — matched pre/post knowledge items, scenario-based behavioural-intent items, and contextual-factors items administered in a single session — is sector-agnostic. The substantive content of the training is the variable; the measurement architecture is constant. Programmes that train workers on safety, system processes, compliance, customer engagement, or any other content where the operative question is whether knowledge gained translates to action taken can be measured with the same architecture. The methodology has been applied across public-health, civic-engagement, financial-literacy, and workforce-development contexts, with consistent operational characteristics: integration with existing training delivery, no additional time burden on participants, and aggregate reporting suitable for programme management.

The assessment measures knowledge and behavioural intent within a single session; it does not measure observed behaviour in subsequent workplace situations. Self-report on contextual factors (peer support, perceived consequences) is subject to known reporting biases, including the possibility that workers’ assessments of their own willingness to refuse non-compliant requests overstate the rate of such refusal in practice. The cross-sectional design permits associative — not causal — inference for the contextual-factors findings. The 33-facility sample is not a probability sample of the broader supplier base and findings should not be extrapolated beyond the participating facilities without further work.

Knowledge change

Workers’ overall accuracy on the matched knowledge battery rose from 74% pre-training to 83% post-training — an increase of 9 percentage points. The increase was statistically significant after controlling for demographic covariates.

The pattern of change varied by topic:

• Topics where pre-training accuracy was below the overall mean showed the largest gains. Several topics showed improvements above 20 percentage points.
• Topics where pre-training accuracy was already high showed modest or negligible change.
• Among workers who answered a given topic’s pre-training question incorrectly, more than half answered the matched post-training question correctly. This pattern held across all topics measured.

Behavioural intent in scenarios

In the scenario battery, more than 80% of workers selected the response associated with declining a non-compliant request across the scenarios presented. Selection of the appropriate response was significantly associated with post-training knowledge of the relevant topic. Workers whose response moved from incorrect (pre-training) to correct (post-training) on the relevant knowledge item were more likely to select an appropriate scenario response than workers whose pre-training response was already correct or who did not show learning.

These findings are consistent with — but do not establish — a causal pathway from training content to workplace decision-making. They establish only that knowledge gained in the training is associated with the type of decision the training is intended to support.

Three contextual factors

Three contextual factors were associated with the likelihood that workers selected an appropriate scenario response:

• Peer support. Workers who reported stronger peer agreement that refusing non-compliant requests is appropriate were more likely to select the appropriate scenario response.
• Perceived consequences. Workers who reported greater perceived likelihood of negative consequences for refusing non-compliant requests were less likely to select the appropriate scenario response, even when post-training knowledge was strong.
• Higher reported confidence in supervisor communication, higher trust in grievance channels, and higher perceived access to accurate information were each associated with greater likelihood of selecting the appropriate scenario response.

The assessment supports the following inferences:

• Training delivery in this format produces measurable knowledge gain at scale, attributable to the training after controlling for demographic covariates.
• The pattern of gain — concentrated where baseline knowledge is lowest — is consistent with effective targeting of content to need.
• Worker-reported behavioural intent in scenarios is associated with knowledge gained from training.
• Contextual factors at the workplace level such as peer norms, perceived consequences, and perceived empowerment are associated with behavioural intent independently of knowledge.

The assessment does not support the following inferences:

• Direct causal claims about whether training-driven knowledge produces workplace behaviour. Behavioural intent in a scenario battery is a useful but imperfect proxy for action.
• Generalisation beyond the 33 facilities and worker population in the assessment without further work.
• Specific causal claims about how facility-level conditions shape worker behaviour. The cross-sectional design supports association, not causation.

The assessment was designed to fit within existing training operations rather than to operate as a separate research project. Pre-training items, post-training items, scenarios, and contextual-factors items were administered as part of the same digital training session on the Goodbit platform, with no additional time burden on the worker or scheduling burden on the facility.

This design pattern — assessment integrated with training delivery rather than separated from it — is the methodological contribution of the assessment. It is intended to be replicable across other training programmes that operate at scale and have an interest in extending measurement beyond completion and content metrics.

The programme is a digital labour-rights training delivered to supplier employees as part of the Apple Supplier Employee Development Fund (SEDF), a $50 million commitment Apple launched in 2022 to strengthen and scale worker rights training, worker voice platforms, and educational opportunities across Apple’s global supply chain.

The 2025 impact assessment described in this case study was conducted in partnership with Goodbit. Goodbit is a platform that combines short, anonymous, game-based learning with AI-guided conversation and rigorous before-and-after measurement — engaging people directly while measuring how their understanding and views shift. The methodology has been peer-reviewed in adjacent contexts including a Public Health Agency of Canada randomised controlled trial.

To learn more about the Supplier Employee Development Fund, refer to Apple’s People and Environment in Our Supply Chain Annual Update. For inquiries about the methodology or about applying the same approach to other training programmes, contact Goodbit at

hello@madebygoodbit.com