Manufacturers are collecting extensive information about product quality without always capturing the process conditions needed to explain failures. Research suggests this gap between measurement and prevention is limiting the value of manufacturing analytics despite continued investment.
Manufacturers have become increasingly effective at recording the quality of what comes off a production line, yet a detailed account of rejected parts does not necessarily reveal what went wrong during manufacture. Inspection systems identify a dimensional fault and record the loss, while the equipment conditions that preceded it remain unrecorded or difficult to reconstruct. The factory knows the outcome with considerable precision but may still struggle to intervene before the next defect occurs.
The imbalance emerges in The State of Manufacturing Data Collection, an October 2026 report published by Minitab Insights, the research and thought leadership group of manufacturing analytics company Minitab. Based on survey responses from 582 manufacturing professionals across 66 countries and ten manufacturing sectors, supplemented by practitioner interviews, the study finds that almost nine in ten respondents collect defect or quality data. Yet only 6.8 percent describe their use of data as predictive or optimised, while nearly half rely primarily on basic reporting. Measuring quality is evidently much more widespread than anticipating its deterioration. Collected through email, webinar and event polling, however, the survey is not a representative census of global manufacturing, so its findings should not be treated as industry-wide estimates.
A record of failure is not an explanation
Quality measurement is often the most developed part of a manufacturer’s data collection operation because the need to determine whether a product meets specification is immediate and unavoidable. Minitab’s research shows a different pattern for information that might help explain those results: around three in five respondents collect machine uptime and downtime data, slightly fewer track throughput and cycle time, and only around a third collect operator performance information. Although these categories do not cover every process input, they expose an imbalance between recording outcomes and monitoring production conditions.
Predicting defects demands more than knowing how many occurred on a particular shift or machine. Engineers need to establish which changes in process conditions are associated with variation in the finished product, whether those relationships persist across production runs and whether the warning arrives early enough to act. Historical quality records become more useful when examined alongside relevant settings and measurements, rather than separately from production performance.
The research also exposes the extent to which collecting those measurements remains a manual exercise. Of respondents answering the data-capture question, almost a third reported mainly manual collection, roughly two-thirds partial automation and just 4 percent fully automated capture. At a packaging operation in Mexico, an operations manager describes manually recording temperatures, pressures and speeds twice a shift, while the quality laboratory draws data directly from inspection equipment. The contrast is revealing: a reliable record of the finished result can coexist with infrequent observations of the process that produced it.
Manual collection need not make information inaccurate, and older equipment does not automatically prevent useful analysis. It does, however, affect the continuity and timing of the evidence available to investigate a defect. If measurements are entered only at fixed intervals, a short-lived change in operating conditions may pass between them, leaving engineers with a quality failure but an incomplete account of the events preceding it. Adding more inspection capacity will identify more of the consequences without necessarily making the causes any clearer.
When information arrives after the decision
The value of quality information also depends on whether it reaches someone who can change the process while production continues. Around one in five respondents primarily use data reactively, after issues have occurred, while nearly half rely on periodic basic reporting. Just over a quarter use real-time monitoring, including dashboards and alerts, and only a small minority use data predictively. These are respondents’ descriptions, not measurements of how quickly factories resolve problems.
A dashboard can make an emerging problem more visible without explaining its origin. Engineers may be able to see a rise in rejection rates as it happens, but still must investigate machine settings, material conditions or earlier process steps before deciding what to change. Where that work relies on exporting files, reconciling records or asking colleagues to retrieve measurements, the information may arrive after further defective output has been produced. Earlier notification improves the opportunity to respond, but it is not equivalent to understanding what is likely to fail next.
Minitab reports that 62 percent of respondents fell into its categories of no or limited data integration, compared with just 6 percent reporting full integration. Quality measurements are rarely sufficient on their own to explain why performance changes. The survey found that roughly a quarter of respondents collecting quality data alone reported moderate or full integration, compared with nearly half of those collecting quality data and at least two additional types of information. That association does not prove that collecting extra signals improves integration or quality, but it exposes the limits of examining quality alone.
Building a usable account of the process
The study’s comparison of more and less integrated manufacturers offers a further indication of what distinguishes stronger data practices. Roughly six in ten fully integrated respondents identified process equipment as a primary data source, compared with about three in ten of all other sites. Conversely, fewer than one in five fully integrated sites relied on legacy equipment as a primary source, compared with nearly two in five elsewhere. These differences do not imply that replacing older machines guarantees better decisions; they suggest that direct access to process information is associated with a more usable data environment.
For predictive quality, the practical requirement is to bring the right process evidence into the same analysis as the quality result. A defect record needs enough context to connect it to the relevant product, operation and time period, so that changes in temperature, pressure, speed or other measured conditions can be examined against what happened to the finished part. Consistent definitions and dependable timing are as important as the quantity of data collected. A model trained on mismatched records may produce an apparently persuasive pattern without identifying a condition that engineers can reliably influence.
There is also a difference between detecting a statistical relationship and establishing a useful intervention. Control charts can show that a process has changed, while regression or other analytical methods may help identify which inputs deserve investigation, as the Minitab report recommends. Those findings must still be interpreted against engineering knowledge and tested in production. A prediction only improves quality when it supports a decision that people or control systems can make safely and soon enough to affect the outcome.
The survey indicates a pronounced relationship between integration and respondents’ reported ability to act earlier. Among those with no integration, around two in five described primarily reactive decisions and approximately one in twenty predictive use. At the fully integrated end, predictive use rose to 21 percent, real-time monitoring was widespread, and fewer than one in ten were primarily reactive. The fully integrated comparison group comprised only 34 respondents, and these are observational associations rather than proof that integration itself causes predictive capability. Nevertheless, the contrast is consistent with the operational need to connect process conditions with quality outcomes before a factory can anticipate problems rather than document them.
Invest in the missing evidence
The report’s investment findings suggest that manufacturers have not lost confidence in collecting more data, although their spending is generally cautious. Just over half of respondents said they had increased investment in data collection during the previous two to three years, while around three in ten held it steady and very few reduced it. More mature sites were particularly likely to report substantial increases: roughly two in five fully integrated respondents had significantly raised spending, compared with fewer than one in five of the others. These are investment patterns, not returns, and do not establish whether spending addresses the missing information.
The stronger case for investment begins with a recurring quality loss and works backwards to the process decisions that might prevent it. That could mean adding measurements to an older machine, linking inspection results to operating conditions or improving existing data. It need not begin with an exhaustive programme to connect every asset. A focused exercise can establish whether the necessary signals are available, whether engineers trust them and whether the resulting analysis changes what happens during production.
Accurate defect records remain essential for quality assurance, compliance and improvement. But a factory that can account for every rejected part while repeatedly investigating the same unresolved causes has achieved only part of what its data could deliver. Progress should increasingly be judged by whether information helps people recognise deteriorating conditions and make effective adjustments before the next batch of defects is produced, rather than by the volume of quality data available for the following day’s report.
