Raising the bar: why data quality is built in, not just cleaned up later

Data quality has become one of the defining challenges facing the insights sector. From fraudulent respondents to the growing influence of AI-generated answers, ensuring robust, reliable data is increasingly complex, and increasingly critical.

In this piece, James Endersby, chair of the Market Research Society and CEO at Opinium, argues that the industry must move beyond treating data quality as a post-field fix. Instead, he calls for a holistic, end-to-end approach that embeds quality at every stage of the research process, and rebuilds trust in the insights we deliver.

Why data quality must be built in, not cleaned up later.

Trust has always been the foundation of market research. Our value as an industry depends on the confidence that clients, stakeholders and the public place in the data we provide. But that trust is under increasing pressure.

For too long, market research has treated data quality as a downstream issue, something to fix once fieldwork is complete. But as fraud, automation and disengagement increase, that mindset is no longer sustainable.

Data quality is not a checkpoint. It is a system.

As chair of the Market Research Society, I see an industry grappling with rising complexity: AI-generated responses, professional respondents and increasingly sophisticated fraud networks. These are not isolated challenges, they are structural. And they demand a structural response.

The answer lies in thinking in terms of a data quality stack, a holistic approach that protects integrity from start to finish.

It begins even before a survey is written, with who we allow into our research ecosystem. Strong panel recruitment, identity checks and responsible sourcing are critical first lines of defence. If we fail here, we introduce risk before a single question is asked.

From there, questionnaire design plays a central role. Length, clarity and engagement are not just experience considerations; they are quality controls. Poorly designed surveys actively create bad data by encouraging speeding, straight-lining and disengagement. If we make participation a chore, we should not be surprised when respondents behave accordingly.

Equally important is testing and piloting. Soft launches, timing checks and early-stage validation help identify issues before they scale, preventing avoidable quality problems in full field.

At Opinium, we have embedded this end-to-end mindset into our processes. Data quality is not owned by one team or one stage; it is a shared responsibility across the entire research lifecycle.

But even with strong foundations, the threat landscape has evolved. Bots, click farms and incentivised fraud now operate at a scale and sophistication that traditional checks alone cannot contain.

This is where the industry must go further.

For us, that has meant combining strong design and sampling with continuous, in-survey quality controls and real-time monitoring. By integrating cutting edge and advanced AI driven solutions, we are able to identify and remove fraudulent / fake and low-quality participants as they occur, in real time during the survey (in-field), protecting data while it is being collected, not just after the fact.

This layered approach is critical. Technology enhances quality, but it does not replace good research practice. The strongest outcomes come from combining thoughtful design, robust sourcing and intelligent, real-time intervention.

And even then, the process does not stop at fieldwork. Post-field validation, transparency and, where possible, triangulation with other data sources remain essential to ensuring confidence in the final insight.

Because ultimately, this is about more than methodology, it is about reputation.

Every dataset we deliver either strengthens or weakens trust in our industry. In an environment where bad data is easier than ever to generate, the consequences of getting it wrong are significant, not just for individual projects, but for the credibility of insights as a whole.

The broader lesson for the sector is clear. There is no single fix for data quality. It must be embedded as a continuous discipline, from onboarding and design through to delivery and validation.

Because if we want to protect the future of market research, we must first protect the trust it is built on.