Dirty Data Is the New Bad Brief
Why data quality is the most important conversation in research right now
There’s a conversation happening across the research industry that isn’t getting enough airtime. Everyone is talking about AI, how it’s transforming how we analyse data, how it’s generating synthetic respondents, how it’s writing survey questions. What we’re not talking about loudly enough is what happens when the data feeding all of that AI is compromised from the start.
In my view, data quality is the defining issue in market research right now. And it’s only getting harder.
The bot problem is real – and it’s accelerating
In the past 12 months, cleaning survey data has become genuinely difficult. AI-trained bots are now sophisticated enough to pass standard quality checks, complete attention tests, and mimic plausible respondent behaviour. We’re catching them – but it takes time, robust fraud detection tools, and a level of methodological rigour that not every research provider is applying.
When we ran a multi-year research program for Serko, the New Zealand-founded corporate travel technology company, we were tracking AI adoption rates among U.S. corporate travel managers. Early data was showing unusually high adoption figures. This is precisely the environment where panel integrity becomes critical: industry data suggests up to 30% of online panel respondents may be inattentive, fraudulent, or bots – which is why we deployed Sentry, the leading data quality solution, across this program. Sentry combines real-time behavioural analysis, on-screen event recording, and AI-assisted scoring to identify and eliminate bad data before it enters your dataset – and it shows. Rather than report inflated numbers at face value, we redesigned the research alongside that quality assurance layer. We segmented AI types – generic, enterprise, and agentic – and built question frameworks specifically designed to surface real-world usage versus perceived usage. The result was a far more credible, actionable dataset. And critically, it led Serko to a completely different strategic decision: rather than leading with product, they led with education, launching a free certified training program for travel managers that has now reached over 500 professionals.
That pivot was only possible because the data was trustworthy.
The survey monkey problem hasn’t gone away
AI has introduced new threats to data quality, but it hasn’t replaced the old ones. I still see companies running DIY surveys with biased question framing, recruiting from friendly CRM lists, and presenting findings as if they represent the market. One focus group with pre-warmed participants is not market research. A SurveyMonkey link pushed to your existing customers will tell you what they think of you – not what the market thinks of you.
For our client Duracell Australia, the brief was to understand how consumers shop the battery category and where the brand could justify a premium positioning. This required genuine objectivity – consumers who hadn’t been primed, recruited across the full competitive set, with methodology designed to capture unprompted behaviour. We ran an online community first, capturing real purchase moments and real language, then validated those findings through a robust n=1,000 quantitative survey. The outcome was a clear, retailer-ready story that Duracell could take into commercial conversations with confidence. That kind of rigour doesn’t happen by accident.
AI is only as good as what you feed it
Here’s the uncomfortable truth: AI tools applied to low-quality data don’t produce better insights – they produce confident-sounding bad ones. Speed is not a substitute for quality. Automation is not a substitute for rigour.
The AI era actually raises the stakes for data quality, because the outputs are more persuasive, more polished, and therefore more dangerous when wrong. Decision-makers are going to look at AI-generated analysis and trust it. Which means the data underpinning that analysis needs to be beyond reproach.
My recommendation to any business commissioning research in this environment: ask your provider exactly how they are handling bot detection, panel quality, and respondent verification. If they can’t give you a clear answer, that’s your answer.
The value of getting it right
Companies often treat market research as an upfront cost to be minimised. I’d argue the opposite: the cost of acting on bad data dwarfs any investment in getting it right. If you’re about to spend six figures on marketing, a product launch, or a market entry, and your foundational data is compromised, you’re not saving money. You’re amplifying risk.
Data quality isn’t a technical footnote. It’s the foundation everything else is built on. In the age of AI, that’s never been more true.
Nichola Quail, Founder & CEO, Insights Exchange