The Numbers Behind the Quality Problem: What the Latest GRBN and GDQ Data Actually Show
Today I had the privilege of joining Andrew Cannon GRBN (Global Research Business Network), Chuck Miller Insights Association Global Data Quality Initiative, Debrah Harding MRS, and Anita Watkins for a webinar presenting the latest wave of two critically important industry data quality studies. This is a summary of what we covered, plus my answers to several questions we didn’t have time to get to. The full Q&A is long — worth it if you care about quality :)
What the data actually said
The GRBN Online Sample Buyers Sentiment Survey (Wave 2, May 2026) and the IA Global Data Quality Benchmarking Report (Wave 2, H2 2025) were released in the same week, and presenting them together almost seemed intentional. One measures how buyers feel about quality. The other measures what’s actually happening in the data. Together, they paint a picture the industry needs to understand.
On the buyer side: B2B satisfaction has continued to erode. The GRBN data shows the dissatisfied share of B2B buyers rising from 40% to 43%, while satisfied buyers dropped from 42% to 36%. The B2B quality satisfaction sub-index fell from 85 to 78 — well below the 100 baseline. Meanwhile, B2C satisfaction held relatively stable at 55% satisfied.
What’s driving it? In part, removal rates. Four in ten B2B sample buyers are now removing 30% or more of their records — up from 32% in Wave 1. That’s not a QC process. That’s a structural problem.
The IA GDQ benchmarks put numbers around what that looks like at the supply side. Across nearly 2 million records from 46 companies in 13 countries: research agencies are running pre- and in-survey removal rates of ~28.5% combined; suppliers are at ~21.2%. For General B2B specifically, the post-survey clean-out rate is 59.6%. That’s not a typo. Nearly 6 in 10 B2B completes are being removed after the survey closes. The highest-risk segment in the study, by a significant margin.
One more number worth noting: roughly 60% of surveys are now completed on mobile. The industry’s survey design hasn’t caught up. We are creating disengaged respondents with poor UX and then flagging them for removal. A self-inflicted wound.
Questions we didn’t get to — my answers
Q: Why are buyers increasingly questioning the value proposition of B2B sample?
It’s not purely price sensitivity — it’s a value-perception gap. Buyers are paying the higher rates, but also absorbing operational costs that aren’t priced in: cleaning cycles, extended field times, open quotas, etc. When you’re routinely removing 30–40% of completes post-field, the cost-per-complete on your invoice is a fiction. The real number is often 40–60% higher once you account for what happens after delivery. Buyers are doing that math. And a UK industry estimate that was referenced on the call (Thanks Debrah!) put the cost of data quality remediation — refilling, incentives, technology — at approximately £200M per year. That’s not a rounding error.
Q: When buyers remove 30%, 40%, even 50% of records — what are the actual consequences?
The consequences cascade in ways that aren’t always visible upstream. Quota completion gets messy fast. Timelines slip. Client confidence erodes. But the hidden cost nobody talks about enough is analyst time. Someone has to run those QC checks, make judgment calls on gray area respondents, and (ideally) document the cleaning process. That falls on the research team that bought the sample — not the vendor who supplied it.
And the part that’s easy to miss, which I’m very passionate about: removing records doesn’t necessarily fix the problem. What’s left may be structurally biased toward whoever was easiest to reach. A cleaned dataset isn’t automatically a representative one. Chuck said it well during the webinar and I wanted to hug him – sometimes we clean too much.
Q: Buying intentions remain strong despite the quality complaints. Isn’t that a contradiction?
Online sample is the infrastructure for this industry. There isn’t a credible alternative, especially one that can scale — yet. So buyers complain, remove 35% of their completes, just accept it, and come back next quarter. Strong buying intentions combined with declining satisfaction is a warning sign, not a reassurance. It’s the pattern you see in markets before a structural break. Buyers are staying in, but they’re also quietly building alternatives — hybrid designs, building their own verified panels, moving towards communities and smaller-base solutions, and synthetic supplements. When a credible alternative achieves scale, the buying-intention numbers will move fast.
Q: How much of today’s quality challenge is actually a transparency challenge?
More than most admit. There are suppliers doing serious, highly sophisticated quality work who are losing business to cheaper alternatives because buyers can’t evaluate what they’re buying. That’s a transparency failure with real market consequences — it degrades the incentive to invest in quality infrastructure in the first place.
Some of the “transparency” being offered right now is theater. Publishing a fraud rate without disclosing your methodology, or citing a “multi-layer QC process” without specifying what those layers actually catch, isn’t transparency — it’s marketing. What buyers need is standardized, provable metrics. That’s exactly what the GDQ benchmarking initiative, led by the Insights Association’s CDI, is trying to build. This is important work and the industry should be participating. We need more data!
Q: Prevention vs. detection — which matters more?
Prevention wins on ROI, but detection is your safety net, and you need both. The detection-only model — let respondents in and clean on the back end — is expensive, lagged, and structurally incomplete. You can never be fully certain what you missed. Pre-survey blocking at the device and identity level is dramatically more efficient: you’re not paying for fraudulent completes, you’re not running cleaning cycles, and you’re not guessing at what slipped through.
One important caveat I’d add: the industry is trending toward layering more and more supplemental fraud and digital fingerprinting platforms, and we haven’t studied well how they complement each other. From my perspective, stacking these tools without understanding their interaction effects creates real risk — Type 1 and Type 2 errors compound, you may be blocking legitimate respondents at scale, and the data can look cleaner while actually introducing noise that puts research-based decisions at risk. More tools is not automatically better. And erroneous decision making may be the outcome.
Q: What metrics should buyers start requesting from suppliers?
Pre-survey block rates should be table stakes. If a supplier can’t tell you what percentage of attempted entries were blocked before reaching your screener, they either don’t have that infrastructure or they don’t want you to see the number. Either answer is informative. Also, I’d love to see what happens to these respondents. Of the respondents that were removed in the past year, what happened to them? Were they kicked out of the panel? Go through survey training? Just keep taking surveys? No one knows.
Beyond that: incidence variance across sources in a blended design, because convergence is a quality signal. Post-survey cleanout rates by flag type — not just total removal percentage. And for B2B specifically, professional verification rates. Not “we target IT decision-makers” but “here is our verification methodology and here is our match rate.” If they can’t answer that question, you don’t actually know who you’re talking to.
Q: What concerns you most looking ahead 2–3 years?
AI-assisted fraud. The fraud we’re detecting today was built to fool systems from two years ago for the most part. What’s being built now is optimized against current detection methods. I’m less confident the gap stays manageable.
Synthetic respondents get a lot of attention — and I’ve written about the legitimate applications of synthetic data, which are real. But identity spoofing at scale is the more immediate threat. The ability to construct a verified-looking respondent profile — real device, real location, plausible behavioral signature — is advancing faster than the industry’s collective ability to detect it.
Q: One metric or trend you most hope to see improve in the next wave?
Pre-survey verification adoption rates. Everything downstream gets easier when you know who you’re talking to before the survey starts. If the next GDQ wave shows meaningful movement in that number — suppliers investing in identity infrastructure, buyers requiring it as a procurement standard — that tells me the market has actually internalized the problem rather than just talking about it at webinars.
But here’s the flip side I want to name directly: the trend right now is to keep layering in new supplemental fraud and digital fingerprinting platforms, and I think that’s heading somewhere risky. We haven’t studied how these tools complement each other — or whether they do. From where I sit, stacking detection layers without understanding their interaction creates compounding Type 1 and Type 2 errors. You block enough real people that the data starts to look cleaner, but you may actually be adding so much noise that the decisions coming out of the research are at risk. Cleaner-looking isn’t the same as more accurate. That distinction matters enormously when a client is making a business decision off your data.
The goal isn’t maximum detection. The goal is knowing who you’re actually talking to — from the front end, with verified identity. That’s where the investment should go.
Q: What gives you the most optimism looking ahead 2–3 years?
Two things, and I’ll keep it quick.
First, the amount of capital invested in this industry is staggering. Brands, agencies, and decision-makers around the world are paying for the insights we produce because those insights drive real business decisions. That level of investment is a signal that what we do matters — and it creates a powerful incentive to get quality right.
Second, the collaboration. The GDQ initiative alone spans dozens of companies, multiple global associations, and practitioners volunteering their time and expertise from across the world. And honestly, some of my closest friends in this industry are my biggest competitors. We work together anyway. That’s unusual. Most industries don’t operate that way, and it’s something worth protecting.
Debrah put it well on the call: the sheer number of practitioners willing to give their time to building standards and tools for GDQ is evidence of a community that’s genuinely committed — not just talking about quality, but building the infrastructure to measure and improve it. Chuck and Andrew echoed that: the associations are rowing in the same direction, the benchmarks are getting sharper, and the attention this problem is finally getting is real.
The problems are real. So is the will to fix them.
