The Data Quality Benchmarking report from the Insights Association and the Global Data Quality initiative is the closest thing online research has to shared numbers on its own plumbing. Wave 1 pooled data collected between January and June 2025 from 51 companies, covering 78 countries and roughly two million records.
The headline most people repeated was removal rates. The numbers we would pin to the wall of anyone who writes quotes are elsewhere in the report. This is our reading of them. The report itself is short and worth reading in full; the figures below are global unless stated.
Removal: suppliers catch more, earlier
| Research agencies | Suppliers | |
|---|---|---|
| Removed pre-survey | 2.8% | 7.4% |
| Removed in-survey for fraud | 2.4% | 1.3% |
| Removed in-survey for behaviour | 4.2% | 5.0% |
| Total removed before or during the survey | 9.4% | 13.7% |
| Post-survey cleanout of qualified completes | 6.6% | 7.4% |
Two things matter here. Suppliers remove more, and more of it before the survey, which is what you would expect from the party closest to the traffic. And even after all of that, roughly one qualified complete in every fourteen was still cleaned out afterwards.
For feasibility: if you need 500 clean completes, you do not need 500 completes. At a 7% post-survey cleanout, plan for roughly 540, and say so in the quote rather than discovering it in week two.
Incidence is sold high
| Sold incidence | Actual incidence | |
|---|---|---|
| Research agencies | 61.6% | 55.2% |
| Suppliers | 50.8% | 41.1% |
By study type the gaps are larger in places. General B2B was sold at 35.4% and came in at 31.3%. Healthcare provider work was sold at 65.7% and came in at 42.5%.
Lower actual incidence means more entries per complete, more screen-outs for suppliers to reward, and longer field time. The report says it plainly: the industry appears to consistently under-estimate how hard its own studies are to fill. We read that as optimistic quoting rather than bad luck.
For feasibility: quote incidence from the tightest qualifying condition, soft launch, and re-quote when actual incidence comes in materially lower. The incidence, LOI and CPI guide covers how to build the estimate.
B2B costs more than its CPI suggests
| General B2C | General B2B | |
|---|---|---|
| Removed before or during the survey | 13.1% | 15.3% |
| Post-survey cleanout of qualified completes | 5.6% | 18.9% |
| Median LOI of qualified completes | 12 minutes | 17 minutes |
Nearly one in five B2B completes was removed after the survey. That is the highest post-survey rate of any study type in the report, and it happens after the buyer has already paid for the traffic, the programming time and the field days.
For feasibility: on B2B, budget an overage that reflects post-survey cleaning, block suspicious sessions at entry rather than just flagging them, and put knowledge questions in the screener. Our pre-survey screening defaults follow that logic.
Link security still has a gap
91.5% of research agency records used server-to-server links, link encryption, formula-based links or secure mobile entry. For suppliers the figure was 75.2%. The report's own summary says link encryption has significantly improved but still needs strengthening, particularly for suppliers.
A quarter of supplier traffic arriving on links that can be edited is not a small number. Signing redirects is cheap by comparison.
Caveats worth keeping in mind
- The data is contributed by participating companies and weighted by self-reported company size. Companies with worse numbers may have less reason to take part.
- Some cuts are small. The B2B post-survey figure rests on around 23,000 qualified completes, and the healthcare provider removal figures on about 21,000 records.
- Each company classifies removals in its own systems before mapping them to the report's definitions, so some variation is likely.
- It is one wave. Trends need later waves.
None of that makes the numbers wrong. It makes them a floor for how careful to be, not a ceiling.
If you change one thing
Put post-survey cleanout into feasibility. Every quote that assumes 100% of completes survive cleaning is a quote that will be renegotiated.