Data quality benchmarks used to be a supplier talking point. Since Esomar Congress in Valencia (1 to 4 September 2026), they have become a buyer's question.
Agencies are now being asked by their own clients how they know the data is good. "We clean it at the end" no longer satisfies an insight team that wants to see what was checked, at which stage and by whom. A single clean-out rate does not answer that.
This article explains which data quality benchmarks are public, how to read them without overreaching, and which fieldwork partner questions belong in a brief. It finishes with seven items for a sample quality RFP checklist that you can adapt for your next study.
Two ideas run through everything below. First, a number only means something when you know who reported it, for which audience and at which stage of the process. Second, a public figure is a prompt for a better question, not a pass mark. Hold on to both ideas and data quality benchmarks become useful in a brief instead of risky in a proposal.
The sections follow the order in which a buyer usually needs them: what is public, how to read it, where removals happen, how to turn it into a brief, and a closing checklist.
Table of contents
What the latest data quality benchmarks now show
The most useful public reference is the Insights Association H1 2026 Global Data Quality Benchmarking Report, released on 16 June 2026 for the Global Data Quality (GDQ) initiative. It draws on nearly 1.8 million survey records from 13 countries. It is now more than three months old, so treat it as background rather than news, and read the report itself rather than a summary of it.
A note on naming. The Insights Association (IA) calls this release the third wave of the GDQ Benchmarking Project, while some secondary sources call it Wave 2. Use the IA's wording in anything you publish and check it against the report. Secondary sources also mention about 46 contributing companies and expanded country reporting for the US, UK, Canada, Australia, Germany and France.
Why does this matter to buyers? The report shows what the industry norm looks like in a given country and audience. It does not tell you what your project should achieve. Used that way, data quality benchmarks help you ask better questions rather than set pass marks.
The most practical use of the report is calibration. If a supplier tells you a B2B study will lose almost nothing to cleaning, you have a reason to ask how that is possible. If a supplier quotes a figure far from what the report shows for the same country and audience, you have a reason to ask what is being counted and at which stage. Neither situation proves anything. Both produce better conversations than a bare percentage.
How to read a public benchmark report without overreaching
Before you apply any data quality benchmarks to a brief, run through a short reading routine. It is quick, and it prevents most misquotes.
- Confirm the source and the date. Is the figure in the primary report, or in someone's summary of it? Anything older than 12 months is background, not a current reference.
- Confirm the population. Check the country, the audience (B2B or B2C) and whether the figure describes suppliers, agencies or buyers. The Wave 1 report found that agencies globally reported lower fraud removal than suppliers, so their figures are not interchangeable.
- Confirm the definition. Which stage does the removal figure cover: pre-survey, in-survey or post-survey? The UK Wave 2 source notes that stage definitions were clarified, which is exactly why figures from different waves may not line up.
- Check the sample behind the number. The same source cautions that the UK B2B sample is small. A small base turns a precise-looking percentage into a rough indication.
- Decide what the number is for. Public data quality benchmarks are a prompt for a conversation with your fieldwork partner. They are not a pass mark.
A figure that fails one of these checks can still serve as context. It should not appear in a proposal as a target.
Say a slide shows a removal rate with no country, no audience and no stage. It fails three of the five checks at once. That does not make it wrong. It makes it unusable as a target, and you should ask whoever produced it for the missing detail before it goes anywhere near a client document.
What buyers say they do with their own data
The GRBN Online Sample Buyers Sentiment Survey Wave 2 (May 2026) adds the buyer's view. As reported by Angelfish Fieldwork, 43% of B2B sample buyers remove 30% or more of cases in data cleaning, compared with 25% of B2C sample buyers.
We have taken that figure from the Angelfish summary, not from the GRBN document. Check it against the GRBN PDF, which is listed in the GDQ resources for buyers, before it goes into a client deck. Buyer-side data quality benchmarks like this one are only as safe as their source.
The gap between B2B and B2C buyers is worth a moment's thought. If B2B buyers remove a larger share of cases in cleaning, B2B plans need more room for replacements, and the terms for those replacements should be agreed before fieldwork starts rather than after the cleaning pass.
In practice, that means a B2B brief should say how replacements will be sourced, how long they will take and who carries the cost. It should also say what happens to quotas that were already full when a record was removed. These points are dull to write and expensive to leave out.
Claims to verify before you repeat them
You may see it said that B2B is the hardest environment for data quality, and that the gap between sold and actual incidence is narrowing globally. Both claims may be in the full IA report, but we have not confirmed them there, so we do not rely on them here.
If you want to use either claim, find the page in the primary report and cite it. The same applies to any healthcare patient or B2B incidence detail: use only what the full report states.
Secondhand data quality benchmarks travel quickly through conference slides, sales decks and social posts, and the caveats rarely travel with them. A simple habit helps: every number in a client document gets a source line with the report name, the page and the date you checked it.
Older background: Wave 1
Wave 1 (H1 2025) is more than 12 months old, so it is context and not a target for 2026 quotes. According to the Wave 1 report hosted by MRS, general B2B had the highest pre- and in-survey removal at 15.3% and much higher post-survey cleaning at 18.9%.
The same report found that agencies globally reported lower fraud removal than suppliers. It attributes this to suppliers cleaning points of entry upstream. That finding leads straight to the question of where removals happen.
Use Wave 1 as a reminder of the pattern, not as a number for 2026. The pattern is that figures differ by audience and by who is reporting, and that data quality benchmarks only make sense when you know both.
One more caution on comparing waves. Because stage definitions were clarified in the later wave, a change between two reports may reflect a change in how things were counted rather than a change in the fieldwork itself. If you want to describe a trend, say which waves you compared and confirm that the definitions match.
Related reading
- Global Economic Outlook 2025: A Data-Driven Perspective
- Big Data and Predictive Analytics in Market Research
- Privacy-First Research Methodologies: The Ultimate Guide to Ethical Data Collection in 2025
Where removals happen matters as much as the headline rate
A headline clean-out rate is the number most buyers ask for first. On its own it is the weakest of the data quality benchmarks, because it hides who did the cleaning and when.
The three stages in plain terms
- Pre-survey removals happen at the door. Duplicates, bots, mismatched locations and ineligible people are stopped before they see a question.
- In-survey removals happen during the interview. Speeding, failed attention checks, contradictory answers or poor open-text responses end the interview.
- Post-survey removals happen after delivery, in the data file. Records are flagged, removed and, where terms allow, replaced.
Suppose two projects both report 15% total removal. In one, almost everything was stopped at the door, so the timeline was barely touched. In the other, the same share was removed from the data file after delivery, which means replacements, delays and difficult conversations. Asking for removal rates by stage is how you tell them apart.
The table below puts the three stages side by side, so you can ask for data quality benchmarks stage by stage instead of accepting one blended number.
| Stage | Typical examples | Effect on your timeline | Question for your partner |
|---|---|---|---|
| Pre-survey | Duplicates, bots, mismatched locations, ineligible people | Lowest disruption, because fieldwork has not been touched | Which checks run at entry, and when? |
| In-survey | Speeding, failed attention checks, contradictory answers, poor open-text | Moderate, because quotas keep filling while interviews are ended | Which rules end an interview, and who set the thresholds? |
| Post-survey | Records flagged, removed and, where terms allow, replaced | Highest disruption: replacements, delays, reopened cells | What is the replacement window, and who pays? |
Read the table from top to bottom as a cost curve. The earlier a bad record is stopped, the cheaper it is for everyone. A partner that can show strong pre-survey and in-survey controls is usually offering you a calmer delivery week, because fewer records are left to argue about at the end.
The reverse is not automatically a warning sign. A low post-survey rate is only good news if the earlier checks were doing their job. If a partner reports almost no removals at any stage, the right response is curiosity about what was checked, not congratulation.
One reported UK data point
Angelfish Fieldwork's summary of UK Wave 2 (published June 2026) gives stage-level data quality benchmarks for the UK. Suppliers removed 18.6% of records pre- and in-survey, against 11.2% for agencies. We have not carried the post-survey cleaning figures from that summary into this article, because they need to be checked against the primary report first. Look them up there before you quote them.
The direction of the gap fits the Wave 1 pattern, in which suppliers do more of their cleaning at the points of entry. It is a reminder that a supplier figure and an agency figure answer different questions. Two cautions apply. The UK Wave 2 source says stage definitions were clarified, and it describes the UK B2B sample as small.
Treat this as an illustration of how stage-level data quality benchmarks can be presented, not as a UK target for your next quote. If you plan to use the figures, read the primary report, confirm the population and the definitions, and add a source line.
Turning public benchmarks into a brief you can run
Public data quality benchmarks only earn their place in a brief when they change what you ask for. The steps below are a practical way to do that. They work for online data collection, and most of them apply equally to CATI and CAPI projects.
- Fix the audience and the country first. A B2B study of a narrow job function and a consumer study of general adults behave differently, so decide which public reference, if any, is comparable before you borrow a number from it.
- Ask for removals by stage. Request pre-survey, in-survey and post-survey rates separately, so that your partner's data quality benchmarks can be compared with the public ones on the same basis.
- Write the rules down. List the checks that apply at each stage, the thresholds that trigger a removal and who is allowed to change them. Speeding rules and attention checks should not be tuned halfway through fieldwork without a record of why.
- Agree replacement terms before launch. State the replacement window, the source of replacement sample and who pays. This is the point where most disputes start, and it is much easier to settle on paper than under deadline.
- Ask for interim reporting. A removal summary at the end tells you what went wrong. An interim summary lets you act while quotas are still open.
- Keep an audit trail. Ask for record-level flags with a reason for each removal, so that you can answer a client who asks why a particular respondent was excluded.
None of these steps needs a public benchmark to be perfect. They need your partner to be specific, and the public figures give you a reasonable starting point for what to expect.
A worked example: planning a B2B study
Consider an illustrative case, with no real client behind it. An agency is planning a B2B survey in two countries, with quotas by job function and company size. The agency's client has asked for a written account of how the data will be checked.
Without stage-level data quality benchmarks to hand, the agency can only ask its fieldwork partner for a total removal rate. With them, the questions become sharper. The agency asks which checks run before the survey, such as duplicate detection and location checks, which rules end an interview in progress, and which flags are applied only after delivery. It asks for each stage to be reported separately, by country and by audience. It also asks for a replacement window to be written into the agreement, with a clear statement of who pays.
Midway through fieldwork, the partner sends an interim report. The agency can see whether removals are happening at the door or in the data file, and can act while quotas are still open. If most removals are landing post-survey, the conversation happens early rather than on the day of delivery.
At the end, the agency has a one-page quality note for its client: what was checked, at which stage, by whom, and what was replaced. That note answers the question the client actually asked.
When your partner's figures differ from the public ones
When a partner's numbers sit well away from public data quality benchmarks, do not assume that either side is wrong. Work through the likely reasons first:
- Different definitions. The partner may count a removal at a different stage from the report, or may count flagged records and removed records as the same thing.
- Different audience mix. A study of a narrow professional group is not the same as a general B2B sample, even if both are labelled B2B.
- Different reporter. Suppliers and agencies report different figures, as the Wave 1 report showed.
- A small base. A single project is a small sample compared with a report built on nearly 1.8 million records, so one project can sit well away from the norm without anything being wrong.
- Different thresholds. A stricter speeding rule or an extra attention check will remove more records, and may be the right call for your study.
Ask for the explanation in writing. A partner who understands its own process can usually give one in a few lines. A partner who cannot is telling you something useful too.
Questions to put to your fieldwork partner
The questions below are the ones that separate a partner with a documented process from one with a general reassurance. The table pairs each topic with a weak answer and a useful one, so you can recognize the difference in a call or a proposal. Use it alongside your own data quality benchmarks when you score responses.
| Topic | Weak answer | Useful answer |
|---|---|---|
| Removal rate | "We clean about the usual amount." | Rates reported by stage, by country and by audience |
| Definitions | One blended figure with no explanation | A written definition of each stage and the rule behind each removal |
| Thresholds | "Standard settings." | Named thresholds, who set them and when they can change |
| Replacements | "We will see how it goes." | Replacement window, source of replacements and cost stated up front |
| Evidence | A summary at the end of fieldwork | Interim reports and record-level flags with reason codes |
| Public references | A figure quoted without a source | Report name, page and date checked |
You do not need a perfect score on every row. You need to know where your partner is strong and where you will need to add your own checks. The point of asking for data quality benchmarks in this form is that the answers can be compared across suppliers, which a blended clean-out rate never allows.
Common pitfalls
Most problems with quality claims come from a short list of habits. They are easy to avoid once you know to look for them.
- Quoting a public figure as a target. A report shows what the industry norm looks like. It does not tell you what your study should achieve, and a contract clause built on it will cause arguments.
- Mixing waves and definitions. Putting a Wave 1 figure beside a later figure without checking the definitions can create a trend that is not there.
- Comparing suppliers with agencies. The two groups report different things. Compare like with like.
- Reading a small base as a fact. The UK B2B sample is described as small. A precise-looking percentage from a small base is a rough indication.
- Treating a low removal rate as proof of quality. It may simply mean that the checks were light. Ask what ran before you celebrate.
- Leaving replacement terms until the cleaning pass. By then the schedule is already under pressure and neither side is in the mood to negotiate.
- Repeating secondary figures without a source line. Data quality benchmarks lose their caveats each time they are copied, so keep the report name, page and date with every number.
- Changing the rules mid-field. If a threshold is tightened after launch, early and late completes are no longer treated alike. Record any change and the reason for it.
Frequently asked questions
What are acceptable data quality benchmarks for online surveys?
There is no single acceptable figure. Benchmarks are industry norms that vary by country, audience and study type, so compare your quote with a matching market and audience in the GDQ report. Ask for removals by stage and treat any global average as context, not a target or a guarantee.
What is a normal removal rate for survey data?
It depends on the audience and on where cleaning happens. In the GRBN buyer survey from May 2026, 43% of B2B sample buyers said they remove 30% or more of cases, against 25% of B2C buyers. Ask for removals split by stage, because one headline rate hides who did the cleaning.
What questions should I ask a sample provider before fielding a survey?
Ask where sample comes from, how panelists are recruited, whether sources are blended or sub-contracted, and how removals are reported by stage. Also ask for sold versus achieved incidence and replacement terms. Esomar's 37 Questions and the MRS Data Quality Transparency Checklist give you ready-made wording to paste into a brief.
What should a market research RFP include about data quality?
Ask for stage-by-stage removal reporting, sold and achieved incidence, source disclosure, replacement and re-contact terms, and who pays for replacements. Add a right to review sub-contracting and blending. Reference the GDQ Buyers of Research Services Procurement Guide so vendors know the evidence you will score in review.
What is the Global Data Quality (GDQ) initiative?
The Global Data Quality initiative is an industry effort that publishes benchmarking reports through the Insights Association, plus buyer tools such as a procurement guide, a transparency checklist and a Feedback Loop for classifying sample issues. The H1 2026 report covers nearly 1.8 million survey records from 13 countries.
A sample quality RFP checklist: seven items
Use the list below as a starting point and adapt it to the audience, the country and the method. It is deliberately short, so that it gets read.
- Audience, country and sample source. State who the respondents are, where they live and where the sample will come from. This is the base for choosing any public reference.
- Removal reporting by stage. Require pre-survey, in-survey and post-survey removal rates to be reported separately, so that the figures can be set beside public data quality benchmarks on a like-for-like basis.
- Definitions and thresholds. Ask for a written list of checks, the rule behind each one and who can change it. Request that any change during fieldwork is recorded with a reason, so that your data quality benchmarks stay consistent across the study.
- Replacement terms. Set the replacement window, the source of replacements and who pays, before fieldwork starts.
- Reporting cadence. Agree when interim removal summaries will arrive, so that problems surface while quotas are still open.
- Record-level audit trail. Ask for flags with a reason for each removal, so you can explain any exclusion to your client.
- Source lines for public references. Require the report name, page and date checked for any public figure used in the proposal, and ask for the primary report rather than a summary. Public data quality benchmarks should never appear without them.
If a prospective partner can answer all seven items clearly, you have the beginnings of a quality process you can describe to your own client. If it cannot, you know where to probe before the project starts rather than after it ends.
The wider point is simple. Data quality benchmarks are most useful when they sharpen your questions, and least useful when they are used as a shortcut to a pass mark. Ask for stage-level detail, agree the rules in writing and keep a source line for every number. That is what lets an agency answer a client who asks how the data is known to be good.
