1. Home
  2. Data quality
Verification

Data quality and fraud prevention for B2B sample

We check the claim, not just the click. Every respondent is screened on device, role and category knowledge before entering your survey.

B2B fraud is a genuine person claiming a role they do not hold. Device and bot checks cannot catch it. This page explains what does.

The problem

Two different kinds of fraud

Consumer grade quality tools were built for the column on the left. Most of what breaks a B2B study is in the column on the right, and it passes an identity check cleanly.

In practice

What the review stages do to a batch

Illustrated for 1,000 screener starts at 18% incidence, the IT decision maker figure. The share each layer removes varies by audience and is reported on every project.

159 verified completes delivered, 12% of qualified completes removed before deliveryIllustrative for 1,000 starts at 18% incidence. The share removed at each stage varies by audience and is reported per project. This is a single batch before replacement: the removed completes are refielded at no cost, so the delivered count still meets your quota.
The layers

Three layers, then the record

  1. Layer 1

    Participation

    Device, network, location and duplicate signals are reviewed the moment a respondent enters, across every supply source on the project.

    What it assesses
    Whether the same person or device has entered before, whether the connection is masked, and whether the respondent is in the target market
    What it cannot establish
    Who the person is professionally. A clean device tells you nothing about the job title, and a VPN flag on its own is not proof of fraud
  2. Layer 2

    Role

    Every respondent states their job title, seniority and company size, with company size asked as a number, and describes the role in their own words before entering your survey. Our own panel members have also verified a work email at the employer they name, when they joined and again every 90 days.

    What it assesses
    Whether title, seniority, company size and the described responsibilities hold together, and for panel members, whether they currently receive email at the employer they name
    What it cannot establish
    Buying authority. A work email shows a current link to an employer, not what the person decides. Partner sources do not share email addresses, so partner-supplied respondents are judged on their answers
  3. Layer 3

    Category knowledge

    One open ended question with an answer a practitioner tends to give specifically: which ERP, which identity provider, how the close runs. Answers are checked against the stated role and company before the respondent is sent to your survey. When we host the survey, a person also reviews the screening answers and removes anyone whose answers do not hold up.

    What it assesses
    Whether the respondent describes the category the way someone who works in it does, and whether the answer is consistent with the role and employer they reported
    What it cannot establish
    Certainty. A plausible answer can be generated, and a terse one can come from a real practitioner. Borderline cases are reviewed against agreed criteria rather than removed by default
  4. Delivery

    Documentation

    Every removal we make is logged with the reason and the stage. When we host the survey, that covers every removal. When you host it, completes you reject are replaced at no cost and logged with the reason you give us.

    What it assesses
    Which completes were removed, why, at which stage, and how each was replaced, so base sizes can be defended in a debrief
    What it cannot establish
    It is a record, not a check. It exists so the three review stages above can be audited, by you or by your client
Rejection criteria

What gets a complete rejected

  • Duplicate device, account or person across sources
  • Location outside the target market. A VPN or proxy flag on its own leads to review, not removal
  • Work email domain that does not match the stated employer (panel members)
  • Company size inconsistent with the named employer
  • Category knowledge answer that is wrong, or stays generic when reviewed against the stated role
  • Speeding below 40% of median length
  • Straightlining on grids combined with any other flag
  • Open ends that are gibberish, copied or off topic
  • Internal inconsistency between stated role and described responsibilities

When we host the survey, a person reviews the screening answers and the survey data, and removes anyone whose answers do not hold up, including speeding, straightlining and weak open ends. When you host it, we cannot see that data, so those checks are yours, and completes you reject are replaced at no cost.

Rejected completes are replaced at no cost and never invoiced. The share removed depends on the audience and the source mix, and it is reported per project with the denominator stated: removals as a share of qualified completes, before delivery. A high removal rate is not a quality score; the goal is a defensible accepted sample, not the largest possible exclusion list.

Our guide to preventing survey fraud and AI bots in B2B research explains the reasoning behind each layer, including the false positives we design against.

Afterwards

After delivery

If you find a delivered complete that fails our own criteria within 30 days, we replace it and update the log. We would rather hear about it than not.

Respondent privacy

Verification uses data respondents have consented to share, such as the work email panel members verify. We do not deliver personally identifiable information with survey data unless the study is designed for it and the respondent has explicitly agreed. Handling is consistent with GDPR and CCPA. Our respondent privacy notice explains what is collected and why.

Industry standards

Answers to the ESOMAR 37 questions for online sample are available on request, along with a description of the supply sources used on your project at category level.

What you receive

The rejection log

Every removal we make is recorded with the layer that caught it and what we did about it. When we host the survey, the log covers every removal. When you host it, completes you reject are replaced and logged with the reason you give us.

rejection-log.csv extract
Table comparing Respondent, Layer, Reason and Action
RespondentLayerReasonAction
VN-4417RoleWork email domain does not match stated employerReplaced
VN-4422KnowledgeNamed a consumer product, not an enterprise platformReplaced
VN-4431IdentityDuplicate device across two supply sourcesReplaced
VN-4448RoleRole description does not match stated titleReplaced
VN-4460IdentityCompleted in 4 min against a 14 min medianReplaced

Illustrative extract from a survey we hosted. Every project comes with a log of the removals we made, so the base sizes can be defended.

Questions

B2B data quality: frequently asked questions

Why is B2B verification different from consumer data quality?

Consumer quality tools are built for identity fraud: bots, duplicates, wrong country. B2B fraud is mostly a real, unique person claiming a role they do not hold, which identity checks are not designed to detect. Catching it means checking the role claim and testing category knowledge, which is what layers two and three do.

Does verification slow field down?

Slightly. Verification runs continuously during field rather than at the end, so the effect on the window is usually a day. The field window quoted at feasibility already includes it.

Do I pay for rejected completes?

No. Only verified completes that land in a quota cell are invoiced. Replacement of rejected completes is included.

Can I see the verification data for each respondent?

You receive the rejection log, which shows what was removed and why, and a description of the method for delivered completes. Individual respondents' identity and employment data are not delivered, because they are personal information.

What is your rejection rate?

It depends on the audience and the source mix, so we report it on every project with the denominator stated: removals as a share of qualified completes, before delivery. Senior audiences see more removals than broad ones. A figure near zero on a decision maker audience is worth a question from any vendor, and a high figure is not a quality score on its own.

Send the spec. Get real numbers back.

Audience, market, target n, expected interview length. Feasibility the same business day for standard audiences, with reachable counts, an estimated qualification rate we believe and a quoted cost per complete that stays fixed for the agreed brief.

Request a project quote
Request a quote