B2B survey sample glossary

A completed survey that has passed the agreed eligibility and quality criteria and counts toward your quota. The number that matters commercially, and a different number from raw completes.

See alsoCompleteQuality termination

An answer a real respondent produces with help from an AI tool. Not automatically fraudulent, and not detectable from fluency alone, which is why open ends are read for category specificity rather than for polish.

See alsoBotSynthetic respondent

Processing data so that a person can no longer be identified from it, by anyone, taking account of what could reasonably be combined with it. Removing names and email addresses is rarely enough on its own: a job title, employer and location together can still single someone out. Data that keeps a key linking back is pseudonymized, not anonymized, and remains personal data.

See alsoPseudonymizationPII

A question inserted to test whether a respondent is reading, for example an instruction to select a specific answer. It catches inattention and some bots. It does not catch a genuine, attentive person who is not the decision maker they claim to be.

See alsoClaim fraudStraightlining

The rate at which panelists stop responding or leave a panel. High attrition is what turns a healthy tracker into a declining one, because the people who remain are systematically the most survey willing.

See alsoPanel fatigueTracker

Business professionals recruited to take part in research, selected by job role, seniority, company size and industry, as distinct from a general consumer audience.

See alsoFirmographicsJob function

The number of respondents behind a given figure, whether that is the whole sample or one cut of it. The base that matters is the smallest group you will report separately, because that is where precision is weakest. Sampling error formulas assume random selection, so on opt-in samples read small bases as indicative.

See alsoMargin of errorSampling error

Sample drawn from more than one source for a single project. Blending reduces the risk of one panel's skew becoming the finding, and for low incidence B2B audiences it is usually the only way to reach the target at all.

See alsoPanelRiver sample

An automated script completing surveys for incentive. Device, network and behavioral checks catch many of them at entry, and detection is an arms race rather than a settled problem: automated and AI-assisted responses can now produce plausible open ends, which is why checks are layered rather than relied on singly.

See alsoClaim fraudDigital fingerprinting

An open ended question that a practitioner tends to answer specifically and a guesser does not, for example which identity provider or ERP is in use. Reviewed by a person, it is the check that catches respondents who pass every technical test but do not do the job.

See alsoClaim fraudOpen end

A respondent who is a real, unique person but does not hold the job title, seniority or company size they report. Nothing about the device, network or account is wrong, so identity checks pass it cleanly. This is the failure mode specific to B2B research and only verifying the claim itself catches it.

See alsoEmployment verificationSeniority inflation

A question with predefined answer options. Fast to answer and easy to analyze, and useful in verification through factual checks, internal consistency and screening logic. What it cannot do on its own is show whether someone can describe the category in their own words, which is why B2B verification usually pairs it with an open end.

See alsoOpen endCategory knowledge question

A finished survey interview from a respondent who qualified and answered every required question. Completes are the unit B2B sample is sold in, which is why price is quoted as cost per complete.

See alsoCost per complete (CPI)Screen out

The share of people who start a survey and finish it. The denominator has to be stated, because completion among qualified starters and completion among all starters are different numbers.

See alsoDrop offSurvey start rate

The range produced around an estimate at a stated confidence level. The confidence level is the setting; the interval is the result.

See alsoConfidence levelMargin of error

How often the interval produced by a sampling procedure would contain the true value if the study were repeated. Raising it widens the interval: at the same base size, 99% confidence produces a wider range than 95%, and 90% a narrower one. 95% is the research default.

See alsoConfidence intervalMargin of error

A method that infers how respondents value individual product attributes by asking them to choose between realistic bundles rather than rate features in isolation. It needs a longer interview and a larger base, both of which raise cost per complete.

See alsoMaxDiffLength of interview (LOI)

The price paid for one accepted completed interview. Usually abbreviated CPI, which strictly stands for cost per interview; cost per complete and cost per interview are used interchangeably in practice, and CPC appears occasionally for the same thing. Driven mainly by incidence rate, the seniority of the audience and the length of the interview.

See alsoIncidence rate (IR)Accepted complete

Error introduced when part of the target population cannot appear in the sample frame at all. It is decided before a single answer is collected, and no amount of weighting fixes it.

See alsoSample frameNonresponse bias

A table showing one variable broken out by another, for example intent to buy by company size. Crosstabs are where small bases get exposed: a cut of 40 respondents carries a range wide enough to erase most apparent differences, and on an opt-in sample that range is itself an approximation.

See alsoBase sizeSampling error

Reviewing delivered responses for invalid, inconsistent or unusable data, and deciding what to correct, flag or remove. Distinct from verification, which happens before a complete is accepted.

See alsoQuality terminationReconciliation

Whether a respondent approves, shares approval, recommends, influences or only uses a business purchase. The single most over claimed attribute in B2B research, and the one a screener has to pin down explicitly.

See alsoSeniority inflationScreener

Removing respondents who have already entered the same study, including through a different supply source. In blended B2B sample this is the check that matters most, because the same scarce professional sits on several panels.

See alsoDigital fingerprintingBlended sample

How much a sample design, usually weighting or clustering, inflates variance compared with a simple random sample. A design effect of 1.5 means the sample behaves as though it were a third smaller.

See alsoEffective sample sizeWeighting

Identifying a device from browser, hardware and network characteristics so duplicates and known bad actors can be recognized across sources. Useful and not infallible; determined fraudsters can alter a fingerprint.

See alsoDeduplicationBot

The structured outline a moderator works from in a qualitative session. It sets the session length, which drives the incentive and the recruitment cost.

See alsoIn-depth interview (IDI)Incentive

A recruitment standard where a new panelist confirms membership by clicking a link sent to their email, proving the address is real and theirs. It slows recruitment and is one of the cheapest defenses against fake accounts.

See alsoPanelBot

A qualified respondent who starts the survey and does not finish. Ten to twenty percent is normal for a twelve to fifteen minute B2B interview and rises sharply beyond twenty minutes.

See alsoLength of interview (LOI)Incidence rate (IR)

The base size a weighted sample actually behaves like once the design effect is applied. It is the number that should drive precision claims, not the raw count of completes.

See alsoDesign effectMargin of error

Checking a respondent's stated employer and job title against evidence outside the screener, usually a corporate email domain or a third party professional record. A domain establishes an association with an employer; it does not by itself establish job title, seniority or purchasing authority, so it is one layer among several rather than a complete answer.

See alsoClaim fraudCategory knowledge question

A published set of 37 questions buyers can ask an online sample provider about sourcing, quality and data handling. Answering them is a transparency exercise, not a membership or a certification.

See alsoSource disclosure

A legitimate respondent flagged as fraudulent or low quality. Over aggressive exclusion has a cost of its own: it removes real people, skews the sample and inflates the removal rate.

See alsoVPN maskingQuality termination

A supplier's estimate of what can be delivered for a given specification: how many respondents are reachable, what incidence to expect, how long the field will take and what it will cost. A feasibility read should arrive before you quote your own client.

See alsoReachable nIncidence rate (IR)

The number of days a study is open for responses. B2B studies run five to ten business days for a standard audience; executive audiences need longer because the population is small and responds on its own schedule.

See alsoSoft launch

Company characteristics used to define a business audience: industry, employee count, revenue, ownership and location. The B2B equivalent of demographics.

See alsoB2B sampleInterlocking quotas

A complete recorded by manipulating the return link rather than by answering the survey. Plain redirects with guessable return URLs are the easy case; signed or hashed redirects and server to server confirmation both raise the cost of faking one. Worth asking a supplier which of these a project uses.

See alsoServer to server (S2S)Redirect delivery

A block of questions sharing one answer scale. Efficient to write, tiring to answer, and the place low effort respondents reveal themselves by straightlining.

See alsoStraightlining

An audience that cannot be drawn from existing supply at the volume and speed required. Low incidence is one cause; so are a small population, restricted access, geography and heavy screening. There is no universal threshold. Valid N prices this work per confirmed participant, because the cost sits in recruitment rather than interviewing.

See alsoIncidence rate (IR)Purposive sampling

A one to one qualitative session, usually 30 to 90 minutes, recruited and scheduled rather than fielded. Priced per confirmed participant.

See alsoQualitative researchDiscussion guide

The reward a respondent receives for completing a survey, also called an honorarium for professional audiences. It rises with seniority and interview length, and it is a real component of cost per complete rather than a rounding error.

See alsoCost per complete (CPI)Professional respondent

The share of people who start a screener and qualify. An 18% incidence means 18 of every 100 starts qualify. It is the single biggest driver of B2B sample cost, because every start that does not qualify still costs money to generate. Above 50% is comfortable; 20 to 50% is normal for targeted B2B; below 20% costs rise sharply; below 5% expect a large premium.

See alsoCost per complete (CPI)Screen out

A short study using only the screening questions, fielded before the real project, to measure incidence rather than assume it. Worth doing whenever the audience is unusual or the budget is tight.

See alsoFeasibilitySoft launch

Giving a participant what they need to decide whether to take part: what is collected, what it is used for, who receives it and how to withdraw.

See alsoPIIRecontact

A survey served at a point of experience, such as after a purchase or on a website, capturing feedback while it is fresh. It only reaches people who touch that channel.

See alsoRiver sample

Targets set on combinations of characteristics, such as industry by company size, rather than on each independently. They constrain recruitment far more tightly than the same targets set separately.

See alsoQuotaNatural fallout

The area of work someone is responsible for, such as finance or IT, which is not the same as their title or their seniority. Screening on function and authority separately is what keeps a B2B sample on target.

See alsoDecision-making authorityFirmographics

How long the questionnaire takes, usually reported as the median in minutes. Every minute past ten raises both price and drop off, and senior audiences stop finishing near fifteen.

See alsoDrop offCost per complete (CPI)

A rating scale measuring agreement, usually five or seven points from strongly disagree to strongly agree. Long grids of Likert items are where straightlining shows up, so they are worth keeping short in a B2B instrument.

See alsoGridStraightlining

Research that follows the same respondents over time rather than sampling fresh people each wave. It produces the cleanest read on change and depends entirely on recontact consent and low attrition.

See alsoTrackerAttrition

The range around a result attributable to sampling variation, at a stated confidence level. The familiar figures assume simple random sampling: at n=300 and 95% confidence a 50% result carries roughly plus or minus 5.7 points. Opt-in panel samples are not random samples, so treat it as an indication of precision rather than a guarantee, and report it with that caveat.

See alsoConfidence levelEffective sample size

A method that ranks a long list of items by repeatedly asking which of a small subset a respondent values most and least. More reliable than a rating scale for prioritization, and it lengthens the interview, which raises cost.

See alsoConjoint analysisLength of interview (LOI)

Error introduced by the questionnaire rather than the sample: leading wording, unbalanced scales, double barreled questions. Good sample cannot rescue a bad instrument.

See alsoNonresponse bias

Letting a subgroup split emerge from recruitment instead of imposing a quota on it. Cheaper and faster, and it means the split is not guaranteed to be readable.

See alsoQuotaOversampling

Selection where the probability of any person being chosen is unknown, which covers most opt-in panel research. Results can be highly useful and should not be reported as though drawn from a random sample.

See alsoProbability samplingRepresentative sample

Error introduced when the people who answer differ systematically from those who do not. Long B2B surveys create it by quietly filtering out the busiest and most senior respondents.

See alsoDrop offCoverage bias

A free text answer. In B2B it does double duty: it produces verbatim insight and it is the most reliable place to detect a respondent who does not know the category.

See alsoCategory knowledge question

A respondent who qualifies but is turned away because the quota cell they belong to is already full. Tight quota designs raise the over quota rate and therefore the cost.

See alsoQuotaScreen out

Recruiting extra respondents from a subgroup so it can be analyzed on its own, then weighting it back to its true proportion for total level reporting.

See alsoWeightingBase size

A recruited, profiled group of people who have agreed to take surveys. Panels give fast access to profiled respondents. For senior or narrow B2B audiences, a single panel often cannot supply the target on its own, and repeatedly drawing from the same pool risks a skew that is hard to see in the data.

See alsoBlended samplePanel fatigue

Changes in how someone answers caused by having taken part in research before. It is the reason screening questions are varied between waves of a tracker.

See alsoPanel fatigueProfessional respondent

The decline in quality and incidence that comes from surveying the same pool repeatedly. It shows up in trackers as incidence decay wave over wave, which is easy to mistake for a real market change.

See alsoPanelTracker

Information collected about panelists ahead of any particular study, used to target invitations. Profile data goes stale as people change jobs, which is why it is a targeting aid rather than a substitute for screening.

See alsoPrescreeningPanel

Personally identifiable information: anything that identifies a specific person, such as a name, email address or phone number. Research data should reach a client without it unless the study is designed for recontact and the respondent has agreed.

See alsoAnonymizationVerbatim

Carrying a respondent's earlier answer into the wording of a later question, so the survey refers to the platform or vendor they actually named. It makes an instrument feel less generic and is one way a category knowledge check is made specific.

See alsoSkip logicCategory knowledge question

Checking likely eligibility before a respondent enters the main survey, using profile data or a short set of questions. It raises incidence in the survey itself without changing the underlying population.

See alsoPanel profilingScreener

Selection where every member of the target population has a known, non zero chance of being chosen. It is what sampling error formulas assume, and it is rare in commercial B2B research.

See alsoNonprobability samplingSampling error

Someone who takes surveys often. Frequency is not dishonesty, and engaged panelists can give better quality answers than first timers. The risk worth managing is familiarity with screeners, which is addressed by varying screening questions rather than by excluding frequent participants.

See alsoPanel conditioningAttention check

Replacing direct identifiers with a code while keeping a separate key that can reconnect them. It reduces risk and does not make the data anonymous; it is still personal data.

See alsoAnonymizationPII

Selecting participants deliberately for their expertise or role rather than at random. The normal approach for executive and specialist B2B research, where insight quality matters more than statistical generalization.

See alsoHard to reach

Research that explores reasoning and context through interviews, discussions and diaries, rather than measuring incidence across a sample. Recruited per participant, not fielded per complete.

See alsoIn-depth interview (IDI)Quantitative research

Ending a respondent's participation because the response failed an agreed quality or fraud criterion. What the criteria are, and who decides, should be agreed before field.

See alsoRejection logFalse positive

Research that measures how widely something holds across a defined population, using a structured questionnaire and a base size large enough to report on.

See alsoQualitative researchBase size

A cap on how many completes come from a defined group, used to make a sample match a known population shape or to guarantee readable base sizes for the cuts you plan to report.

See alsoOver quotaWeighting

Recruiting non randomly until each target cell is full. Common, practical, and not a probability method, so the resulting sample supports description rather than formal inference.

See alsoQuotaStratified sampling

Varying the order of questions, answer options or stimuli between respondents so that position does not systematically shape the result.

See alsoGridMeasurement bias

The size of the qualified pool for an audience: people who hold the role and can be put in front of a survey. It is not the same as the number of accepted completes deliverable on your project, which is lower once your markets, quotas, timing and review criteria are applied. Ask for both numbers.

See alsoIncidence rate (IR)Accepted complete

Matching the supplier's record of completes against the survey platform's record, so both sides agree what is accepted and what is invoiced. Do it before the invoice, not after.

See alsoAccepted completeRejection log

Inviting someone who has already taken part to a further stage, such as depth interviews after a survey. It needs consent captured at the first stage, so plan it into the original brief.

See alsoInformed consentLongitudinal study

Passing respondents into a survey hosted elsewhere by URL, with separate return links for complete, screen out and over quota. The standard way sample reaches a hosted questionnaire.

See alsoRouter

A record of every complete removed before delivery, the reason and the replacement. It is the evidence a buyer needs to defend base sizes to a client, and the fastest way to judge whether a vendor actually checks.

See alsoEmployment verificationReplacement

Refielding a complete that failed the agreed review criteria so the delivered count still meets quota. Whether it is free, and on what criteria, is a supplier policy rather than an industry rule, so agree it in writing before field. Valid N replaces at no cost against the criteria set at feasibility.

See alsoRejection logQuality termination

A sample whose composition matches the target population on stated characteristics. It is only meaningful when the population and the characteristics are named, and matching on demographics does not make a sample representative on attitudes.

See alsoNonprobability samplingWeighting

The identifier passed between the supplier and the survey platform so a participant can be tracked, deduplicated and reconciled without exchanging personal data.

See alsoRedirect deliveryReconciliation

In survey methodology, the share of eligible sampled people who complete an interview. The denominator has to be stated, because different conventions treat unknown eligibility differently. Opt-in panel work usually reports a participation rate instead, which is not the same thing. What most suppliers quote is a survey start rate: the share of invitations that produce a start.

See alsoSurvey start rateIncidence rate (IR)

Respondents recruited in the moment from advertising or website traffic rather than a standing panel. It widens reach and avoids panel fatigue, but needs strong quality controls to filter duplicates, bots and low effort responses.

See alsoPanelIntercept

Software that screens an incoming respondent once and then assigns them to whichever live study they qualify for. Efficient for suppliers; it can frustrate respondents who are screened repeatedly, which shows up as lower data quality.

See alsoScreen outRedirect delivery

The list or database respondents are drawn from, such as a research panel, a professional record set or a customer file. If the frame does not match the population you care about, the study has coverage bias before it starts.

See alsoCoverage biasPanel

The number of respondents in a study or in a particular analysis. Always check which one a stated n refers to, because the total and the cut you care about are rarely the same number.

See alsoBase sizeMargin of error

The variation in a result that comes from measuring a sample rather than the whole population. It is one source of error among several and does not cover coverage, nonresponse or measurement problems.

See alsoMargin of errorCoverage bias

A respondent who starts the screener and does not qualify. They cost the supplier money and produce no data, which is why low incidence is expensive.

See alsoIncidence rate (IR)Over quota

The set of qualifying questions at the front of a survey. In B2B it carries most of the weight: a screener that asks about seniority in vague terms will qualify people who should have screened out and the study will look fine and be wrong.

See alsoIncidence rate (IR)Seniority inflation

A respondent reporting a title more senior than the one they hold, usually in order to qualify. It is a recognized form of claim fraud in B2B and one reason seniority is checked against employment evidence rather than accepted from the screener alone.

See alsoClaim fraudEmployment verification

Confirming respondent status between the supplier and the survey platform directly, rather than relying on the respondent's browser to carry it back. It reduces the risk of a manipulated return link being accepted. It is a control, not a guarantee, and signed redirect integrations offer similar protection.

See alsoGhost completeRedirect delivery

Rules that route a respondent past questions that do not apply to them. Good skip logic shortens the real interview length, which lowers drop off and cost; bad skip logic hides a screening failure until the data comes back.

See alsoLength of interview (LOI)Screener

Fielding a small share of the sample first, typically around ten percent, to confirm incidence, interview length and data quality before the full launch.

See alsoField windowIncidence test

A supplier naming the supply sources used on a project, at least at category level. Reasonable to ask for and a fair test of how a provider thinks about transparency.

See alsoESOMAR 37Blended sample

A respondent who completes far faster than the median, usually below about 40% of it, indicating they did not read the questions.

See alsoStraightliningAttention check

A result that falls below an agreed threshold under a specified statistical model, given the assumptions of that model. It does not establish that a difference is real, important, or large enough to act on, and it is not the probability that the finding is true. Read it alongside the effect size and the base size.

See alsoConfidence intervalBase size

Selecting the same answer down a grid of questions regardless of content. A common low effort pattern and a standard quality flag, especially combined with speeding.

See alsoGridSpeeder

A probability method: the population is divided into strata and a random selection is drawn within each, with known selection probabilities. Filling company size or seniority targets by recruiting from an opt-in panel is quota sampling, not stratification, and having quotas does not make a study a probability sample.

See alsoQuota samplingProbability sampling

The share of invitations that produce a survey start. Frequently quoted as a response rate, which it is not: a response rate is calculated on eligible sampled cases and interview outcomes, not on clicks.

See alsoResponse rateIncidence rate (IR)

A simulated respondent generated by a model rather than a person. Legitimate as a disclosed research technique, and fraud when passed off as a human participant.

See alsoAI-assisted responseBot

The complete group a study aims to describe, defined before any sampling happens. Everything downstream, including whether a sample can be called representative, refers back to it.

See alsoSample frameRepresentative sample

A study repeated on the same instrument at intervals to measure change. Sample consistency matters more than anything else in a tracker, because a shift in sourcing produces a trend that looks real and is not.

See alsoPanel fatigueBlended sample

A screening item with a plausible but fictitious answer option, included to catch respondents who claim familiarity with everything. Effective against guessers and low effort respondents, and it should be used sparingly so it does not screen out honest people.

See alsoAttention checkClaim fraud

The exact text a respondent typed in an open end, delivered unedited. Verbatims carry the insight a closed question cannot, and they are the single most reliable place to spot a respondent who does not know the category.

See alsoOpen endCategory knowledge question

Using a proxy or VPN to appear to be somewhere other than where you are. Network and geolocation checks flag it, but a flag is not proof of anything: corporate VPNs are normal in business environments, so the signal needs context and supporting evidence before a complete is removed.

See alsoDigital fingerprintingFalse positive

Adjusting the influence of responses after field so the sample matches known population proportions. Weighting corrects imbalance; it cannot create respondents who were never reached, and it costs you effective sample size.

See alsoDesign effectEffective sample size