This page explains how respondents reach an MX8 Labs study, which responsibilities remain with the sample provider, and how each source is measured during fielding. For what happens after a respondent arrives, see Data quality methodology. For configuration instructions, see How to set up respondent sources.
The sourcing model
MX8 Labs does not operate a proprietary consumer panel.
Every study draws from one or more respondent sources, each configured on the survey and measured independently. A respondent source can be an integrated sample provider, an external sample provider, a first-party customer list, a phone or SMS route, an ad unit, or responses collected elsewhere and imported into the platform.
The model has three practical consequences:
Source selection is explicit. Each source is configured on the study, and its fielding performance is reported separately.
Sources can be combined. Where one provider has limited reach in a market or audience, the study can field through additional providers or other source types.
External providers are supported. A sample provider can be configured through either a dedicated integration or the generic bring your own sample flow.
Where recruitment and incentives sit
Recruitment and incentive design belong to the sample provider, not to MX8 Labs.
Each provider runs its own recruitment model and its own reward structure: loyalty and rewards programs, publisher and app intercepts, affiliate acquisition, community panels, specialist business-to-business and healthcare panels. Incentive levels vary by provider, market, and audience, and are set by the provider against local expectations.
MX8 Labs applies validation after respondents enter the study and reports fielding outcomes separately for each source. Validation is documented in Data quality methodology and summarized under Measuring respondent sources in field below.
The observable measures are live incidence, termination patterns, and the proportion of entrants rejected by the quality pipeline. These describe what each source delivers to the study; they do not expose the provider's upstream recruitment process.
Respondent source types
Online panel is one route among several, and studies routinely combine them.
| Respondent source | What it is | Typical use |
|---|---|---|
| Integrated providers | Automated sourcing and fielding against demographic targeting, with quotas passed through to the provider | General population and profiled audiences at scale |
| External providers | A generic integration flow for any sample provider not already integrated | Existing sample-provider relationships |
| First-party data | Upload a customer list and generate up to one million unique survey links | Customer, lapsed-customer, and CRM-segment research |
| Twilio interactive SMS and text-to-web | Surveys delivered and answered over text | Voter files, texting houses, audiences not on panel |
| Twilio voice | Interactive AI voice surveys | Audiences reached better by phone than by screen |
| Call center | Links issued to an interviewer-administered operation | Complex or assisted interviews |
| Airtory in-ad | Surveys launched from inside a display ad unit, with in-ad questions and click-through to the rest of the survey | Reaching an audience in the media environment itself |
| Third-party import | Ingest completed responses collected outside the platform | Reporting alongside data collected elsewhere |
| Synthetic | Synthetic Twins trained on first-party respondent data, or Synthetic Profiles generated from a prompt | Extending real data, never replacing measurement |
Targeting for integrated providers is derived from the survey's quotas and can be extended with any additional targets the provider exposes. It is edited as validated structured configuration, so a malformed definition is caught before fielding.
Synthetic respondent sources are labeled distinctly throughout the platform and are excluded from human-quality screening by construction. Their statistical treatment is documented in Synthetic data and statistical inference.
Sustaining coverage market by market
Coverage in a given market rests on four mechanisms rather than on a single provider relationship.
Several respondent sources per market. Where a market is served by more than one integrated provider, a study can configure several. A provider that underdelivers in one market is not a single point of failure for that market.
Local targeting taxonomies. Every sample provider expresses demographics differently. Demographic definitions and screening questions are matched to each local provider's own taxonomy rather than forcing one market's category scheme onto another.
Local language. Surveys are translated automatically into the local language, with a review interface and an export and import round-trip so a translation house can work on the file and return it without losing question identifiers. Respondents are served their language automatically from browser and operating system preferences. See Multilingual surveys and Fielding surveys internationally.
Standard screeners. Reference screeners exist for common markets, so a general population definition stays consistent from study to study rather than being rebuilt each time. See Standard screeners.
Where data must remain in a region, studies can be hosted in a specific region. See Local data residency.
Check coverage before fielding
A feasibility run checks a target audience against an estimated incidence rate and expected survey duration.
It returns achievability and expected cost for that specification. Feasibility runs in the platform and through the public API and MCP integration, allowing candidate audiences or markets to be compared before fielding. See MCP and OpenAPI integration.
Run feasibility against the real target rather than relying on a headline coverage figure. The question that matters is whether this audience is reachable in this market at this incidence.
Measuring respondent sources in field
Once a respondent source is live it is measured continuously, and separately from every other source on the study.
The summary tab shows a live incidence rate per source, calculated from respondents who have entered the survey rather than from the planning estimate. Hovering the figure breaks it into four outcomes:
- Completed - finished the survey and passed quality checks. This is the numerator of the incidence calculation.
- In progress - entered but not yet finished. A persistently high figure usually indicates a long survey or a confusing question.
- Terminated - screened out by a termination question or a failed quality gate. Normally the largest driver of incidence, and the most useful figure to compare across sources.
- Poor quality - flagged and removed by the fraud and quality systems. A spike here is a reason to investigate or pause the source.
Because the figure is computed per source, a blended underperformance can be attributed to a specific respondent source rather than to the audience as a whole, and that source can be paused or re-targeted on its own. Live incidence moves during the first few hundred completes, so we would not act on it until a source has at least 50 to 100 entries.
For a fuller view of what is happening inside a source, covering where respondents terminate, which devices struggle, and how each demographic group is progressing, see Field reports.
Soft launch
Before releasing the full sample, field a small batch through a separate respondent source and confirm incidence and quota behavior against expectation. Keeping the soft launch on its own respondent source makes the early read visible in isolation and easier to act on. See Setting up quotas.
Boost sources
Adding respondents from a particular group without distorting the topline is handled with a boost: a separate respondent source with its own targeting definition, configured to weight back to the primary respondent source. The extra interviews raise the achieved base for that group, and the weighting scheme prevents the boost from moving the overall distribution.
See Reaching low-incidence audiences and, for the calibration behind it, Weighting methodology.
Scope and limitations
Three things this model does not do.
It does not make provider recruitment transparent. We measure what a respondent source delivers; we do not audit how the provider recruited it. Where a study requires a documented recruitment chain, that has to be agreed with the provider directly.
It does not eliminate coverage bias. Online respondent sources, however well combined, do not constitute a probability sample of a national population. Calibration to known population margins addresses composition, and the limits of that framework are set out in Weighting methodology.
It does not guarantee coverage. Some audiences are not reachable at a workable incidence through the available providers. Feasibility identifies those cases before fielding begins.