The primary challenge in B2B research is not “finding a specific job title,” but ensuring that responses represent the individual, the establishment, the account, or the entire enterprise—and that respondents genuinely possess the corresponding information. Purchasing, deployment, daily usage, compliance review, and renewal are typically handled by different roles; targeting only senior executives may yield budget and strategic insights but miss actual friction points; targeting only active users fails to explain contract terms, risks, and switching costs. Reliable samples must separate business entities from individual reporters and restrict each role to answering questions grounded in their direct experience.
Job titles are at best a recruiting clue, not proof of qualification. An “IT Manager” in a small company might simultaneously approve budgets, configure systems, and handle support, while in a large enterprise they may only oversee a single region. The real screening criteria should focus on what someone has done in a recent specific decision, which records they have access to, and which stage of the process they can influence—not what is printed on a business card.
Clarify the Sample Unit, Reporting Unit, and Analysis Unit First
Business research involves at least three units that are easily confused. The sample unit is the entity drawn from the list—possibly a company, a customer account, or an establishment; the reporting unit is the person or organizational unit that actually provides information; and the analysis unit is the entity that conclusions ultimately describe. These three can be identical or distinct.
The UK Office for National Statistics’ business register methodology distinguishes between enterprise groups, enterprises, local units, and the reporting units that receive questionnaires; a reporting unit may cover an entire enterprise or only a subset of local units. Similarly, the U.S. Census Bureau’s economic survey standards note that the sample unit, reporting unit, and tabulation unit need not be the same—large enterprises may provide data disaggregated by region, line of business, or establishment. These distinctions are highly practical for commercial B2B research:
- When studying the usability of a specific function, the analysis unit is typically actual users and their tasks.
- When studying renewal risk, the analysis unit is typically the customer account, but multiple roles may be needed to provide evidence.
- When studying deployment across retail locations, the establishment may be the unit, and headquarters respondents may not know what is happening at each store.
- When studying market penetration rates, the unit should be the target business population; website-registered accounts cannot substitute for all enterprises.
At the project launch, specify “how many respondents per company,” “whether one person can represent multiple roles,” “how to aggregate multiple establishments into an enterprise,” and “whether final tabulation will be per person or per account.” Without these rules, a seemingly large sample may simply indicate that a few major clients contributed many respondents each.
Translate the Buying Committee into Verifiable Behaviors
Do not simply ask “Are you a decision-maker?” Instead, anchor on a recent qualifying purchase or renewal and ask respondents to select the specific steps they actually performed:
- Needs Identification: Defined the problem to solve, the using teams, and the success criteria;
- Candidate Development: Searched for, recommended, or approved vendors for evaluation;
- Technical and Security Evaluation: Verified integration, permissions, data, security, and operational conditions;
- Economic Evaluation: Compared budgets, contracts, procurement processes, and total cost of ownership;
- Final Approval: Held signing authority, budget authority, or veto power;
- Deployment Management: Oversaw configuration, migration, training, support, and usage governance;
- Daily Usage: Uses the product to perform concrete tasks and bears the process costs.
One person may play multiple roles, but “knowing the final choice” is not the same as participating in the choice. Eligibility questions should also verify the time window, product category, organizational scope, and depth of involvement. For example, when studying enterprise software renewals in the past twelve months, participants should at minimum state the month of the most recent renewal, the steps they performed, whom they collaborated with, and the final output produced. Do not request vendor names, contract amounts, or confidential screenshots, and do not list all correct answers in public recruitment invitations.
The U.S. Census Bureau’s guidance on business questionnaire pretesting emphasizes understanding which records respondents can access, how they obtain information from colleagues, and what estimation processes they will use. Therefore, when asking about finances, usage figures, or security incidents, first ask about the data source: system records, internal reports, colleague input, or personal estimates. Respondents who cannot access records should not be forced to provide seemingly precise company-level numbers.
150 Respondents May Represent Only 120 Accounts
Here is an illustrative calculation. A business software team obtains 150 valid respondents covering 120 customer accounts. Among these, 110 accounts each have a single respondent, while ten large accounts each have four respondents. If account identifiers are ignored, a researcher might treat 150 responses as independent evidence.
Assume that among the 40 respondents from the ten large accounts, 28 express a positive attitude toward renewal—a rate of 70%. Among the remaining 110 single-respondent accounts, 55 express a positive attitude—a rate of 50%. When pooled at the individual level, 83 positive responses out of 150 yield 55.3%.
If the goal is “the attitude of a typical account,” you should first aggregate multi-role responses within each account using a predefined rule, then give each account a single contribution. Assuming the within-account average positivity for the ten large accounts remains 70%, and the other 110 accounts at 50%, the equal-weighted account result is (10 × 70% + 110 × 50%) / 120 = 51.7%. The individual-level estimate is 3.6 percentage points higher than the account-level estimate, simply because positive respondents from large accounts were overrepresented.
Neither method is inherently correct: if the question is “what is a random user’s experience,” the individual may be the analysis unit; if the question is “whether a random customer account is at risk of non-renewal,” the account is the unit. The error occurs when using person-level sample sizes to answer account-level questions, or when presenting account-level aggregated results as representing all employees’ experiences. The report should simultaneously present the number of distinct accounts, number of individuals, distribution of respondents per account, and the aggregation rule.
Multiple Roles Are Not for Voting: They Locate Evidence Conflicts
When the economic buyer and the administrator within the same account give conflicting answers, it does not necessarily mean one of them is wrong. Different roles see different outcomes. Another illustrative example: 42 accounts have both an economic buyer and an administrator interview. The questions are “Did deployment meet business objectives?” (buyer) and “Is routine account provisioning easy?” (administrator).
In 15 accounts, both are positive; in 18 accounts, the buyer is positive but the administrator is not; in 3 accounts, the administrator is positive but the buyer is not; and in 6 accounts, both are negative. The buyer is positive in 33 accounts (78.6%), while the administrator is positive in only 18 accounts (42.9%); 21 pairs of answers are discordant, representing exactly half of the 42 accounts.
These results cannot be collapsed into an “average satisfaction of 60.7%.” A more informative interpretation is: 18 accounts show business-level endorsement but daily operational friction, which may foreshadow future usage and renewal risk; the 3 reverse accounts may indicate smooth operations but unproven business value. The next step should separately examine usage coverage, permission tickets, goal attainment evidence, and the person responsible for renewal—not force the two respondents to reconcile onto a single score.
When designing multi-role modules, include a common core plus role-specific sections. The common core enables comparison of different roles within the same account on goals, outcomes, and risks; the role-specific modules only ask about tasks within that role’s domain. The account evidence packet can retain the source, timestamp, and confidence level for each judgment, without forcing the system to prematurely synthesize a single “truth score.”
Four Recruitment Sources: Who Does Each One Miss?
Customer or account lists are the easiest way to verify real relationships and allow sampling by account status, size, and usage stage, but they can only represent current or historical customers. If invitations are hand-picked by sales managers, they tend to favor accounts with strong relationships, recent activity, or ease of contact. A more robust approach is to sample systematically from the full list based on explicit rules, and then have the client team help update contact information.
Professional panels and occupational panels can reach competitors’ customers and non-customers, but business identity, category experience, and role must be re-verified within the project. A supplier’s “decision-maker” tag cannot replace step-level screening. The report should state how panel members are recruited, how often qualifications are refreshed, and how deduplication was performed for this project.
Industry associations, conferences, and professional communities work well for highly specialized roles but may overrepresent people who actively seek learning, are comfortable with public visibility, and join associations. Referrals from sales, partners, or respondents can build trust and reach hard-to-access groups, but tend to concentrate within existing relationship networks. Multi-source mixes should retain a source field to compare business structure, roles, and key answers; if channel differences are substantial, simply merging into a single “market proportion” is not acceptable.
Business Qualification Verification Must Balance Fraud, Change, and Confidentiality
Business size, industry, job function, and usage status all change over time, and external databases have built-in delays. Qualification can be verified in layers: first, use a trusted account list or sample frame for basic attributes; second, confirm current status within the questionnaire; and third, validate roles using recent behaviors. Corporate domain email, professional profiles, licenses, or company documents can serve as supplementary evidence when project needs require them; but freelancers, small businesses, and genuinely qualified roles under confidentiality restrictions may lack these artifacts.
Quality rules should be defined before fieldwork begins. Clear timeline contradictions, repeated participation with the same identity, or an inability to describe any relevant step may trigger review, but brief answers, fast response times, or reluctance to name specific clients should not independently determine fraud. Identity information and research answers should be stored with separated access; analysis tables should retain only randomized account IDs, qualification results, and roles. If responses will be linked back to the customer’s account, both the invitation and consent must explicitly state this; when anonymity is promised, individual-level answers must not be sent back to sales teams.
The Sample Plan Should Prioritize Account Coverage, with Role Coverage as a Secondary Goal
A renewal study can first draw accounts by size, industry, region, lifecycle stage, and risk status, then specify mandatory roles for each account. Not every account needs a full buying committee: small businesses often have one person covering multiple roles, while large enterprises may require supplemental recruitment for critical conflicts. Set parameters in advance:
- The number of distinct accounts needed per stratum, not just the number of completed individuals;
- Qualified reporting roles and substitute roles for each research question;
- The maximum number of respondents per single account, and the within-account aggregation method;
- How to handle non-response, single-role-only, or role-conflict accounts;
- Whether large-account case studies and market-level estimates are presented separately.
Respondents from the same enterprise share processes, culture, and contracts, creating clustering in statistical inference. For complex quantitative studies, cluster-robust standard errors, multi-level models, or prior aggregation to the account level are options; the choice depends on the research design and statistical capacity. The most basic bottom line is to avoid inflating account coverage with individual counts and to refrain from overly detailed industry rankings when only a handful of enterprises are represented.
A Disclosure Template Ready for B2B Samples
The methodology note could state: “This study used customer accounts as the primary analysis unit. Accounts were drawn from lists that met product-status and geographic conditions using stratified sampling. A total of 150 qualified participants were obtained, covering 120 distinct accounts; ten accounts contributed four participants each, and the remaining accounts contributed one each. Participants confirmed their role through specific behaviors within the last twelve months related to need identification, technical evaluation, economic approval, deployment management, or daily use. Account-level metrics were first combined within each account using a pre-specified rule, then weighted equally across accounts; individual experience tables are reported per person, and clustering is handled at the account level in uncertainty analyses.”
Next, disclose sources, refusals and substitute contacts, account and role base sizes, incentives, deduplication, linkage permissions, and limitations. When a business population frame needs to be defined, refer to defining the target audience and sample frame; for rare occupations or decision roles without a complete list, see recruiting hard-to-reach participants; after fieldwork, address design-based differences in business structure using survey weighting and effective sample size—but never use weighting to mask account and role gaps.
Methodological Basis and Further Reading
- UK Office for National Statistics: Business activity, size and location methodology: enterprise groups, enterprises, local units, and reporting units.
- U.S. Census Bureau Appendix D3-B: sample units, reporting units, tabulation units, and complex enterprise treatment in economic surveys.
- U.S. Census Bureau Appendix A2: Business respondents’ data access, colleague collaboration, estimation processes, and cognitive testing.
- AAPOR Transparency Initiative: Disclosure of sources, screening, recruitment, incentives, and coverage gaps in non-probability business samples.