Brainstorm

Help the researcher discover, sharpen, and assess research questions. Start from what they actually bring. Develop a defensible question and a feasible next learning step, allowing the question and proposed contribution to change as evidence accumulates. A clear one-sentence question is useful; an anticipated finding is not required.

Distinguish the starting point (what prompted interest), the approach to inquiry (how to learn), and the potential contribution (what knowledge could change). These need not have matching labels. A new phenomenon may invite empirics-first exploration, a focused theory test, or an assumption challenge. A familiar phenomenon can also warrant empirics-first investigation.

Work with the researcher

Read supplied observations, notes, papers, and data descriptions before asking for more context. Ask a few questions that change the research direction or feasibility; avoid a questionnaire covering every possible concern. When useful, propose a small set of substantively different questions and explain their tradeoffs. If the researcher already has a focused question, stress-test it directly.

Separate documented observations, literature-supported claims, conjectures, and unresolved access or measurement questions. Search or inspect sources when novelty, an attributed assumption, a current phenomenon, or data availability matters. Scope absence claims to the literature actually checked. If verification is unavailable, identify the specific uncertainty and continue with conditional reasoning.

The agent can generate alternatives and critique reasoning. Keep the researcher’s substantive interests visible and explain why a proposed direction is promising. Do not require the user to arrive with a polished argument before helping.

Recognize the starting point

Use the relevant guidance below and combine approaches when helpful. These are entry points, not mutually exclusive study types or mandatory stages.

Starting with a new phenomenon

Give the phenomenon an empirical identity: define what counts as an instance, distinguish neighboring occurrences, and establish facts and usable constructs before demanding a complete explanation. This draws on phenomenon-based research’s role in making emerging domains investigable [3].

For this skill, develop a short account of:

• Occurrence and evidence: what has been observed, the source and date, and what remains anecdotal or unverified. A product announcement establishes availability claims, not adoption or behavioral effects.
• Distinctiveness: what changed in behavior, actors, institutions, technology, or relationships? Compare an earlier or neighboring practice. Distinguish new-to-the-world, newly widespread, newly observable, and newly noticed by the researcher.
• Consequences: whose choices, experiences, or outcomes may change? Explain the substantive uncertainty; popularity in the press alone does not establish research value.
• Initial questions: what is it, how prevalent is it, how does it unfold, what varies, or what consequences follow? Definition, classification, measurement, and prevalence may be the immediate research tasks.
• Next observation: identify a feasible way to learn more, such as inspecting actual instances, comparing cases, conducting interviews, piloting a measure, or checking a data source.

Treat existing concepts as candidates to examine. Explain both useful analogies and consequential differences before declaring an entirely new construct or a failure of existing theory. Where suitable theory already makes a clear prediction, a phenomenon-originating project can proceed deductively.

Developing an empirics-first inquiry

Golder et al. describe empirics-first research as beginning with a substantive real-world observation or problem and developing knowledge through empirical investigation; testing or creating theory is a possible outcome, not a requirement [1]. Use it when existing explanations offer limited guidance, several accounts remain plausible, or basic regularities are unsettled.

Adapt their iterative progression to brainstorming:

1. Identify an opportunity: articulate the substantive uncertainty and its relevance; consult literature to assess what guidance exists.
2. Explore the terrain: formulate open questions, identify evidence to obtain, and revise the questions as observations suggest where to investigate more deeply or broadly.
3. Advance understanding: consider whether the work could establish a regularity, improve concepts or measurement, support an explanation, or inform a decision [1, Figure 1 and Table 3].

Empirics-first inquiry can use experiments, qualitative evidence, surveys, or existing records. It does not require possession of a large dataset. Literature can inform the investigation throughout [1].

The Journal of Marketing’s empirics-first guidance emphasizes abstraction beyond the immediate setting, learning from failed robustness checks, and reporting discovery as it occurred [2]. Apply that here by stating what a pattern might teach beyond a single case and what additional evidence could change the interpretation. A failed check can suggest a boundary or measurement problem while also weakening the original claim.

Keep an account of consequential changes to the question and analyses already inspected. Label explanations formed after seeing results as exploratory; do not retrospectively portray them as prior hypotheses. Where stronger confirmation is needed, identify suitable new evidence, a genuinely untouched holdout, or a prospective study. Repeated exploration of the same data is not independent confirmation. Null or mixed findings can be informative; assess uncertainty and substantive magnitude instead of selecting a preferred sign.

Developing questions through literature and theory

Gap-spotting: identify unresolved knowledge, why it matters, and what an answer would change [4]. This may involve competing accounts, insufficient evidence, or extending an informative perspective to an overlooked area. Check whether seemingly conflicting studies examine comparable constructs, populations, outcomes, and conditions. A new setting or dataset needs an argument for the knowledge it enables.

Problematization: identify a consequential assumption underlying an existing explanation, document where that assumption is made, explain why reconsidering it matters, and develop an alternative with different implications [5]. Examine assumptions in the preferred alternative too. Adding a variable or importing another theory does not by itself establish an assumption challenge. Use the term in this stricter sense here; if a course uses broader categories such as incompleteness or inadequacy, preserve and explain that distinction.

Theory-led inquiry: articulate the prediction or unresolved implication, its conditions, and the observation that would discriminate among accounts. A known direction can leave an important magnitude, scope, or decision threshold uncertain. A formal model may clarify incentives, interactions, aggregation, or welfare when those are central to the question.

An untested challenge motivates inquiry; demonstrated theory failure requires evidence inconsistent with a specified prediction under its relevant assumptions. Do not make theoretical novelty or an assumption challenge a condition for valuable empirical work. Gap-spotting and problematization can combine, but neither is mandatory for a phenomenon-originating project.

Connect the question to evidence

State the actor or unit, focal construct or outcome, context, comparison where relevant, and time horizon. Distinguish constructs from indicators and interventions. A concept can be useful before it has a validated measure; treat developing that measure as work to do.

Match the evidence task to the question:

Use a sub-question tree when it clarifies dependencies. Each branch should answer a specific uncertainty and name the evidence needed; do not require every branch to estimate a causal effect or fill a quota of mechanism, heterogeneity, and robustness tests.

Distinguish mechanisms from moderators. A subgroup difference or treatment effect alone does not establish mediation. For observational adoption, identify selection and concurrent changes before proposing a causal interpretation. For structural explanations or welfare counterfactuals, specify what identifies the parameters and what depends on modeling assumptions.

Check feasible access, linkage keys, observation timing, measurement resolution, and missing actors or outcomes. Mark proposed data collection or linkage as unverified until supported. If the preferred design is unavailable, consider a narrower question or a different evidence source and make the change in scope explicit.

Converge provisionally and hand off

Compare candidates by the importance of the uncertainty, likely knowledge gain, interpretability, feasible evidence, and cost of resolving the largest unknown. Distinguish a weak idea from an appealing idea with a current access problem. Avoid numerical rankings that imply unsupported precision.

Choose a provisional direction with the researcher where a choice is needed. Scale the output to the task; a short discussion need not become a full proposal. A useful research brief contains:

• Question and motivation: one clear sentence plus the phenomenon, unresolved knowledge, assumption, or decision that motivates it.
• What is known and open: observations and sources, candidate explanations where relevant, and important uncertainties.
• Potential contribution: what could be learned, stated conditionally. Empirical characterization, measurement, prediction, explanation, method, and conceptual development are possible contributions; a primary emphasis can coexist with others.
• Evidence plan: the smallest useful set of observations or sub-questions, access status, and assumptions needed for the proposed claims.
• Stress test: evidence that would weaken the favored explanation or require reframing; whether a precise null, heterogeneous pattern, or unresolved mechanism would still teach something.
• Next action: the most useful feasible check and how its possible outcomes would affect the decision to continue, narrow, or redirect.

Use “We ask whether…” or “This could establish…” before findings exist. Keep an alternative direction alive when a consequential uncertainty prevents choosing. The exit criterion is a defensible next learning step, not a known punchline or a complete paper outline.

For deeper source verification, use the literature-review skill; for data inspection, use eda. Paper-writing develops the resulting manuscript argument, including conceptual grounding. The separate conceptual-framework skill draws a model from supplied research when a figure is useful. These are optional handoffs when available, not prerequisites for brainstorming.

Worked example: a new shopping practice

This example is hypothetical. It assumes no verified adoption pattern, available dataset, or finding.

Starting observation: the researcher reports seeing consumers delegate product searches to AI shopping assistants. First establish what “delegate” includes: obtaining suggestions, creating a shortlist, choosing a product, or executing a purchase. Collect concrete instances and compare them with conventional search and recommendation tools.

Phenomenon question: which shopping tasks are delegated, and how do consumers retain or relinquish control? A pilot combining observed sessions with interviews could assess whether these distinctions are meaningful and observable. It would not establish population prevalence without an appropriate sample.

Empirics-first direction: investigate how search paths, considered products, and purchases vary across forms of delegation. Request a data dictionary or run a small prospective observation pilot; do not assume browsing logs can be linked to purchases. Any observed association may reflect which consumers and tasks select into delegation.

Possible explanatory follow-up: reduced search effort, narrower exposure, and pre-existing preferences could produce overlapping patterns. Evidence on effort and exposure may help distinguish accounts. A causal follow-up might randomize assistant access if feasible, while distinguishing access from actual use.

Potential contribution: characterize forms of delegation and their empirical relationships with shopping behavior; further evidence may support causal or conceptual claims. Theory development remains optional. An assumption-challenging version would first need sources establishing the specific premise it questions.

Next step: inspect a small set of real sessions and available fields to decide whether task delegation can be measured. If only assistant availability is observable, narrow the question or collect different evidence. Do not claim a result about search or purchases yet.

Sources and scope

The workflow and example are an applied synthesis for this skill. Source-specific ideas are attributed above; the workflow is not a universal journal template. Sources checked on 2026-09-30:

1. Golder, P. N., Dekimpe, M. G., An, J. T., van Heerde, H. J., Kim, D. S. U., & Alba, J. W. (2023). Learning from Data: An Empirics-First Approach to Relevant Knowledge Generation. Journal of Marketing, 87(3), 319–336. Publisher full text; especially Figure 1, Table 3, and the discussion of literature’s role. First published online in 2022.
2. Journal of Marketing (2025). Call for Papers: Special Issue on Empirics First. Official editorial guidance, especially “What Does EF Research Look Like?” and the discussion of consumer research. Used for methodological guidance, not as a claim that submissions remain open or that its criteria apply to every journal.
3. von Krogh, G., Rossi-Lamastra, C., & Haefliger, S. (2012). Phenomenon-based Research in Management and Organisation Science: When Is It Rigorous and Does It Matter?. Long Range Planning, 45(4), 277–298. Author repository record; abstract and indexed accepted-manuscript passage on identifying and distinguishing phenomena. Full manuscript retrieval was unavailable in this update; no detailed five-activity procedure is attributed to it here.
4. Sandberg, J., & Alvesson, M. (2011). Ways of Constructing Research Questions: Gap-Spotting or Problematization?. Organization, 18(1), 23–44. Publisher abstract supports the distinction between identifying gaps and challenging assumptions; the practical prompts above are a synthesis.
5. Alvesson, M., & Sandberg, J. (2011). Generating Research Questions Through Problematization. Academy of Management Review, 36(2), 247–271. Publisher abstract supports the assumption-challenging definition.