Formulating Analytics Research Questions

Formulating Analytics Research Questions

Quick Answer: After framing the situation, analyzing current state, defining future state and scope, CBDA practitioners formulate research questions—specific, data-answerable questions tied to a decision. Good questions name the population, outcome or relationship of interest, and how the answer will be used; bad questions restate tools, rest solutions, or ask things data cannot resolve.

Identify the Research Questions is Domain 1 of the CBDA blueprint (about 20% of exam weight). Earlier Domain 1 work frames the business situation, maps current and future state, and sets scope, assumptions, and approach. This section closes the Domain 1 loop: you turn that framing into research questions that drive sourcing (Domain 2), analysis (Domain 3), interpretation (Domain 4), and decision influence (Domain 5).

On scenario items, the highest-scoring move is often to rewrite a vague or solution-shaped ask into one or more testable research questions before any technique or tool is chosen.

From Business Situation to Research Question

A business situation describes context, need, decision owners, and uncertainty. A research question is the precise inquiry analytics will attempt to answer so that uncertainty shrinks enough for action.

Translation pattern

Use a repeatable pattern so questions stay decision-linked:

1. Start with the decision — Who will do what differently if we learn the answer?
2. Name the population — Customers, claims, SKUs, regions, time window.
3. State the unknown — Driver, difference, size, likelihood, or ranking—not a tool.
4. Link to actionability — The answer should map to levers the organization can pull.
5. Check answerability — Data exists or can be obtained within scope, ethics, and time.

Weak translation: Situation = "churn is high" → Question = "Why is churn high?" (too broad; no population, no decision path).

Stronger translation: Situation = retention VP can fund one major intervention this quarter → Question = "Among U.S. online subscribers who canceled voluntarily in the last 12 months, which product, engagement, and support factors most strongly associate with cancelation, ranked by estimated revenue impact of reversible drivers?"

Characteristics of Good vs Bad Research Questions

CBDA-aligned research questions share four traits. Exam stems often bury a bad question inside a confident-sounding request; your job is to recognize which trait fails.

Specific

Specificity is not buzzwords; it is operational detail. Prefer "median days from lead to closed-won for enterprise software deals sourced by partner channel in FY2025" over "sales cycle length." Specific questions prevent teams from analyzing the wrong grain (account vs opportunity, order vs customer).

Answerable with data

Not every important business question is an analytics research question. Values conflicts ("Should we exit the market on principle?"), pure political choices, or unknowns that require experiments the organization will not run may be business questions without an analytics path. CBDA practitioners surface that distinction instead of promising insights they cannot produce.

Answerability also respects scope and assumptions from earlier Domain 1 work. If identity resolution across CRM and billing is unproven, a household-level lifetime value question may need to be deferred or rephrased at account grain.

Decision-linked

Ask: If we answered this perfectly, what would change Monday morning? If the honest answer is "nothing—people would just know more," the question is incomplete. Link answers to budget, staffing, pricing, outreach, process design, risk appetite, or product prioritization.

Neutral on technique

Technique selection comes after category fit (next section). A research question should not smuggle in "build a neural net," "use Snowflake," or "create a real-time dashboard." Those are design choices, not research questions.

Hypotheses vs Exploratory Questions

Both forms appear in CBDA work. Confusing them leads to either confirmation bias (testing only favorite stories) or undirected fishing (endless exploration without a decision).

Using hypotheses well

Hypotheses should be falsifiable and pre-specified enough that success criteria are clear. Stakeholder beliefs are hypothesis sources, not proven facts. Document them as assumptions until tested. On the exam, if a sponsor asserts "price is the only reason we lose," the strong move is often to treat that as a hypothesis and define how data would support or challenge it—not to accept it as scope.

Using exploratory questions well

Exploration is legitimate when the current state shows definition chaos, unknown driver sets, or new products/markets with thin history. Guardrails keep exploration professional:

• Time-box exploratory work.
• Pre-commit how findings will be promoted into confirmatory questions.
• Avoid presenting exploratory patterns as causal proof.
• Keep the decision owner in the loop so exploration does not become entertainment.

Sequence often seen in mature teams: exploratory questions to map the space → prioritized hypotheses → confirmatory analysis → refined recommendations.

Multiple Related Questions and Prioritization

Real business situations rarely collapse into one perfect question. A retail conversion problem may need questions about stage drop-off, cohort differences, fulfillment failures, and promo cannibalization. CBDA practitioners decompose the situation into a question set, then prioritize.

Decomposition

Group related questions by decision theme:

1. Sizing — How large is the problem or opportunity?
2. Localization — Where (segment, stage, region, product) does it concentrate?
3. Drivers — What factors associate with better or worse outcomes?
4. Levers — Which controllable actions relate to improvement?
5. Constraints — What limits (capacity, policy, cost) bound the solution space?

You do not need all five on every initiative, but missing sizing or localization often produces precise answers to the wrong slice of the problem.

Prioritization criteria

Example prioritized set (subscription retention):

1. P1 — Sizing/localization: Which tenure and plan cohorts account for the most voluntary churn revenue loss in the last 12 months?
2. P2 — Drivers: Within the top two loss cohorts, which engagement and support events most strongly associate with cancelation in the 30 days before cancel?
3. P3 — Lever readiness: For the top associated drivers, which are controllable by retention ops within current policy and staffing?

Lower priority (deferred): brand sentiment NLP across all social channels—interesting, but not required for this quarter's intervention choice.

Official-Style Sample: Sales Conversion-Time Questions

CBDA competencies emphasize research questions that look like real business analytics work—not academic curiosities. A classic flavor is sales conversion and time-to-conversion analysis.

Framed situation: B2B leadership believes the sales cycle is "too long," pipeline reviews are contentious, and capacity planning for account executives is unreliable. Sales ops can change territory design, specialist assignment, and stage-exit criteria—but only with evidence.

Weak questions (exam traps):

• "Should we buy a new CRM AI package?" (solution restatement)
• "Why is sales bad?" (not specific, not decision-linked)
• "Predict every deal's close date with deep learning." (technique-first; may skip baseline)

Strong research questions (sample flavor):

1. Descriptive baseline: What is the median and 90th-percentile days from first qualified opportunity to closed-won (and to closed-lost) by segment, channel, and deal size for the last four complete quarters?
2. Localization: Which stage-to-stage transitions contribute the most elapsed time and drop-off for enterprise deals above $100k?
3. Diagnostic association: How do early product-specialist involvement, proposal turnaround time, and number of stakeholders relate to conversion probability and cycle time, controlling for deal size and industry?
4. Decision-oriented: If leadership can fund only one process change this half, which lever—specialist assignment SLA, proposal SLA, or stage-exit criteria clarity—is associated with the largest improvement in conversion rate or cycle time among controllable factors?

These questions move from what is true today to where friction concentrates to which levers matter—exactly the Domain 1 → later domains handoff CBDA expects.

Trap: Research Question That Restates the Solution

Trap: Stakeholders (or candidates) phrase the research question as a yes/no on a preselected deliverable.

Why this trap is costly: Solution-restating questions skip uncertainty reduction. You can "answer" them with opinions and vendor demos without ever measuring the business. CBDA exam scenarios reward candidates who reframe tool questions into outcome and decision questions.

Self-check before you lock questions

1. Does each question avoid naming a tool, vendor, or algorithm as the object of study?
2. Can you state the decision maker and decision timing?
3. Is the population and time window explicit?
4. Would opposing evidence change a plan (falsifiability for hypotheses)?
5. Is the set prioritized so analysis starts with the highest-impact answerable question?

Lightweight Research-Question Register

Document questions in a short register shared with stakeholders:

Approval of the register is a Domain 1 exit criterion: analysis that answers unapproved questions is often wasted effort.

Formulating research questions is where business analysis skill becomes analytics rigor. Master the translation from situation to specific, answerable, decision-linked questions; use hypotheses and exploratory forms intentionally; prioritize multi-question sets; and never let a solution restatement masquerade as research.