BCG Casey Dataset Question: How to Pick 3–4 of 8

Part ofBCG Casey Online Case: The Complete Guide

Casey's first question is not math. It shows you about eight datasets and asks which ones you need to solve the case. The right answer is the smallest set that covers every driver of the client's objective, usually three or four. Below: what the question looks like, what candidates report about scoring, a four-test elimination rule and six worked examples you can use as a quiz. For the rest of the test, see the complete BCG online case guide.

datasets offered in the opening question, as candidates report

~8

Source: managementconsulted.com

of multi-select options that are correct, per one prep site

30–60%

Source: casebasix.com

questions in the full Casey case

8–10

Source: igotanoffer.com

average time per question

~3 min

Source: hackingthecaseinterview.com

What is the dataset-selection question in Casey?

Casey is BCG's chatbot-led online case: one business problem, 8 to 10 sequential questions and a short video recommendation at the end, according to Management Consulted and IGotAnOffer. Some candidates now receive it as a HireQuotient invitation; Management Consulted reports that the underlying case experience is largely the same.

The opening question is "consistently reported as a dataset selection task," in Management Consulted's words. Casey shows roughly eight data sources and asks which ones are most relevant to the client's problem. IGotAnOffer describes it as a multiple-choice question that is usually followed by an open-ended question asking you to justify your choice. No source states a fixed number to pick. CaseBasix reports that only 30 to 60% of the options are correct, which on an eight-option list works out to about three to five, and three or four in most cases.

BCG itself does not publish Casey specifications. Its official case interview page describes the underlying skill: structure your approach and "identify the most important factors." That is what this question tests, before you have seen a single exhibit.

How is the dataset question scored?

Nobody outside BCG knows the exact rule. Here is what is reported:

Is it worse to pick too many or too few?

With the rule unknown, pick a strategy that holds up under any plausible one.

• Over-selecting (picking 6 of 8 "to be safe") shows the opposite of the skill being tested. If a justification question follows, you have to defend every extra pick in writing, and a pick you cannot defend costs you there too.
• Under-selecting leaves a driver uncovered. A BCG consultant quoted by IGotAnOffer warns that some datasets "will leave a gap that is impossible to guess through," and that a sub-optimal selection "will have a knock-on effect on the rest of the case."

The tiebreaker when you are torn between a fourth and a fifth pick: if I removed this dataset, which driver would go uncovered? If the answer is "none," drop it.

Which datasets should you pick? The four-test elimination rule

Hacking the Case Interview states the rule in one line: "choose the fewest options that cover the most ground." Here is how to apply it. Start by writing the objective as a decision or an equation, such as profit = volume × price − costs. Its terms are your drivers. Then put each dataset through four tests, in order.

1. Irrelevant. Does it feed any driver? Org charts, social media followers and employee surveys rarely do.
2. Redundant. Is its content already inside a dataset you are keeping? This is the most-cited mistake. The example from IGotAnOffer is an itemized P&L plus an inventory list with purchase costs, where the P&L already contains those costs.
3. Out of scope. Right topic, wrong segment, geography or time period, such as the home market when the question is about entering a new one.
4. Too early. A drill-down, such as invoices by supplier, is useful only after you know which cost line moved. The first question is about scoping, not diagnosis.

Then run the gap check. Every driver needs at least one kept dataset. This is the MECE test applied to data: no overlaps (test 2) and no gaps (the gap check).

What do good picks look like? 6 worked examples

These cases are original DrillCase practice examples, not real Casey content. To use them as a quiz, cover the Verdict column and pick your set before you read on.

Profitability: coffee chain, profit down while revenue is flat

Drivers: volume, price, costs, and a market benchmark to tell a company problem from an industry one.

Market entry: European e-bike maker considering the US

Drivers: market attractiveness, competition, entry cost, unit economics.

M&A: private equity fund buying a regional dental-clinic chain

Drivers: the target's standalone performance, market attractiveness, price paid.

Pricing: software firm launching a project-management tool

Drivers: cost floor, competitor reference, customer willingness to pay.

H is the tempting one here. Market size matters for the business case, but the question is what to charge.

Growth: snack brand aiming to double revenue in 5 years

Doubling in five years needs about 15% a year (2^(1/5) ≈ 1.149), which usually means looking beyond the core. Drivers: core baseline, market growth pockets, distribution white space, acquisitions.

Operations: hospital emergency department, waits doubled from 2 to 4 hours

Drivers: demand, capacity, process time.

Compare F here with G in Example 1. A benchmark earns its place only when you cannot observe the driver directly. In the hospital case, arrivals data measures demand directly, so the benchmark adds nothing.

What mistakes do candidates make on the dataset question?

1. Picking the summary and the detail. A P&L and a cost list, or a market report and the same numbers in an analyst note. According to the BCG consultant quoted by IGotAnOffer, most candidates make this mistake: two datasets that give the same information, with a crucial piece still missing.
2. Choosing interesting data over decision data. Satisfaction scores and brand surveys feel relevant but rarely change the answer.
3. Diagnosing too early. Supplier invoices and clinic-level splits come after you know where the problem is.
4. Scope drift. The home market, the wrong segment, or a period that doesn't match the case.
5. Skimming the prompt. Hacking the Case Interview advises an extra 30 seconds on the objective, because every later answer depends on it.
6. Rushing to bank time. The consultant quoted by IGotAnOffer notes that candidates often rush the early questions and finish with time to spare, "at the cost of accuracy." Online Case Secrets suggests investing extra time in the first question but capping any single question at four to five minutes.

How can you practice the dataset question?

Practice the scoping step on its own, then inside a full case:

• Self-quiz. Go back to the six tables, cover the Verdict column and give yourself three minutes per case. Score a point for each driver you covered, and lose one for each redundant or irrelevant pick.
• Write the justification. In two sentences, name the driver each pick covers. If you can't, the pick is padding.
• Full-case practice. DrillCase's Casey-style simulations include multi-select scoping questions like this one, followed by exhibits and math. One demo case is free.

Once your scoping is reliable, move on to the last step of the case, the 60-second video recommendation. For a side-by-side view of how chatbot cases differ across firms, see the chatbot case format breakdown.