The BCG Online Case (Casey), and how to prepare for it

What actually is the BCG Online Case, and who takes it?

Before any live interviewer sees your name, a share of BCG applicants are asked to complete a chatbot-based case, widely known by the name Casey. It is one of more than one online assessment BCG has used at various points, alongside things like a timed cognitive test and a career-fit questionnaire, and exactly which ones you meet, in what order, if any, appears to vary by office, role and year. This guide is specifically about the case-shaped chatbot conversation that ends in a recorded video, because it is the one candidates most often search for by name and understand least.

Not every applicant sees it. By most accounts, an invitation goes out to a subset of candidates who have cleared an initial resume screen, commonly reported as arriving somewhere around one to two weeks after an application deadline rather than immediately. Treat the exact timing and the exact trigger as unconfirmed for your specific process. BCG does not publish the mechanics of who gets invited, and what circulates online is candidate-reported rather than official.

It is also worth saying plainly what this guide is not covering. Candidates sometimes use "BCG online case" loosely to mean any of the digital steps in the process, and BCG has reportedly used other tools at different points, including a timed cognitive-style test and a game-based or questionnaire-style career-fit assessment. Those test different things through different mechanics, and lumping them together with Casey in your preparation wastes time you could spend on the format you are actually about to sit.

How does Casey differ from BCG's other online assessments?

BCG's digital hiring tools have changed more than once, and different offices and roles report different combinations of them. Rather than assume you know which one you have been sent, it helps to know the rough shape of each by name, because the name in your invitation email is the reliable signal, not a generic phrase like "online test".

None of these three are guaranteed for any given candidate, and BCG does not publish which offices or roles use which combination in a given hiring cycle. If your invitation names a tool this table does not describe well, trust the invitation and whatever practice materials it links to over any guide, including this one.

How does the chatbot format actually work?

The chatbot portion is commonly reported as running somewhere in the 25 to 35 minute range, structured as a single business case delivered as a written prompt. A short business situation appears at the top of the chat window, then a sequence of questions arrives one at a time. Most accounts put the question count at six to ten. You answer inside the chat interface itself: no scratchpad shared with a person, no spoken back-and-forth, just typed or selected answers against a running countdown.

The detail that changes behaviour more than any other: once you submit an answer, it is commonly reported as gone. By most accounts there is no way to revisit a previous question, no way to revise a number after you have moved on, and no way to pause the session once it starts. A live interviewer will let a slow start recover over the next twenty minutes. Casey, by these accounts, will not, because there is no next twenty minutes with the same question still open in it.

How is this different from sitting across from a real interviewer?

Most of what makes a live case interview forgiving is missing here. In a live case, thinking aloud earns you something: the interviewer hears the false start, hears you correct it, and can credit the correction. In the chatbot, only the final typed answer exists. A confused first thirty seconds that resolves into a clean answer would score fine live and, on these accounts, scores on the clean answer alone here, provided you reach it before time runs out.

• There is no clarifying question you can ask back. If a prompt reads as ambiguous, you answer the most reasonable interpretation of it and move on, rather than asking what was meant
• There is no exhibit handed over at a moment chosen for you. Data typically appears attached to the specific question that needs it, so you read fast rather than requesting it
• There is no credit for the structure you didn't type. A structure held only in your head, however good, is invisible to a chatbot reading a text box
• There is no interviewer to read for a steer. The chatbot doesn't nod, doesn't lean forward, and gives you none of the small human cues candidates use live to sense whether they're on the right track

The net effect cuts both ways. Some candidates who feel pressure under a live interviewer's gaze report doing better here, because the pressure is a countdown timer rather than a person. Others who think well out loud, and build an answer through the act of talking, report doing worse here, because the chatbot only ever sees the destination, never the route that got you there.

The same content, a different set of things that go wrong

Take one moment and play it both ways. In a live case, the interviewer hands you a table and asks what you make of it. You can say "give me twenty seconds", read it in silence, then speak, and that pause costs you nothing. On Casey, the same table appears attached to a multiple-choice or short-text question with a visible countdown running, and the equivalent pause has a number attached to it that you can watch shrink. The reading itself does not get harder. The decision about how long to spend reading it, before you have to commit to something, gets harder, because the cost of hesitation is made explicit instead of invisible.

The reverse is also true. A live interviewer occasionally rescues a candidate who has clearly misread the question, with a small "just to clarify" nudge that resets the exchange. Casey extends no such courtesy. Answer the question you think was asked, not the question you wish had been asked, and if the prompt genuinely supports two readings, say which one you are answering rather than trying to cover both in a field built for one.

What logistical details should I sort out before I start it?

Because Casey commonly allows one attempt with no pause and no do-over, the logistics matter as much as the case preparation, and they are the part candidates most often leave until the last minute. None of this is about the case content. All of it is about not losing points to something you could have fixed in advance.

• Read the invitation email fully before you click start, and note whether it gives you a window to choose from or a hard deadline, since by most accounts once the session begins there is no pausing it
• Test your camera and microphone in whatever application you'll actually use beforehand, not for the first time during the one-shot video segment at the end
• Close anything that could interrupt you mid-session: chat apps, calendar pop-ups, a second monitor with notifications on
• Have paper and a pen, or a blank note file, open specifically for your own scratch work, since nothing you type into a question box is visible again once you submit it
• Pick a stretch of time when you are not rushed into or out of it. Casey has no sense of whether you have a call in ten minutes, only a clock of its own

None of this guarantees a strong performance. It removes a specific category of preventable loss: a shaky internet connection discovered mid-video, a scratch calculation done on the back of an envelope you can't read, a session started with eleven minutes free when the actual test needs closer to forty. Fix what is fixable before the clock starts, so that what actually gets tested is the case thinking, not the setup around it.

What is Casey actually testing?

By consistent account across independent sources, Casey is built to test the same underlying skills a live case tests: reading a business situation and structuring an approach to it, doing case-style arithmetic correctly and with the right units, drawing a defensible conclusion from data you have only seen briefly, and communicating a recommendation inside a hard constraint. What differs is the delivery mechanism, not the target.

How the multiple-choice and numeric questions are actually scored, whether partial credit exists anywhere in the system, whether some questions carry more weight than others: BCG has not published that, and candidate theories about it online are guesses dressed as certainty. Treat every specific claim about Casey's internal scoring that you read anywhere, including implicitly in this guide, as unverified. What is well supported by repeated, independent accounts is the shape of the test, not its grading internals.

BCG's own materials describe the firm's approach to problem-solving as hypothesis-driven: form a working answer early, then test and refine it as evidence arrives, rather than exhaustively working through every possibility before committing to a view. That habit is worth carrying into a Casey session specifically, because the format leaves no room for the alternative. A structuring answer that names a leading hypothesis and one clear reason for it reads better in four lines than a list of five equally weighted areas to investigate, because the second version answers the question "what could this be about" rather than the question actually asked, which is closer to "what do you think this is about, and why".

Independent accounts also tend to describe the question sequence as moving through recognisable content categories rather than being randomly ordered: a structuring or information-gathering question near the start, one or two quantitative questions built on a table or chart, a question asking for judgement or a ranked set of priorities, and a closing question that asks you to synthesise everything into a view. That sequence maps closely onto how a live case is built, which is one more reason ordinary case practice transfers here more than chatbot-specific drilling does.

Where do prepared candidates lose points without noticing?

None of the six traps below are about case ability. Every candidate who reaches Casey has already cleared a resume screen and knows roughly how a case works. What trips people up here is almost entirely mechanical: habits built for a live conversation, applied unchanged to a timed, unforgiving text interface where they no longer fit.

The common thread across all six is that Casey punishes the gap between what you know and what you actually submit before the countdown ends. A candidate with weaker case fundamentals who reliably gets a complete, unit-labelled, conclusion-first answer into the box before time runs out will often outperform a stronger case thinker who spends the same budget on a beautifully structured answer to a question that closed while they were still writing it.

What about the recorded video recommendation?

The chat portion ends and, by most accounts, you get roughly a minute to prepare and roughly a minute to record a spoken recommendation, on camera, with no retake. This is the one part of Casey that behaves like a live case: you are talking, not typing, and whoever eventually reviews the recording is judging delivery as well as content.

Notice the shape: recommendation first, then reasons tied to numbers already established earlier in the chat, a named risk, a next step. It is the same closing discipline a live case rewards, compressed into a minute with no interviewer there to rescue a rambling start.

Treat the recording itself as a real presentation, not a formality tacked onto the end of a test. Find a quiet room, sit rather than slouch, and look at the camera lens rather than at your own preview image, which is a small habit that reads as far more confident than it feels while you're doing it. None of that compensates for a weak recommendation, but a strong recommendation delivered while glancing sideways at your own reflection undersells itself for no reason.

How should you prepare for it specifically?

If Casey is somewhere ahead of you, the fastest useful preparation is not more chatbot simulators. It's the underlying case skills done under a clock: structuring fast, doing clean arithmetic with the right units, and closing with a recommendation before you're asked to. Our case rooms run those same skills live, with a voice on the other end instead of a countdown, which is a harder and more realistic rehearsal for the thinking Casey is trying to measure, even though the delivery mechanism looks nothing alike. A candidate who can build a clean structure out loud, on the spot, under a real clock, generally has little trouble compressing that same habit into a text box. The reverse is less reliable: drilling chatbot-shaped questions in isolation teaches you the interface, not the underlying judgement a live interviewer, or a written scorecard on the other end of one, is actually looking for.

All of our coaches are AI personas, designed by ex-MBB interviewers, and we say so everywhere: you are talking to a rubric with a voice around it, not a person. The first case is free.