Issue tree (ours)
Breaks the problem into questions. Use it when you don't yet know enough to guess.
- Branch
- "Did volume fall?"
- Best for
- The start of a case
Hypothesis-driven case interviews, guess, test, revise
A hypothesis-driven case interview means you state your best guess at the answer early, then ask for the data that would prove or disprove it, and change the guess when the data says so. It replaces asking for everything with testing the likeliest cause first. A Bain university recruiter, writing on Bain's careers blog, tells candidates to be prepared to make hypotheses about which factors matter most and take a position. Soreno, an AI voice interview practice app for consulting, investment banking and other finance candidates, runs a free profitability case with no sign-up, where you can test a first guess against real data, and full mock cases with scored feedback.
Updated
- The loop
- Guess, test, revise
- State a likely answer, ask for the data that tests it, update
- Firms
- McKinsey, Bain, BCG
- Each uses the word on its own careers or capability pages
- Worked case
- 1 branch ruled out, 1 guess confirmed
- Invented by us, every number shown
- Free practice
- No sign-up
- A full profitability case with a timer
What is a hypothesis-driven approach?
A hypothesis is a sentence that could be wrong: "profit fell because bean costs rose and prices didn't follow." Being hypothesis-driven means you say that sentence early, pick the one piece of data that would test it, and let the result decide where you go next. The alternative, asking for every number on every branch, burns the clock. In our experience, interviewers don't mark you down for a wrong first guess. They mark you down for not testing it, or for keeping it after the data said no.
Hypothesis tree (ours)
Breaks one claim into the things that must be true for it to hold.
- Branch
- "Bean costs rose more than price"
- Best for
- Once a clue points somewhere
Worked example: Corbel Coffee Roasters (ours)6 steps
Corbel Coffee Roasters, an invented roaster that sells coffee to cafes by the pound, made $12 million last year. This year it made $5.69 million. The CEO says sales volume barely moved. Why did profit fall, and what should Corbel do? This case and every number in it are invented by Soreno for practice.
First guess
Costs, not volume
Profit fell 52.6%, from $12 million to $5.69 million. The CEO says volume barely moved, so a fall that big is more likely a margin problem. Say it: "My hypothesis is that a cost rose faster than price, most likely green coffee, since it is the biggest input. To test that, I'd like price and cost per pound for both years."
Structure
The tree, with the guess marked
Draw the profit tree: volume, price, variable cost per pound (green coffee, roasting and freight) and fixed costs. Then point at the branch you are testing first. The tree keeps you MECE. The hypothesis tells the interviewer where you will start and why.
Issue trees, with examplesTest 1: volume
Ruled out: volume, minus $420,000
Pounds sold went from 6,000,000 to 5,880,000, down 2%. At last year's margin of $3.50 a pound, that costs $420,000, about 7% of the drop. Volume is not the story. Say so in one line and move on.
Test 2: margin
Confirmed: green coffee minus $6.47 million
Green coffee went from $4.00 to $5.10 a pound, up 27.5%. On 5,880,000 pounds that is $6.47 million. Price rose only $0.20 a pound, plus $1.18 million, so Corbel passed on 18% of the bean increase. The margin per pound fell from $3.50 to $2.55.
Check it adds up
The bridge sums to $6.31 million
By driver: green coffee minus $6.47 million, price plus $1.18 million, volume minus $420,000, fixed costs minus $300,000, roasting and freight minus $294,000. Together they explain the full $6.31 million drop. Green coffee alone, at $6.47 million, is bigger than the whole drop, and the price rise clawed back only $1.18 million of it. Now the hypothesis is a finding.
Next hypothesis
Price up about $1.07 a pound
Revise toward the fix: "Winning back $6.31 million on 5,880,000 pounds needs about $1.07 more a pound, 10.5% on today's $10.20. My new hypothesis is that cafes will accept most of that, because every roaster faces the same bean prices. To test it, I'd want to know what rivals charged this year and how many cafes left after the last increase."
Weak vs strong, and what to do when the data says no (ours)3 scripts
Three short scripts on the same Corbel case, written by us.
Weak opening
Asks for everything
"Can I get revenue, costs, customers, competitors and anything on the market?" Every item may be useful, but nothing says why, or what would change if the answer came back one way or the other.
Strong opening
One claim, one test
"You said volume barely moved, so I think this is margin. My first guess is green coffee, since beans are the main input. Can I see price and cost per pound for both years? If green coffee is flat, I'll move to roasting and freight."
If the data had said no
Pivot in one sentence
"Green coffee is flat, so that guess is out. That leaves fixed costs and the other variable costs. Roasting and freight are the larger pool at about $15 million a year, against $9.3 million of fixed costs, so I'll test those next." A clean pivot names what died, what is left and what comes next.
Candidate-led vs interviewer-led cases2 formats
Boston University's 2021 workshop deck describes both formats. In an interviewer-led case you get a set of pre-determined questions regardless of your structure, so it calls the case a big case made of mini cases. The scripts are ours.
Candidate-led
Your hypothesis picks the path
You choose the next branch, so the guess steers the case: "I'd like to start with margin, since volume is flat. Does that work?"
Interviewer-led
Your hypothesis fits inside each question
The interviewer picks the next question, so don't fight for your branch. Give a mini guess on the question asked, then tie the result back: "My guess is the bean cost explains most of this. Let me check. It does, which supports margin as the main problem."
Where it fits in a 30 minute case (ours)5 moments
A typical shape from our practice cases, not a firm's rule. The timing moves with the case.
Minutes 0 to 4
Clarify and structure
Confirm the goal, take a moment to think and lay out the tree. Don't force a guess before you understand the question.
Around minute 4
First hypothesis
Pick the first branch and give the reason from the prompt.
Minutes 5 to 15
Test, then pivot if needed
Each number confirms, kills or sharpens the guess. Say which.
Minutes 15 to 25
Hypothesis on the fix
Once you know why, guess the fix, size it and name the risk.
Last 2 to 3 minutes
Close answer first
The tested hypothesis becomes your recommendation, then the reasons, one risk and the next step.
When to state a hypothesis, and when not toOur view
Prep guides disagree on timing. No firm page we read sets a rule. This is what we have seen work in practice cases.
Yes
After the structure, if the prompt gives a clue
"Volume barely moved" or "costs are up across the industry" is a clue. Use it to pick the first branch and say why.
Yes
After every piece of data
Each chart or number should confirm, kill or sharpen the current guess. Say which.
Wait
When the prompt gives you nothing
A blind guess in the first minute sounds like a framework reflex. Ask one or two questions, then commit.
Skip
Market sizing and brainstorming
An estimate has no hypothesis to test, and a brainstorm needs breadth first. Structure those instead.
How do you run a case hypothesis-first?
Five habits we listen for in practice cases. None needs the word hypothesis said out loud, though saying it once early helps the interviewer follow.
Pick a branch and say why
"I'll start with costs, since you said volume is flat." One reason, tied to a fact from the prompt.
Ask for the data that would kill the guess
Not "can I see the financials?" but "can I see cost per pound for both years?" Every request should say what it is testing.
Update out loud
"Volume explains under a tenth of the drop, so I'll drop it and look at margin." The update is the point the interviewer is scoring.
Make the numbers add up
A confirmed hypothesis comes with a bridge: each driver's share of the change, summing to the total.
Turn the finding into the next guess
Once you know why, hypothesize the fix and what would test it. That is how a diagnosis becomes a recommendation.
6 case questions to practice hypothesis-first
Each prompt carries a clue that points to a first branch. Under each is what a strong opening hypothesis does.
Why did profit fall?
Diagnose first, then fix.
An airline is flying more passengers but making less money. Why?
Uses the clue (more passengers) to guess that revenue per passenger or cost per flight moved, then asks for fares and costs per seat to test it. Practice it, see a model answer
A regional grocery chain's profits are down 20% over two years. What happened?
States a first guess on price, mix or cost, names the one number that would test it and drops the branch the data rules out. Practice it, see a model answer
A manufacturer's unit costs are rising. Where would you look?
Guesses which cost line moved, such as materials or labor per unit, and checks it against volume before going wider. Practice it, see a model answer
Something got worse
A metric moved and the client wants to know why.
A streaming service's churn has doubled. What would you look at?
Splits churn by customer group and timing, guesses one cause such as a price change, and tests it with churn before and after. Practice it, see a model answer
A hospital's emergency room wait times have doubled. Why?
Guesses demand or capacity, asks for arrivals and staffed beds by hour, and updates when one of them stays flat. Practice it, see a model answer
A manufacturer's on-time delivery is falling. What would you do?
Picks the likeliest step in the chain, such as production or shipping, and tests it with lateness by step before suggesting fixes. Practice it, see a model answer
Hypothesis mistakes we hear most
From practice cases. Bain's recruiting post names the second.
Guessing with no reason
"My hypothesis is the market shrank" with nothing in the prompt behind it is a coin toss. Tie the guess to a fact you were given.
Force-fitting a framework
Reciting a memorized structure, then a memorized hypothesis, misses the case in front of you. Tailor the branches to this client.
Defending a dead hypothesis
When the data says no, say no and move. Holding on to a wrong guess is the most expensive mistake in a case.
Asking for data without saying why
Each request should name what it tests. Otherwise it sounds like fishing, and the interviewer can't help you.
Never closing the loop
A diagnosis with no fix is half a case. End with what the client should do and the number behind it.
What McKinsey, Bain and BCG publish about hypotheses
None of the firm pages we read sets a rule for when to state a hypothesis in an interview. What they do publish, read on October 31, 2026, points the same way: structure, a point of view, and testing it.
McKinsey
- In the case
- Tailored, not standard frameworks
- On the job
- Formulate and test hypotheses
McKinsey's interviewing page says there is not one logical solution, and that candidates should listen carefully and share tailored solutions rather than standard frameworks. One of its associate job postings lists formulating and testing hypotheses as part of the work.
Bain
- In the case
- Hypotheses on what matters most
- Mistake
- Force-fitting a framework
A Bain recruiting post tells candidates to be prepared to make hypotheses about which factors are most important and take a position supported by clear reasoning. It names force-fitting a pre-learned framework as a common mistake. Bain's strategy page says a hypothesis-driven approach lets executives quickly identify no regrets moves.
BCG
- In the case
- Find the most important factors
- At the firm
- Traditional hypothesis-driven approach
BCG's case prep page asks you to structure your approach, analyze data and identify the most important factors. Its customer journey page calls hypothesis-driven work BCG's traditional approach, and its customer insight page says those experts prioritize the data that will best solve the problems at hand instead of a massive volume of data.
Boston University career office
- Step 3
- Analyze and test hypotheses
- Tip
- Trim the branches
BU School of Public Health's career office teaches a four-step case: listen and clarify, structure, analyze and test hypotheses, then summarize and recommend. Its workshop deck tells you to narrow down the likely problem and trim the branches from your issue tree.
Guess, test and revise, out loud
Soreno's AI interviewer runs a full case, hands over data only when you ask for it, pushes back when your guess has no reason and scores the session. Start your free week.
Questions people ask
What does hypothesis-driven mean in consulting?
Starting from a likely answer and testing it, instead of collecting all the data first. BCG's customer journey page calls it BCG's traditional approach, and a McKinsey associate posting lists formulating and testing hypotheses as part of the job.
Do McKinsey, BCG and Bain expect a hypothesis in the case interview?
A Bain university recruiter, writing on Bain's careers blog, says to be prepared to make hypotheses about which factors matter most and take a position. McKinsey and BCG don't use the word on their interview pages. McKinsey asks for tailored solutions rather than standard frameworks, and BCG asks you to identify the most important factors. In our experience, a hypothesis is the fastest way to do both.
What is the difference between an issue tree and a hypothesis tree?
No firm page defines the two. In our usage, an issue tree breaks the problem into questions to explore, and a hypothesis tree breaks one claim into what must be true for it to hold. Most cases start with the first and move to the second once a clue points somewhere.
When should I state my hypothesis?
In our experience, right after your structure if the prompt gives a clue, and after each piece of data. If the prompt gives nothing, ask one or two questions first so the guess has a reason.
What if my hypothesis is wrong?
Say so and move on. A guess the data kills in one line is a good result: it rules out a branch. Keep it only as long as the data supports it.
Should I say "my hypothesis is" out loud?
Once, early, it helps the interviewer follow. After that, "I'll test costs next because..." does the same job without sounding scripted.
Do I need a hypothesis for market sizing?
No. An estimate has nothing to prove or disprove. Lay out the formula, state your assumptions and sanity-check the result.
Is a hypothesis the same as answer first?
Close. A hypothesis is an answer you are still testing. The pyramid principle's answer first is the tested answer at the end, said before the reasons.