All chapters
Hypothesis-driven thinking
Core toolkit - Chapter 7
07

Hypothesis-driven thinking

Use hypotheses as testable working views: notice clues, form a direction, test it with evidence, update, and choose the next analysis.

Read time
9 min
Chapter
07
Level
Core toolkit
What you'll take away
  • Form a light working hypothesis from real clues, not vibes.
  • Name the evidence that would prove or disprove your view.
  • Update your hypothesis when new data changes the case.

In the last chapter, you learned how to make your thinking easy to follow. Now we need to give that thinking a direction.

This is where hypotheses come in.

The word can feel more intimidating than the idea. Some candidates hear "hypothesis" and think they are supposed to magically know the answer early. Others become afraid of being wrong, so they avoid taking any view at all. Neither is the point.

The whole idea
A hypothesis is not a guess you defend. It is a starting point you test.

The intuition

Hypothesis-driven thinking is how you move through a case without wandering.

Your structure gives you a map. Your hypothesis gives you a compass.

The map

Here are the areas we could investigate.

The compass

Based on what we know so far, this is where I would look first.

That is the whole idea. You met a lighter version of this in Structuring, where you learned to say where you would start. There, the starting point came from the shape of the problem. A hypothesis sharpens the same decision: now it comes from what the clues suggest is most likely, and you stay ready to change course as evidence arrives.

A useful hypothesis sounds like
I have an early view of what might be going on, and I know what evidence would test it.
Hunch

I think it is competition.

Hypothesis

Given the timing and the revenue decline, competition could be driving lower traffic. I would test that by looking at traffic by region before and after the competitor entered.

One is a hunch. The other is a plan.

Why it matters

Case interviews reward direction. Not fake certainty. Direction.

In a real client problem, you rarely have time to analyze everything with equal depth. You have to decide what is most likely to matter, test it, update, and move. That is why interviewers watch whether you can focus the analysis instead of drifting from bucket to bucket.

Good hypothesis-driven thinking shows:

  • You can connect clues into a working view.
  • You can prioritize the next analysis.
  • You know what evidence would change your mind.
  • You can update when the data disagrees.
  • You can stay decisive without becoming rigid.
The useful question
What is my current best view, and what would I need to test next?

That question gives your brain somewhere to stand.

The core moves

Hypothesis loopUse a view to choose the next test
  1. 1
    NoticeWhat clues do we have?
  2. 2
    FormWhat is my current working view?
  3. 3
    TestWhat evidence would confirm or disprove it?
  4. 4
    UpdateWhat changed after the data?
  5. 5
    MoveWhat should we analyze next?
A hypothesis earns its place only if you know how to test it.
Sentence frame
My working hypothesis is [view], because [clue]. I would test that by looking at [evidence]. If that is true, I would [next step]; if not, I would look at [alternative].

You will not always say the full frame out loud. Sometimes it will be too heavy. But the logic should be in your head every time.

Throughout this chapter, we will keep using the same example: a coffee chain whose profits dropped 15% last quarter.

1. Notice: start with clues, not vibes.

A useful hypothesis starts from something real. That something might be in the prompt, a clarifying answer, a chart, a number, or a business pattern you know. It does not have to be conclusive. It just has to be enough to justify where you look first.

For the coffee chain, the initial prompt is light:

Profits dropped 15% last quarter.

At that point, you do not know whether the issue is revenue, costs, region, product mix, competition, or something else. So a strong early view should be modest.

Too far

My hypothesis is that a competitor caused the decline.

Better early view

At this stage, I would not want to assume the cause yet. My first working view is that we should quickly test whether the decline is revenue-driven or cost-driven, because that will determine the rest of the analysis.

The hypothesis is not "competitor." The working view is simply that the first split that matters is revenue versus costs.

Later evidenceWhen the clues get sharper
Profit fell 15%
Region B
decline concentrated
  • other regions stable
  • local issue possible
Revenue
down 25%
  • large enough to matter
  • not just noise
Competitor
opened nearby
  • timing lines up
  • traffic loss plausible
Costs
rose slightly
  • not enough alone
  • still worth checking
Now the hypothesis can become sharper because the evidence is sharper.
Sharper working view
My working hypothesis is that the profit decline is mainly revenue-driven and concentrated in Region B, likely from traffic loss after the competitor entered.

2. Form: make the view directional, not dramatic.

Candidates often think a hypothesis needs to sound bold. It does not. It needs to be useful.

Useful phrases
  • My working hypothesis is...
  • Based on what we know so far...
  • I would initially expect...
  • One likely driver could be...
  • The evidence seems to point toward...

These phrases let you take a position without pretending the case is solved.

Broad first test is fine
Based on what we know so far, my working hypothesis is that we need to determine whether the decline is revenue-driven or cost-driven first. I would test that by comparing revenue and major cost categories versus last quarter, then go deeper on whichever side moved more.
When evidence is thin
I do not have enough information to form a specific cause yet, so I would start with a broad revenue-versus-cost test. Once we see which side moved, I can form a sharper hypothesis.

That is not weak. That is honest and structured. Hypothesis-driven thinking is not forcing yourself to state a theory when the evidence is thin. It is staying directional at every stage.

3. Test: name the evidence that would change your mind.

This is what separates a hypothesis from a hunch. If you cannot name the test, you do not have one.

View only

I think it is a traffic issue.

View plus test

I think it may be a traffic issue. I would test that by comparing customer visits in Region B before and after the competitor opened, while checking whether average order value stayed stable.

For the coffee chain, if your working hypothesis is competitor-driven traffic loss, you might test:

Test planCompetitor-driven traffic loss
What evidence would change my mind?
Traffic
before vs after competitor
  • Region B trend
  • closest stores
AOV
average order value
  • stable or down?
  • mix shift?
Customers
segments and dayparts
  • who declined?
  • morning vs afternoon
Alternative
costs or operations
  • give the hypothesis a fair chance to be wrong
A good test does not only confirm your favorite answer. It also gives the hypothesis a fair chance to be wrong.

4. Update: change your view when the case changes.

A hypothesis is a tool, not a promise. If the data disagrees, you update.

New evidenceA different version of the case
Revenue is flat
Ingredient costs
+18%
  • large cost pressure
Labor costs
+10%
  • second cost pressure
Margin
14% -> 9%
  • profitability compressed
The facts now point away from a demand issue and toward costs.
Rigid

I still think it could be a demand issue, so I would keep looking at customers.

Updated

That changes my view. If revenue is flat but ingredient and labor costs rose significantly, the decline now looks cost-driven rather than demand-driven. I would shift the analysis toward the biggest cost increases and whether they are temporary, structural, or controllable.

That update is not an admission of failure. It is the skill. The interviewer does not need you to be right from minute one. They need you to use evidence well.

5. Move: let the hypothesis choose the next analysis.

The point of a hypothesis is not to sound sophisticated. The point is to decide what to do next.

After every meaningful factChoose the next analysis
  1. 1
    More likelyWhat does this evidence support?
  2. 2
    Less likelyWhat does this evidence weaken?
  3. 3
    Next testWhat should we analyze now?
Coffee-chain next move
If Region B revenue is down 25% while other regions are stable, my working hypothesis is that this is a localized revenue issue. I would next test traffic, average order value, and local competition in Region B before spending time on company-wide cost cuts.

Hypotheses connect everything: clarifying questions give you the boundaries, structuring gives you the map, hypotheses help you prioritize the map, math and exhibits test the view, and synthesis turns the updated view into a decision.

Initial, working, and revised hypotheses

It helps to separate three versions.

Hypothesis typesHow the view evolves
Initial hypothesis
First light direction
I would first test whether the decline is revenue- or cost-driven.
Working hypothesis
Current best view
The decline appears revenue-driven and concentrated in Region B.
Revised hypothesis
Updated after new evidence
Given flat revenue and rising ingredient costs, I would now shift toward a cost-driven explanation.
The goal is not to have a perfect initial hypothesis. The goal is to keep improving the working hypothesis as evidence comes in.

What weak looks like

Weak hypothesis · story without evidence
AI
Prompt
A coffee chain's profits dropped 15% last quarter. What happened, and what should they do?
YOU
Weak answer
My hypothesis is that a new competitor entered the market and stole customers. I would look at competitors and then recommend marketing.
Why it falls flat
This jumps to a story without evidence. The candidate may be right by luck, but the thinking is not reliable. They have not separated revenue from costs, clarified whether the decline is broad or concentrated, or named what data would prove the competitor theory.

What strong looks like

Strong hypothesis · light view, clear test
AI
Prompt
A coffee chain's profits dropped 15% last quarter. What happened, and what should they do?
YOU
Strong answer
At this point, I would keep the hypothesis light. Since profits are down 15%, my first working view is that we need to test whether this is revenue-driven or cost-driven. I would start by comparing revenue and major cost categories versus last quarter. If revenue is down, I would then segment by region, channel, traffic, and average order value to find where the decline is concentrated. If costs are up instead, I would look at ingredients, labor, rent, and one-time costs. Once we see which side moved, I can form a sharper hypothesis about the root cause.
Direction without overclaiming: a view, a reason, and the first test.
Later, after Region B evidence
Based on the evidence so far, my working hypothesis is that the profit decline is mainly a Region B revenue issue, likely driven by traffic loss after the new competitor opened. I would test that by comparing Region B traffic before and after the competitor entry, checking average order value, and seeing whether the drop is concentrated in stores closest to the competitor.

When to say the hypothesis out loud

You do not need to announce a hypothesis every three minutes. Say it out loud when it helps the interviewer follow your direction:

  • After the opening structure, if you have enough information for a light initial view.
  • After a major exhibit or calculation changes what seems likely.
  • Before choosing between two possible analysis paths.
  • During interim synthesis, to explain what you believe so far.
  • In the final recommendation, as the conclusion your evidence now supports.
Robotic

My hypothesis is...

More human

This makes me think the issue is more likely traffic than pricing, so I would test customer visits next.

Same skill. Less theater.

Common traps

The usual ways hypotheses go wrong
  • Guessing too early.
  • Sounding certain when the evidence is thin.
  • Treating the first hypothesis like a final answer.
  • Ignoring data that disagrees.
  • Using "hypothesis" as a fancy word for "hunch."
  • Failing to name the test.
  • Jumping straight from hypothesis to recommendation.
  • Forcing every case into one theory when the facts are still messy.

The most common trap is wanting the hypothesis to be impressive. Do not aim for impressive. Aim for testable.

How to use this later in CaseLab

When you are ready for live practice, use this chapter as the lens for how you move from one analysis to the next. You are not trying to prove a hunch; you are practicing how to test and update a working view.

After the opening structure, state a light working view and the first data you would test. After each major finding, update your view out loud:

Update line
This makes X more likely and Y less likely, so I would next test Z.

Later, review the feedback on whether your hypotheses were evidence-based, testable, appropriately humble, and updated when the facts changed.

The standard
Do not just ask, "Was my hypothesis right?" Ask: did my hypothesis help me choose the next smart analysis?
Up next
Chapter 8 · Market sizing
Next, learn how to estimate an unknown market with a transparent path, reasonable assumptions, and a useful business implication.
Continue to Chapter 8
Finished Chapter 7?Mark it complete to track your progress through the handbook.
← PreviousCommunicationUp next →Market sizing