Thinking out loud
The interviewer doesn't just want your answer. They want to watch you get there. This is the most misunderstood skill in case interviews, and the most learnable.
Consulting is a client-facing job. You will spend years explaining complex analyses to executives who don't have time to read everything. The case interview is a proxy for that skill. The interviewer is asking: can this person make their thinking followable in real time?
A candidate who gets the wrong answer but narrates their reasoning clearly will often outscore a candidate who gets the right answer through silent guessing. This is counterintuitive but true.
Signposting
Tell the interviewer where you are and where you're going. Use explicit transition phrases. Signposting keeps the interviewer oriented. Without it, they lose track of where you are in your framework and whether you're making progress.
Moving between sections of your framework, starting a new line of inquiry, shifting from diagnosis to recommendation.
“I've finished looking at the revenue side — I don't see a major issue there. I'm going to move to costs now, starting with the variable cost structure.”
“I've identified the root cause — it's the COGS spike. Let me now think about what that means for our recommendations.”
A signpost has two parts: close the last section (with a finding) and open the next one (with a specific question). Both in one sentence.
Hypothesis narration
State what you think is true, and explain what evidence you're looking for to test it. This mode is powerful because it shows the interviewer you're not just collecting data. You're testing a specific idea.
Forming a hypothesis, asking for data, interpreting information the interviewer gives you.
“My hypothesis at this point is that the cost increase is concentrated in variable costs — specifically COGS — rather than fixed costs. I'd expect to see COGS as a percentage of revenue rising. Could you share what the gross margin has been over the past three years?”
“Interesting — margins are down 4 points. That's consistent with my hypothesis. Let me drill into what's driving the COGS increase — is it input costs, volume-related inefficiency, or a mix shift?”
Always pair a data request with the hypothesis it's meant to test. “Can you tell me what the COGS breakdown looks like?” is weaker than “My hypothesis is that input costs drove the margin compression — can you share the COGS breakdown so I can test that?”
Navigating uncertainty
When you don't know, don't go silent. Narrate the uncertainty and how you're resolving it. Narrating uncertainty correctly, instead of hiding it, actually builds interviewer confidence. It shows you know what you don't know.
You're not sure which direction to go, you're doing mental math, or you've hit an ambiguous data point.
“I'm seeing two possible explanations here. The margin compression could be a supplier cost issue, or it could be a product mix shift — selling more of the low-margin SKUs. These would have different solutions, so I want to figure out which it is. Can you tell me if the product mix has changed this year?”
“Let me do a quick sanity check on this number — if COGS went up $40M and revenue only grew $20M, that's a $20M net impact on profit, which accounts for about 60% of the total decline. So we know cost is the main story but not the only one. Let me look at what else is moving.”
If you need a moment for math, say so: “Give me just a second to work through this number...” Then narrate the math as you do it. Don't go silent for more than 20 seconds mid-analysis.
60% analysis. 40% narration.
A rough guide for how to split your airtime. Less than 40% narration and the interviewer loses the thread. More than 40% and you're filling space without adding substance.
Weak vs. strong narration
Click each tab to compare. Same analysis, completely different impression.
[30 seconds of silence]
“So I think the issue is costs. Specifically I think it's COGS. The gross margin went from 44% to 39% which is 5 points. On $800M revenue that's $40M. And I think it's because of supplier issues. So they should renegotiate their supplier contracts.”
What's wrong
- No signposting, so the interviewer doesn't know what section you're in
- No hypothesis stated before asking for or interpreting data
- No visible reasoning process: silence, then a conclusion dump
- Jumped straight to a recommendation without testing it
- Interviewer can't evaluate your thinking, only your answer
“Okay — I've ruled out revenue as the primary driver since it's up. I want to move to costs now. My hypothesis is that variable costs are the culprit — specifically COGS — because fixed costs would need to jump significantly to explain a $40M profit drop and the business hasn't opened new locations. Can you share gross margin over the past two years?”
[Interviewer: Gross margin fell from 44% to 39%.]
“That's consistent with my hypothesis — a 5-point gross margin compression on $800M revenue is exactly $40M. So COGS is our story. Now I want to understand why COGS went up — there are three explanations: input cost inflation, a shift toward lower-margin products, or operational inefficiency. Which of these would be most useful to look at first?”
What's right
- Signposted the move from revenue to costs
- Stated a hypothesis with reasoning before asking for data
- Connected the data back to the hypothesis explicitly
- Identified three sub-causes — showed structured thinking
- Invited the interviewer to guide the next step
Common narration mistakes
Going silent for more than 20 seconds mid-case
Silence at the start (during structuring) is fine and expected. Silence in the middle of an analysis is different. It breaks the interviewer's ability to follow along and creates an uncomfortable dynamic.
Say: "Give me just a second to work through this number..." Then narrate the math as you do it.
Narrating without advancing
"So there are many things we could look at here... it could be revenue, it could be costs, it could be the competitive environment... there are a lot of factors..."
This is narration with zero analytical content. You're talking but not moving. Every sentence should either advance the analysis or signal where you're going next.
Asking for data without explaining why
"Can you tell me what the COGS breakdown looks like?" gives the interviewer nothing to work with. They don't know what you're testing or where you're headed.
Always pair data requests with a hypothesis: "My hypothesis is that input costs drove the margin compression — can you share the COGS breakdown so I can test that?"
Three moments. Three modes.
Read each moment from a case. Decide which narration mode you'd use, then say your response out loud before revealing the answer.
You just finished analyzing revenue and found it's flat. Time to move to costs.
Which mode do you use? What do you say?
"Revenue is essentially flat — I don't see a meaningful driver there. I'm going to shift to costs now. I'll start by looking at the fixed vs. variable split to figure out which side is growing faster than the business."
- Closed the previous section with a finding
- Announced the pivot to costs
- Stated the specific question you're investigating next
The interviewer just told you COGS jumped 8% this year. You have a theory about why.
Which mode do you use? What do you say?
"An 8% jump in COGS is significant — that's well above inflation. My hypothesis is that this is driven by either input cost increases — raw materials or supplier pricing — or a mix shift toward lower-margin products. These would have different solutions, so I want to isolate which one. Has the product mix changed this year, or has the company been sourcing from the same suppliers at similar volumes?"
- Quantified the significance of the data point
- Offered two competing hypotheses
- Explained why distinguishing between them matters
- Asked a specific question to resolve it
The interviewer gives you data that doesn't fit your hypothesis. You're not sure what it means yet.
Which mode do you use? What do you say?
"Interesting — that's not what I expected. If supplier costs are actually flat, then the COGS increase isn't coming from inputs. That points more toward a mix shift or an operational inefficiency — maybe yield losses or rework costs in production. Let me reconsider. I want to look at the product-level margin data to see if there's been a shift in what they're selling. Could we look at revenue and COGS broken out by product category?"
- Acknowledged the unexpected data explicitly
- Updated the hypothesis in real time
- Didn't panic or go silent
- Identified the new direction and explained why
Rate your narration
Think about your last practice case (or imagine one). Check each habit you consistently do.
Practice narration live.
The best way to build this skill is to use it. Try a full case and focus on narrating every move: signpost, hypothesize, navigate.
Try a case