In a first round of eight candidates, I would expect five to read the exhibit back to me. Revenue grew. The blue segment is the biggest. The line dips in 2024. I wrote the exhibit. I know what it says.
Here is what no guide says plainly: the insight on a case exhibit is never printed on it. It is a number you have to make from two numbers that are. The case writer started with the insight, built the smallest chart that contains its ingredients, and left the insight off the page. Describing the ingredients scores nothing. Combining them scores everything.
Below: how exhibits get built, the five numbers that carry the insight, a table to redo with a pencil, and my mark scheme from the other side of the desk.
How an exhibit gets written
Case writers start with the sentence the candidate should be able to say at the end, such as "the growth is all volume and price per unit is falling," then build an exhibit from which that sentence can be derived but not read. If it could be read straight off the page, the exhibit would test eyesight.
The firms' own practice cases show it. McKinsey's Talbot Trucks case hands you a total-cost-of-ownership comparison and asks, verbatim, "Using this data, what can you infer about the differences in TCO for diesel trucks vs eTrucks?" The follow-up asks for the eTruck price at which the two totals are equal. That price is on nobody's slide; it exists only after you subtract. McKinsey's Electro-Light case gives a $2 price, $1.90 of cost per bottle, and $40 million of fixed cost; the break-even volume, 400 million bottles, appears only when you notice ten cents has to cover forty million dollars. Bain's FashionCo case does the same with two promotions: the "good answer" is the computed comparison, the "better answer" adds risks and next steps.
Three published cases, one design: the exhibit holds the inputs and the scoring holds the output. Every exhibit I have written worked the same way.
The five numbers that are never printed
Almost every exhibit insight I have scored was one of five derived quantities.
| Derived number | Printed ingredients | Question it answers | What it sounds like |
|---|---|---|---|
| Gap | Two values of the same thing, two years or two regions apart | How big is the problem, in the CEO's units? | "The gap is $12 million, two-thirds of the margin decline we are here to explain." |
| Ratio | A money column next to a volume column | What is each unit worth, and is that moving? | "Revenue per visit fell from $80 to $75 while visits rose 23%." |
| Share | A part and a whole, or client growth and market growth | Who is winning? | "We grew 4% in a market growing 9%. At a 20% share, that is nearly a point lost." |
| Rate | Start value, end value, number of years | Does the trend bite before the client runs out of road? | "Costs compound at 8%, revenue at 3%. From a 15-point margin the lines cross in about three and a half years." |
| Product | A number here and one from the previous exhibit | Does this page change the answer? | "400 million bottles at an eighth of a gallon is 50 million gallons; divide by the market size on the last exhibit and that is the share we need." |
Four of the five are one operation: subtract, divide, divide, or divide and annualize. The fifth requires remembering something from three minutes ago, which is why interviewers like it. Candidates who score well on exhibits are not faster at arithmetic; they have decided in advance that the first thing they say will be one of these five.
Where the writer hides it, by chart type
The chart type decides which number gets hidden. In 1984 William Cleveland and Robert McGill showed in the Journal of the American Statistical Association that people judge position along a common scale far more accurately than length or angle: position judgments were 1.4 to 2.5 times as accurate as length judgments and about twice as accurate as angle judgments. A stacked bar asks you to judge length. A pie asks you to judge angle. When a case writer picks one of those formats, it is rarely for decoration.
| Exhibit type | Easy to see | Usually hidden | What to compute |
|---|---|---|---|
| Clustered bars, two periods | Which bar is taller | Each series' growth against the total | Percent change per series vs the total |
| 100% stacked bar | Biggest segment | Absolute size of a segment when the total changed | Share times total, each period |
| Pie chart | Largest slice | Anything within five points, and the total | Every slice as a number, from the footnoted total |
| Table | Everything, which is the problem | The ratio between two columns | Divide, then rank |
| Indexed chart (2019 = 100) | Growth since the base year | Absolute levels; which series is actually bigger | Ask for base-year values |
The defense is the same in every row: stop looking and start writing numbers. If a 100% stacked bar does not print the total behind it, ask. Asking for the denominator is the first signal the interviewer gets that you know what a percentage is.
A worked exhibit you can redo with a pencil
Northgate Veterinary runs fourteen small-animal clinics. Revenue is up 16% in two years and operating margin has fallen from 18% to 12%. The owner wants to know why before she signs the lease for clinic fifteen. Twelve minutes in, the interviewer slides this across.
| Service line | 2023 revenue ($M) | 2025 revenue ($M) | 2023 visits (000s) | 2025 visits (000s) |
|---|---|---|---|---|
| Wellness and vaccination | 24.0 | 27.6 | 300 | 368 |
| Surgery | 18.0 | 19.8 | 30 | 30 |
| Dental | 6.0 | 9.0 | 40 | 72 |
| Pharmacy and retail | 12.0 | 13.2 | n/a | n/a |
| Total | 60.0 | 69.6 | 370 | 470 |
A weak read: "Wellness grew from 24 to 27.6, surgery from 18 to 19.8, dental from 6 to 9, pharmacy a bit. Everything is growing, so the problem must be costs." Every number in that answer is printed. The conclusion is wrong, and it sends the next ten minutes toward costs.
A strong read computes three things, in order.
First, volume against revenue. Visits went from 370,000 to 470,000, up 27%. Revenue rose 16%. Strip out pharmacy, which has no visits, and clinical revenue per visit fell from about $130 to $120, down 7.5%. That gap is the whole case.
Second, revenue per visit by line. Wellness fell from $80 to $75, down 6%. Dental fell from $150 to $125, down 17%. Surgery rose from $600 to $660, up 10%, on flat volume. The two lines that grew both earn less per visit; the one that raised prices added no visits.
Third, a sanity check from the prompt. An 18% margin on $60 million means cost of $49.2 million. A 12% margin on $69.6 million means about $61.2 million, up 24%, almost in step with visits at 27%. Cost per visit held. Revenue per visit did not.
Said aloud, that is four numbers in forty seconds, none of them printed. Two things the strong read skipped: a per-visit figure for pharmacy, which has no visits column, and reading all sixteen cells.
What the scoring sheet actually rewards
No firm I can find publishes an exhibit rubric. McKinsey says the problem-solving interview presents a business case "to evaluate your analytical thinking and approach to solving complex problems". BCG's preparation page lists "analyze data, perform quick calculations, and identify the most important factors". Bain says its case interviews "reward precision, but also creativity". Anything more specific, including the table below, is an interviewer's heuristic; this one is mine, from scoring cases at a Big Four strategy practice.
| What the candidate says about the exhibit | Ned's mark, out of 5 |
|---|---|
| Reads printed values, in the order they appear | 1 |
| Picks the comparison that matters, but still in printed values | 2 |
| Computes one derived number correctly | 3 |
| Derived number plus what it means for the client's question | 4 |
| Derived number, implication, and either the next analysis or the caveat (denominator, period, nominal dollars) | 5 |
A derived number with a shaky implication still outscores a flawless description, because I can fix an implication with one follow-up and I cannot fix a candidate who has not noticed that division is available. The jump from 3 to 5 is not more math; it is one sentence about the client and one about what to look at next.
Ned's rule. If every number you have said so far is printed on the exhibit, you have not started. The first number out of your mouth after the pause should be one I cannot find on the page.
Three traps that survive a correct read
Nominal growth across an inflationary period. A client whose revenue rose from $400 million in 2019 to $480 million in 2024 grew 20%. The US CPI-U annual average rose from 255.657 to 313.689 over the same years, 22.7%, per the BLS historical CPI-U table. In 2019 dollars, 2024 revenue is about $391 million: a 2% real decline. Any exhibit that spans 2021 to 2023 deserves this check, and almost nobody runs it.
The denominator that moved. "Share of respondents who completed the survey." "Revenue per store" when the store count changed. Ask of every percentage: percent of what, and did the "what" change between the periods? The answer is usually in a six-point footnote, which is where case writers put the things they hope you will not read.
Index charts. When every series starts at 100, the fastest-growing series looks biggest even if it is the smallest business: a $2 million line that doubled towers over a $200 million line that grew 5%. Ask for base-year values before you say which line matters.
Saying it in forty seconds
Take the pause. Twenty or thirty seconds of silence to write down two printed numbers and one derived one is fine; say "give me a moment" and take it. Then four beats.
- The derived number, first. "Revenue per visit is down about 7.5%."
- The ingredients, in one clause. "Visits are up 27% against revenue up 16%."
- The implication for the case question. "So the margin problem is price or mix, not cost."
- The next step or the caveat. "I would want the discount policy by line, and I am assuming nominal dollars."
Then stop. Candidates who keep talking after beat four are usually filling silence with printed numbers, which walks the mark back down.
In an interviewer-led case the exhibit arrives with a question attached, which tells you which derived number is wanted. In a candidate-led case it arrives because you asked for something, so read it against the question you asked. McKinsey Solve's Redrock study and BCG's Casey case run on the same principle with a clock attached.
Practice this today
Take five exhibits from anywhere: a segment table in an annual report, a case book, or the CoachNed chart drills. Before you speak, write three lines: the two printed numbers you will use, the operation, and the resulting sentence in the client's language. Then say it aloud in under forty seconds. If your first spoken number was printed on the page, that rep does not count.
The chart drills score exactly this: whether your first sentence is derived or read, and whether it connects to the case question. For exhibits that arrive mid-case, the live voice case with Ned runs about eighteen minutes with a seven-score debrief; the first rep is three typed turns, no account needed. Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month.
CoachNed is independent and has no affiliation with McKinsey, BCG, Bain, or any other firm named here.
Frequently asked questions
How long should I look at a chart before speaking in a case interview?
Twenty to thirty seconds of silence is normal. Use it to write down two printed numbers and one derived number. If you need longer, say so; a stated pause reads as control, an unexplained one as panic.
Do I need exact numbers from a case exhibit, or is an estimate fine?
Estimate to the precision the decision needs and say so. "Roughly 8% less per visit" is fine when the point is direction; if the interviewer wants the exact figure, they will ask.
What should I do if I do not understand an exhibit in a case interview?
Ask about the one thing blocking you: the unit, the denominator, or the base year. Interviewers penalize guessing far more than a precise clarifying question.
Are case interview exhibits different at McKinsey, BCG, and Bain?
The delivery differs more than the exhibit: interviewer-led cases attach a question to it, candidate-led cases hand it over because you asked. None of the three publishes a chart-reading rubric, so claims about a firm's "favorite chart type" are candidate folklore.
Sources
- Talbot Trucks | Careers | McKinsey & Company — exhibit question wording and the follow-up asking for a computed eTruck price. Checked 2026-09-24.
- Electro-Light | Careers | McKinsey & Company — $2 price, $1.90 unit cost, $40 million fixed cost. Checked 2026-09-24.
- Interviewing at McKinsey — what the problem-solving interview evaluates. Checked 2026-09-24.
- Case Interview Preparation | BCG Careers — "analyze data, perform quick calculations" wording. Checked 2026-09-24.
- Interviewing | Bain & Company — "reward precision, but also creativity" wording. Checked 2026-09-24.
- FashionCo. | Bain & Company — "good answer" versus "better answer" on a computed comparison. Checked 2026-09-24.
- Cleveland and McGill, "Graphical Perception," Journal of the American Statistical Association 79(387), 1984 — accuracy of position, length, and angle judgments. Checked 2026-09-24.
- BLS, Historical CPI-U, December 2024 (PDF) — CPI-U annual averages for 2019 and 2024. Checked 2026-09-24.