How to answer a case
Answer any marketing case in five steps: restate, find the deciding number, estimate it, decide, and say what would change your mind; then score it.
On this page
Every exercise in this course hands you a situation with real numbers and asks for one decision or one estimate. The answers that score are not the ones with the most frameworks in them. They find the single number the decision turns on, estimate it with the arithmetic visible, commit to a call, and name the result that would reverse it. That is five steps, the same every time, and the rubric on every exercise is written against them.
The five steps
Restate. One sentence, in your own words, naming which of the five decisions this is and how much money rides on it. A good share of weak answers answer a question that was not asked, and the restatement is where that shows.
Find the deciding number. The one figure that, if you knew it, would make the decision obvious. It is rarely printed in the situation. Usually it is a threshold derived from the situation: the customers a sponsorship must produce, the win rate a positioning must reach, the payback month a channel must beat.
Estimate it. With the arithmetic shown and each input marked as given or assumed, using the course's shared constants where the situation is silent. Round to one or two significant figures; a guessed rate does not deserve four.
Decide. One sentence that could be pasted into a message to the person who asked. "It depends" is not a decision; "not at this price, and here is the price at which yes" is.
Say what would change your mind. A figure and a direction, and ideally the cheapest test that would produce the figure. This step turns an opinion into a position someone can check, and it is the one most often left out.
The shape is not special to marketing. The DSA course's solving loop restates a problem before writing code, and a system-design interview's four steps estimate before they draw a box.
Worked: an $8,000 podcast slot
A podcast with 40,000 downloads an episode offers a host-read sponsorship at $8,000 an episode, with a minimum of three episodes. The product is the course's default customer: $40 of margin a month, 5% monthly churn, $800 of expected margin, and a 12-month payback target, which puts the CAC ceiling at $368.
Restate. Is a host-read slot a cheaper way to win customers than what we do now? A channel decision, with $24,000 at stake at the minimum.
The deciding number. Paying customers per episode. At a $368 ceiling, $8,000 has to buy 8,000 ÷ 368 ≈ 22 of them.
Estimate. Nothing about this audience is given, so assume and say so: 1% of listeners visit (400), 3% of visitors sign up (12), 20% of those pay (2.4). About 2.4 customers an episode, roughly $3,300 each, nine times the ceiling. For the slot to work, this audience has to convert at nine times cold-traffic rates, which happens only when the show's listeners are very nearly the segment.
Decide. Not three episodes. Offer to buy one at the same rate as a test, with a unique link and code; if the seller will not sell one, decline.
What would change your mind. The test episode producing 22 or more paying customers within 60 days. A code count alone undercounts, because listeners type the address and forget the code, so count the codes and also the rise in direct sign-ups in the week after the episode against the four weeks before.
That answer is under 200 words, and every bullet of a rubric can be checked against it. Note what it spends: $8,000 to learn whether a $24,000 commitment is worth making. When an estimate rests on guessed rates, the cheapest test that settles the deciding number is usually the best decision on offer.
How an exercise page is scored
Every exercise has the same five sections: the situation, the question, a strong answer, a rubric, and where answers go wrong; a code exercise adds a sixth, the tested solution. Write your answer before you open the strong one, in prose, with the arithmetic shown, in the five steps. Reading the strong answer first feels like learning and is not: you recognise the reasoning next month and still cannot produce it.
The question is one of five kinds. A case asks what to do and what evidence would change it. An estimate asks for a number from stated assumptions. A design asks for a mechanism, such as a test plan or an agent's permissions, and its defence. A critique hands you a decision or a report and asks you to find its defect and say what to do instead. A code exercise asks for one function, and its tests, not a reviewer, decide whether the number is right. The course has three; payback month from a cohort is the first, and it needs enough Python to read a 20-line function.
The rubric is four to six bullets, each something a reviewer can see on the page. For the podcast it would read:
- restates the question as a channel decision and names the $24,000 at stake
- derives the customers-per-episode threshold from a stated CAC ceiling
- estimates customers per episode with each rate labelled as assumed
- decides, and proposes a test smaller than the commitment
- names the test result that reverses the decision, and how it is counted
Score each bullet 0 or 1, no halves. "Understands CAC" is never a bullet, because nobody can see understanding; they can see 22. Scored honestly, the total says which step to practise next.
Where answers go wrong
The weak answers quoted at the end of each exercise are rarely wrong on facts. They lose for one of four reasons.
- A framework instead of a number. "Run it through AARRR" names a tool and decides nothing.
- The seller's number as the deciding number. "$8,000 for 40,000 downloads is a $200 CPM against $20 for paid social, so no." The call happens to be right for the wrong reason. A CPM prices reach: an audience that converts at ten times the cold rate would buy 24 customers an episode at about $333 each, under the $368 ceiling, so $200 can be a fair price. The answer compared a price with a price instead of a price with a value.
- No reversal condition. A decision with nothing that would change it means the next data point re-opens the argument instead of the number.
- False precision. "$3,333.33 per customer" from three guessed rates. Precision is a claim about how much you know; round to what the inputs support.
In practice
To the seller: "We'll buy one episode at the per-episode rate, with a unique link, and decide on the next two from what it returns." A seller confident in the audience usually agrees, and one who will not sell a single slot has told you something about the expected result. To the team, put the five steps at the top of any budget request as five lines. Anyone can disagree with a line, and the disagreement is then specific.
The same loop, run against a 15-minute clock, is how a marketing case is answered in a hiring conversation; the plan and the interview covers that at the end of the course.
The mistake that costs money is committing the whole budget before writing down the number that would make you stop. The podcast minimum is $24,000 and the test that settles it is $8,000. Without the threshold of 22 customers written down first, the test result gets read to fit whatever the team already wanted.
Check yourself
1. The podcast drops its minimum to one episode but raises the price to $10,000. Do you buy it?
The threshold rises to 10,000 ÷ 368 ≈ 27 customers an episode, against an estimate of 2.4 at cold-traffic rates, so the slot needs more than ten times the cold rate. But the most you can lose falls from $24,000 to $10,000, and the one episode is now the test itself. Buy it only if there is a specific reason to expect that conversion, such as listeners who are very nearly your segment, and write the 27 down before it airs.
2. A founder asks whether a $5,000 conference booth is worth it. 2,000 people attend, you guess 10% are in your segment, and two of you will staff it for two days. What does the booth have to produce?
The cost is not $5,000. Two people × two days × 8 hours × $100 is $3,200 of time, so $8,200 in all, and at a $368 ceiling it must produce 8,200 ÷ 368 ≈ 22 paying customers. That is 11% of the 200 attendees in your segment, far above what any cold channel converts. Decline it for a $50-a-month product, unless the booth's real purpose is something else, such as recruiting or partners, in which case say so and price that instead.
3. A competitor has announced an AI assistant, and your CEO wants to announce one next week. Restate the question and name the number that settles it.
It is a what-we-claim decision, and the number is the win rate against that competitor since the announcement. If deals lost to them have not risen, the announcement answers nothing a buyer asked, and its cost (the build, and a claim the product then has to prove) buys no change in the number. If the loss rate has risen, ask the lost buyers what they cited before writing a word of the claim.