The plan and the interview
Answer a marketing case in 15 minutes with the arithmetic shown, run a 30-60-90 on an AI-first team, and hand a hiring manager a plan they can score.
On this page
A hiring manager for a marketing lead is looking for this course's habits under a clock: a case answered in 15 minutes with the arithmetic on the page, a straight answer to "how would you use AI here", and a plan they could hand to their own board. This is the one lesson in the course written for the interview. It uses the running company, because the case you are given will look like it.
A case in 15 minutes
How to answer a case gave the five steps. In an interview each gets a slice of the clock, and the slices are uneven on purpose: most of the time goes to the estimate, and the last two steps are never skipped.
| Step | Minutes | On the page at the end of it |
|---|---|---|
| Restate | 2 | which of the five decisions, and the money riding on it |
| Find the deciding number | 2 | the threshold, derived from the situation |
| Estimate it | 6 | the arithmetic, each input marked given or assumed |
| Decide | 3 | one sentence you could send to the founder, and the test |
| Say what would change it | 2 | a figure and a direction |
The same discipline runs the system-design course's four-step framework: a budget per step, so the step being scored is not squeezed out by the one before it.
Worked: "should we double paid?"
The case: "We're at $3M ARR and have been flat for two quarters. The founder wants to double paid media, from $50,000 to $100,000 a month. Do it?"
Restate, minutes 0 to 2. "A where-we-say-it decision with $600,000 a year riding on it. Before I answer it, I want to know why we're flat, because doubling spend only helps if acquisition is the problem."
Deciding number, minutes 2 to 4. The cost of a customer bought by the second $50,000, against two lines: the $800 of margin a customer returns over a lifetime, and the 12-month payback ceiling of $368 from the course's constants.
Estimate, minutes 4 to 10.
why flat 5,000 customers x 5% monthly churn = 250 leave a month
96 paid + 154 organic = 250 arrive a month
first $50k 96 customers at $521 pays back in month 21, above the $368 ceiling
second $50k assume 60 customers $833 each, more than the $800 one returns
churn one point at 5,000 customers = 50 customers a month
The first $50,000 already misses the payback ceiling: $521 pays back in month 21, as CAC, LTV and payback derives. The second reaches many of the same people again, so it buys fewer customers at a higher price, the saturation paid media basics prices. At 60 customers it loses money on every one.
Decide, minutes 10 to 13. "No. Hold paid at $50,000 and run a four-week geo holdout on it, because $521 is what attribution credits paid with, not what paid causes. Put the next dollar into retention: at 5,000 customers one point of monthly churn is worth 50 customers a month, about what the second $50,000 would buy, and those customers are already paid for."
What would change it, minutes 13 to 15. "If the holdout shows paid's incremental CAC under $368, and extra spend in a fifth of regions buys customers below it, I'd raise paid in steps of $10,000." The holdout design is holdouts, geo tests and power.
That answer fits in 15 minutes because nothing in it is decoration. Every line is either a number or a decision.
What a hiring manager scores
A good interviewer scores what they can see on the page, and it is the rubric every exercise in this course has used:
- restates the question in one sentence and names the money at stake
- derives the threshold before estimating anything
- shows the arithmetic, rounded, with each input labelled given or assumed
- decides, in a sentence that could be pasted into a message, and proposes a test smaller than the commitment
- names the result that would reverse the decision
Then the AI question, which nearly every loop now asks: "How would you use AI on this team?" The answer that scores is a ledger, not a tool list. "Of the 330 hours of recurring work on a team like yours, about 250 sit in tasks a prompt or a workflow now does. Counted with review and upkeep, the team's hours fall by 40%, not 76%. I'd put the freed hours into customer interviews and lifecycle programs, review every paid ad before it runs, sample the rest, and make my first hire the person who owns the workflows." That is what got automated and what did not in four sentences. If the follow-up is about a generated campaign, answer with the disclosure line and the section it satisfies.
Answers lose points in four recognisable ways: a tour of frameworks with no number, a list of tools, "it depends" with nothing it depends on, and "we'd use AI to produce ten times the content", which prices the drafting and forgets the reading.
The 30-60-90 on an AI-first team
The second thing a hiring manager asks for is the first 90 days. On an AI-first team, where most drafts are already generated, the new lead's value is not more output. It is the standard, the measurement and the plan.
Days 1 to 30: rebuild the numbers. The funnel, the cohort table, and paid and blended CAC, from raw data rather than from the dashboard, because the dashboard is where a flattering number hides; blended versus paid CAC is the usual one. Ten customer interviews. The task ledger. The deliverable at day 30 is one page with the deciding number on it: "we add 250 customers a month and lose 250".
Days 31 to 60: change what is cheap and reversible. The review standard live, with a named steward and every paid ad reviewed. A four-week geo holdout on paid. One workflow shipped with an owner, usually the weekly report, because it returns hours every week and fails visibly. Disclosure lines added to the review checklist.
Days 61 to 90: the 12-month plan. Built on the holdout result and the diagnosis, and written as the next section describes.
The plan you hand over
A 12-month plan a board or a hiring manager can score fits on one page and contains seven things: the deciding number and the target; four quarters, each with one number it moves; the budget by line; the hire, and why that one first; what you will not do, with the number that rules it out; how each result will be measured; and, for each part, the result that would change it. A plan without the last item cannot be corrected; one without numbers cannot even be checked.
The course's last exercise, the 12-month plan for a $3M ARR company, builds that page for this running company, drawing on every module. Practise the 15-minute case aloud against a timer at least twice; running your own mock sets up the practice, and the AI PM course's estimation drills the arithmetic.
In practice
In the room, say the step you are on: "Restating: this is a channel decision with $600,000 a year on it." It lets the interviewer follow and score, and it stops you drifting. When the time is nearly up, skip polishing the estimate and give the decision and the reversal; they are worth more points than a third significant figure. Once hired, the same 30-60-90 is what you show the CEO on day one, so the first conversation is about numbers you will bring back, not about taste.
The mistake that costs money, and in an interview the offer, is a plan of activities with no numbers. "Launch a content engine, build AI workflows, refresh the brand" gives a board nothing to check at month three, so the first quarter's budget is spent before anyone can say whether it worked.
Check yourself
1. At minute 9 the interviewer says time is nearly up. You have the estimate but no decision and no reversal condition. What do you say in the two minutes left?
The decision and the reversal, in a sentence each: "No to doubling; hold at $50,000 and run a geo holdout. I'd change my mind if paid's incremental CAC comes in under $368." Leave the estimate as it stands. Two of the five points in the rubric are those two sentences, and no amount of extra precision earns them back.
2. On day 20 the CEO asks you to approve 500 generated posts a month, before your diagnosis is done. Decide.
Not yet, with the price: 500 posts at under $1 plus a $50 review is about $25,500 a month and 250 hours of reading, about 1.4 of the team's three people at 176 hours each. Offer a decision at day 90 inside the plan, with a one-quarter test of a few dozen reviewed posts against the written ones. The founder wants 500 posts a month works the full critique.
3. Your 30-60-90 holds paid out of 20% of regions for four weeks. Estimate the most the holdout costs in customers, and decide whether to run it.
20% of $50,000 is $10,000 of spend not made for the month. If paid were fully incremental at $521, that is about 19 customers forgone, roughly $15,400 of lifetime margin at $800 each, against $10,000 saved: a net cost of at most about $5,400, and a gain if paid causes less than it is credited with. To settle a $600,000-a-year decision, run it.