6. Lighthouse

Level 2 · Does marketing spend drive revenue?

From Jordan Reyes, Head of Marketing

“Over 36 months our marketing spend and our revenue correlate at 0.92. I'd like to tell the board that spend drives revenue and ask for a bigger budget. Is that a fair thing to say?”

Your goal

Choose the most accurate reply to Jordan.

Average month, indexed: January–October = 100
Average month, indexed: January–October = 100050100150200Jan–Oct, Marketing spend: 100Jan–Oct, Revenue: 100Jan–OctNov–Dec, Marketing spend: 171Nov–Dec, Revenue: 197Nov–DecMarketing spendRevenue
Average month, indexed: January–October = 100050100150200Jan–Oct, Marketing spend: 100Jan–Oct, Revenue: 100Jan–OctNov–Dec, Marketing spend: 171Nov–Dec, Revenue: 197Nov–DecMarketing spendRevenue

Each measure's normal month (January–October) is 100. In November and December spend averages 171 and revenue 197: both jump together, every year.

Show the numbers
Marketing spendRevenue
Jan–Oct100100
Nov–Dec171197
Revenue and marketing spend both rise sharply every November and December, and both grow every year (spend from $382k in 2023 to $511k in 2025, revenue from $1.60M to $2.30M). Which reply is accurate?

Pick one answer, then check it.

The lesson

Correlation is not causation, concretely

Everyone has heard the phrase. The useful part is knowing why two columns can move together without one causing the other, because then you can name the reason in a meeting. There are three common ones:

  1. A confounder: a third thing drives both columns.
  2. Reverse causation: the arrow points the other way.
  3. Chance: with enough columns, some pairs correlate by luck.

The confounder in Harbor & Pine’s numbers

Put monthly marketing spend next to monthly revenue from completed orders, 2023 to 2025:

monthly = pd.DataFrame({"spend": spend_by_month, "revenue": revenue_by_month})
monthly["spend"].corr(monthly["revenue"])   # 0.92

Now split the months:

spend per month revenue per month
January–October $33,153 $141,041
November–December $56,689 $277,176

Both jump in the holiday season. The holiday rush brings shoppers whatever the budget, and the marketing team knows it, so it plans its biggest spend for exactly those weeks. On top of that both columns grow each year. The season and growth are confounders: they raise spend and revenue at the same time, so the two correlate strongly even if spend added nothing.

Reverse causation hides here too

Budgets are often set as a share of expected sales. When revenue is expected to be high, spend goes up. Part of the arrow runs from revenue to spend, not the other way.

What you can claim, and what would settle it

You can say: “Months with higher spend had higher revenue.” You can’t say how much revenue the spend caused. To find out, change spend on purpose and keep the calendar the same for both groups: pause or cut it in randomly chosen regions or weeks (a holdout or geo test) and compare. That is an experiment, which chapter 3 taught you to read.

Common mistakes

  • “0.92 means 92% caused by.” r is not a percentage or a share of anything.
  • “More months make it causal.” More data sharpens the correlation; the confounder stays in.
  • Removing the season and declaring victory. Outside November and December the correlation is still 0.84, because both columns also grow every year. One confounder removed is not all of them.

Cookie settings

Pick what may run on this site. You can change this any time from "Cookie settings" at the bottom of every page.