Funnels and unit economics · The funnel

Which funnel stage to fix

Price three funnel proposals in paying customers a month, cost per added customer and time to read, and decide what one quarter's budget funds.

estimatemediumExercise · 3 min · 17 of 60

On this page

Three proposals each promise more customers from the same funnel. The one at the stage with the worst rate is the obvious choice, and it is the one that costs the most per customer.

The situation

The invoicing app's paid funnel each month: $50,000 of spend brings 100,000 visits ($0.50 a visit); 4% sign up (4,000), 30% of those activate (1,200), 40% of the activated are retained at month 3 (480), and 20% of the retained pay (96). A paying customer costs $521 to buy through paid, so value each added customer at that. Three proposals compete for this quarter's $30,000 and eight engineer-weeks:

A, more traffic: $10,000 a month more paid spend, bringing 16,000 extra visits from a broader audience that activates at 27%, not 30%.
B, onboarding rebuild: $24,000 and six engineer-weeks, expected to lift activation from 30% to 33%.
C, upgrade prompt: $6,000 and two engineer-weeks, expected to lift paid conversion from 20% to 22%.

Treat a lift as readable once the monthly change in the count it moves is twice that count's spread, √(n × p × (1 − p)) for n users at rate p.

The question

Estimate the paying customers each proposal adds a month, what each added customer costs, and how soon B and C can be read; then decide what this quarter's $30,000 funds.

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

Multiply each through the funnel. A: 16,000 × 0.04 × 0.27 × 0.40 × 0.20 = 13.8 paying customers a month for $10,000 a month, $723 each, every month. That is above the $521 it would replace, and the weaker activation cost it a tenth of its customers. The quarter's $30,000 buys three months of A and nothing after.

B: a 10% lift at activation is a 10% lift in paying customers, 96 × 0.1 = 9.6 a month, worth 9.6 × $521 = $5,000 a month of paid spend, so $24,000 repays itself in about five months and keeps paying: over a year, 115 customers at about $210 each. It reads within a month: 120 more activations against a spread of √(4,000 × 0.3 × 0.7) ≈ 29.

C: also 10%, also 9.6 a month, for $6,000, about $52 a customer over a year. But it moves 96 upgrades by about 10 against a spread of √(480 × 0.2 × 0.8) ≈ 9, so one month shows nothing. The lift grows with months and the spread with their square root, so reaching twice the spread takes three to four months.

Fund B and C, which use exactly $30,000 and eight engineer-weeks. They act on different stages, so they multiply: 1.1 × 1.1 = 1.21, about 20 more paying customers a month for a one-off $30,000, against A's 13.8 for $30,000 a quarter. Ship B first: it reads in a month, and its extra activations enlarge the sample C's test needs. Reject A; it buys customers above today's paid CAC.

What would change this: B's first month showing fewer than about 60 extra activations, twice the spread. Then the lift is not there, and the engineer moves on.

Rubric

Multiplies each proposal through every later stage, not only the stage it touches
Applies A's lower activation rate to A's extra visits
Separates A's recurring cost from B's and C's one-off cost, and prices an added customer for each
Uses the spread to say how soon B and C can be read, and that C needs months
Decides within $30,000 and eight engineer-weeks, and notes that B and C multiply
Names the first-month result that would stop B

Where answers go wrong

"Fix sign-up: 4% is the worst rate in the funnel." The lowest rate is not the cheapest one to move. The only proposal at the top is A, at $723 a customer, recurring, and it dilutes the stage after it. This is the mistake the funnel and AARRR warns against.

"C gives the same 9.6 customers for a quarter of B's cost: ship C alone and report next month." C is the cheapest, but next month's reading is about one spread, which is noise. The team will either claim a win it cannot show or kill a lift that is real.

"A adds the most, 13.8 a month against 9.6." The most per month, at $723 each and only while the spend lasts. B and C together add about 20 a month for a one-off sum.