Duolingo

Case Discussion

Making critical product decisions when your company is on a streak.

Presented by:
Larry Vincent,
Professor of the Practice of Marketing

Presented to:
GSBA 509
October 1, 2026

Usage

• Overall
• A
• B
• C

New usage

• Overall
• A
• B
• C

Growth focus

• Overall
• A
• B
• C

Five factors

• Overall
• A
• B
• C

Design thinking?

• Overall
• A
• B
• C

Super App?

• Overall
• A
• B
• C

How do we apply what we’ve learned this week to the product decisions Duolingo must make?

How would design thinking play a role in this?

Design thinking process

What about the five-factor model?

Five factors model

Would conjoint analysis help?

Can we use breakeven analysis?

Breakeven

Breakeven without units

If most users pay nothing and software has almost no marginal production cost, what counts as a unit?

For a retention feature, the economic unit can be a subscriber-month.

Let students wrestle with the prompt first. The goal is to move them away from downloads, lessons, or daily active users. Those can describe activity, but they do not directly pay back the engineering investment. For the assumptions in this exercise, a retained month from a paying subscriber contributes $6.

Baseline subscriber economics

Input	Assumption
Fixed engineering cost	$5,000,000
Monthly contribution	$6
Monthly churn	5%
Expected lifetime	20 months

Expected lifetime uses the steady-state approximation \(1/c\).

\[ L = \frac{1}{c} = \frac{1}{0.05} = 20 \]

\[ CLV = m \times L \]

\[ CLV = \$6 \times 20 = \$120 \]

Use the same undiscounted CLV convention students already learned. CAC does not enter this feature decision because acquisition spending on the installed base is sunk. The relevant cash flow is incremental contribution after launch.

Breakeven in subscriber-months

\[ \text{Breakeven subscriber-months} = \frac{\$5{,}000{,}000}{\$6} = \mathbf{833{,}333\ \text{subscriber-months}} \]

What is a subscriber-month?

• 833,333 subscribers stay one extra month
• 83,333 stay ten extra months
• Or, any equivalent combination

This is the cleanest breakeven result because it makes the fewest behavioral assumptions. It also gives students the closest analogue to units in a conventional breakeven calculation.

What does “one customer saved” mean?

One-month delay

The feature postpones cancellation for exactly one month.

\[ \text{Value} = \mathbf{\$6} \]

Restored expected life

The customer remains and again faces the normal 5% monthly churn rate.

\[ \text{Value} = \mathbf{\$120} \]

Do not let students casually call every delayed cancellation a $120 save. The stronger interpretation requires evidence that the intervention returns the customer to the ordinary retention process. If that claim holds, $5 million / $120 = 41,667 retained subscribers. If the feature merely buys one month, the answer remains 833,333 subscriber-months.

Breakeven in retained subscribers

If a prevented cancellation restores the full expected remaining life:

\[ \text{Breakeven subscribers} = \frac{\$5{,}000{,}000}{\$120} = \mathbf{41{,}667} \]

This result is a shorthand for 41,667 customers × 20 months × $6.

The arithmetic is simple; the causal claim is not. Ask what evidence would justify treating an observed retention event as a restored expected life rather than a short postponement.

The same breakeven as a churn target

Duolingo reported 8 million paid subscribers at the end of Q2 2024.

If the feature affects all 8 million:

\[ \frac{833{,}333\ \text{months}}{8{,}000{,}000\ \text{subscribers}} = 0.104\ \text{month per subscriber} \]

\[ 20.000 + 0.104 = 20.104\ \text{months} \]

\[ c_{new}=\frac{1}{20.104}=\mathbf{4.974\%} \]

Breakeven requires monthly churn to fall from 5.000% to about 4.974%.

That is a reduction of 0.026 percentage points, or roughly 0.52% relative to the baseline churn rate. The result looks small because the fixed cost spreads across a very large installed base. This calculation assumes the benefit applies to every paid subscriber and ignores timing, discounting, taxes, and ongoing feature costs.

How many subscribers will the feature affect?

The feature must create 833,333 additional subscriber-months.

Subscribers affected	Added life needed per subscriber
800,000	1.042 months
2,000,000	0.417 months
4,000,000	0.208 months
8,000,000	0.104 months

\[ \text{Added life needed} = \frac{833{,}333\text{ subscriber-months}} {\text{subscribers affected}} \]

The feature may reach only a subset because of platform, language, geography, eligibility, or adoption. “Affected” should mean users whose behavior could plausibly change, not everyone who technically had access to the feature.

The product decision

What would Duolingo need to believe before committing $5 million?

Evidence

• The feature causes a measurable reduction in paid churn
• The effect persists beyond a novelty period
• Enough subscribers encounter and use it

Economics

• Incremental contribution exceeds the build cost
• Ongoing maintenance and support are included
• Effects on conversion and free-user ad revenue are counted separately

Real subscription companies evaluate product investments this way, usually through cohort retention curves, survival or hazard models, controlled experiments, and incremental NPV rather than the simple 1/churn shortcut alone. End by asking students to design the experiment: treatment, control, primary outcome, observation window, and the minimum detectable effect.