Growth Marketer interview questions

AARRR describes five stages of a customer journey that growth teams measure.

• Acquisition shows how people first reach the product.
• Activation measures whether they experience the product's initial value.
• Retention, referral, and revenue track whether they return, invite others, and generate money.

A North Star metric is one shared measure of the recurring value customers receive from a product.

• It should move when more users reach the product's core value, such as weekly completed rides for a transport app.
• It is usually more useful than raw registrations or page views because those do not prove value was delivered.
• Supporting input metrics explain which behaviors, such as first-week usage, are driving the North Star.

A funnel moves users through stages, while a growth loop feeds an output back in as a new input.

• A funnel makes conversion and drop-off between acquisition, activation, and later stages easy to inspect.
• In a referral loop, an activated user invites someone who can become another activated user.
• Loops can compound, but they still need measurable conversion rates and cycle time to show whether they work.

Acquisition is the stage where a relevant person first arrives through a measurable source.

• Sources can include paid search, social ads, organic search, partnerships, or referrals.
• Reach and clicks describe channel delivery, while qualified visits or signups show movement into the product funnel.
• UTM parameters and ad-platform identifiers help connect each visit to its source and campaign.

Activation means a new user has reached an early moment that demonstrates the product's value.

• The activation event must reflect the product, such as publishing a first form rather than merely creating an account.
• Activation rate is activated new users divided by eligible new users in the same defined window.
• Time to value measures how long users take to reach that event and reveals friction in onboarding.

Retention measures whether a defined group of users returns to a meaningful product behavior over time.

• The return event should represent real use, such as completing a workout, not simply opening an email.
• Day 7 or week 4 retention uses the original cohort as the denominator and checks activity in that later period.
• Retention should be compared across cohorts because one blended percentage can hide improvement or decline.

Referral measures whether existing users bring new people into the product.

• A referral flow has an invitation action, a recipient response, and a target conversion such as signup or purchase.
• Referral rate can mean inviters divided by eligible users, so its exact definition must be stated.
• Referred signups should be tracked separately from sent invitations because many invitations never convert.

Revenue is the stage where user activity produces measurable income for the business.

• Common measures include purchase conversion, average revenue per user, recurring revenue, and renewal rate.
• Gross revenue alone can mislead when refunds, discounts, or payment failures are material.
• Revenue should be connected back to acquisition source and retained customers rather than viewed as an isolated total.

Step conversion is users who complete the next step divided by users who completed the previous step.

• If 1,000 visitors produce 200 signups, visitor-to-signup conversion is 20 percent.
• Drop-off for that step is 80 percent because it is one minus the conversion rate.
• The same user definition, time window, and event order must be used in both numerator and denominator.

A good experiment hypothesis links one change to a measurable outcome for a defined audience and explains why.

• A clear form is: changing X for audience Y will move metric Z because of reason R.
• The expected direction and primary metric should be set before looking at results.
• The change must be specific enough that the control and treatment differ in one interpretable way.

A basic A/B test randomly assigns eligible users to a control or treatment and compares a predefined outcome.

• The control sees the current experience, while the treatment sees the proposed change.
• Random assignment makes the groups comparable on average, so the treatment is the main systematic difference.
• The test needs a planned sample, primary metric, and decision rule before it starts.
• Results should include the estimated lift and uncertainty, not only which variant had the larger number.

Randomization reduces systematic differences between test groups before the treatment is applied.

• It distributes factors such as device, geography, and user intent across variants by chance.
• Alternating users or assigning mobile users to one variant can introduce selection bias.
• Assignment should stay stable for the same user so one person does not switch between experiences.

Statistical significance means the observed result would be sufficiently unusual under the test's no-effect assumption at a chosen threshold.

• A common threshold is 0.05, but it must be chosen before reading the result.
• Significance does not prove the treatment caused a large or valuable improvement.
• Effect size, confidence interval, data quality, and guardrail metrics still matter for the decision.

A p-value is the probability of seeing a result at least this extreme if the null hypothesis and test assumptions are true.

• A p-value of 0.03 does not mean there is a 97 percent chance the treatment is better.
• A smaller p-value is stronger evidence against the null, not a measure of effect size.
• Its interpretation depends on valid randomization, the planned analysis, and no uncorrected repeated testing.

A confidence interval gives a range of effect estimates compatible with the data under the chosen statistical procedure.

• A 95 percent interval for lift might run from 1 to 5 percentage points rather than report only a 3-point estimate.
• A narrow interval gives more precision than a wide interval.
• If the interval includes zero, the data do not rule out no effect at the matching two-sided significance level.

Minimum detectable effect, or MDE, is the smallest true change a planned test is designed to detect reliably.

• A smaller MDE requires more observations because subtle changes are harder to distinguish from noise.
• It should be chosen from the smallest improvement worth acting on, not adjusted after results arrive.
• Baseline conversion, significance level, and statistical power all affect the required sample for that MDE.

Required sample size mainly depends on baseline variability, MDE, significance level, and desired statistical power.

• For a binary conversion metric, the baseline affects variance: at a fixed absolute MDE variance is largest near 50 percent, while at a fixed relative MDE a lower baseline usually needs more observations.
• Detecting a smaller effect or demanding higher power also increases the sample requirement.
• Sample size should be calculated before launch and translated into duration using eligible traffic, not total site traffic.

The primary metric decides whether the hypothesis succeeded, while guardrail metrics catch important harm elsewhere.

• A signup test might use completed signup rate as its single primary metric.
• Guardrails could include payment failures, unsubscribe rate, or page performance depending on the change.
• Both types and their decision thresholds should be defined before the test to prevent cherry-picking.

Repeatedly stopping a standard fixed-horizon test when the p-value looks favorable raises the false-positive rate.

• Random fluctuation can cross the significance threshold early and disappear as more data arrives.
• The safe basic approach is to wait for the planned sample and duration before making the decision.
• Sequential testing methods allow monitored decisions only when their stopping rules are built into the analysis.

An event is a timestamped record that a user or system performed a defined action.

• Examples include signup_completed, report_created, and subscription_started.
• Each event should represent one observable action rather than a vague state such as engaged_user.
• A clear event definition states when it fires, who triggers it, and whether retries can create duplicates.