Product Manager Interview: Design a Referral Program for a Grocery Delivery App

Prompt and Applicable Context

A fictional grocery delivery app has 250,000 monthly active purchasers and adds 50,000 first-time purchasers each month. The 30-day second-order rate for first-time purchasers is 30%, contribution margin per completed order before acquisition and referral rewards is $9, and paid channels cost an average of $27 per first-time purchaser. The team wants existing customers to invite people who have not ordered and acquire incremental retained customers without increasing fraud, spam, or fulfillment problems.

The app, user counts, conversion rate, amounts, time window, and channel data are all interview-case assumptions, not industry benchmarks. In this prompt, contribution margin means order revenue minus variable merchandise, picking, payment, and delivery costs, before acquisition expense and referral rewards. A real answer must first confirm the company's accounting definition, market, legal jurisdiction, and fulfillment capacity.

A public product-design prompt explicitly asks candidates to design a referral system for a grocery app. A real product help page also demonstrates common rules: an existing user shares a code, and a new user must complete a qualifying action before a reward is issued. Published research warns that a larger reward is not automatically better because reward visibility, scheme, and product characteristics can change referral intent. If an incentivized share is an endorsement, US FTC guidance also calls for a clear disclosure of the material connection. These sources support the prompt's relevance to user insight, growth experiments, unit economics, trust, and risk judgment. They do not prove that any named company consistently asks it.

The task is to prove that the program creates incremental new customers who retain. A three-step “share link → sign up → issue coupon” flow cannot establish that result on its own. If someone who would have purchased anyway merely uses a referral code, first orders rise without incrementality. If rewards pay at registration, self-referrals and account farms can consume the budget. If demand jumps while delivery capacity stays fixed, successful acquisition can still damage the customer experience.

What the Interviewer Evaluates

First, can the candidate translate a business goal into a falsifiable customer outcome? A strong answer does not treat shares, clicks, or registrations as final success. It selects a result such as incremental referred customers retained for 30 days, which captures both acquisition and quality.

Second, can the candidate identify credible, motivated referrers? Brand-new accounts have reach but no completed experience and high risk. Customers with several recent successful orders, demonstrated satisfaction, and service in a stable delivery area can make a more credible recommendation. Segmentation should combine potential, relevant moments, and risk, going beyond a list of the highest spenders.

Third, can the candidate close the entire loop? The answer should say when the entry point appears, how a user shares, who qualifies as new, who receives attribution, which action triggers each reward, how refunds and cancellations work, and when rewards expire. Ambiguity in any rule becomes either a customer dispute or an abuse path.

Fourth, can the candidate calculate incremental unit economics? The $27 paid acquisition cost is a comparison baseline, not a referral budget ceiling. Referred customers have different organic conversion, retention, redemption, support, and fraud costs. Only value above the control group's outcome is attributable to the program.

Fifth, can the candidate design a credible experiment? Rewards may cannibalize natural advocacy, social relationships can expose people across experiment arms, and shared delivery capacity can create interference. The answer should define the randomization unit, control, attribution rule, observation period, and when a conventional user-level A/B test is unsuitable.

Sixth, can the candidate turn trust, privacy, compliance, and fulfillment into product requirements? The app should not upload contacts by default or send messages on a user's behalf. Incentivized recommendations should disclose the relationship. Capacity must be checked by area and time before demand expands. Risk guardrails matter as much as growth metrics.

Questions to Clarify Before Answering

• What is the primary goal? Is it lower first-purchase acquisition cost, more 30-day retained customers, reactivation of existing customers, or entry into a new area? This answer assumes profitable new customers who place a second order within 30 days.
• Who counts as new and retained? Must a person have never registered, never ordered, or belong to a household that has never purchased? Here, a new customer is an independent customer with no prior completed order, and 30-day retention means completing a second order within 30 days of the first.
• How are $27 and $9 calculated? Does paid CAC divide spend only by first purchasers? Does $9 already deduct discounts, refunds, support, and fulfillment failures? Different definitions invalidate the comparison.
• What is the natural-referral baseline? How many new customers already cite a friend, share an ordinary link, or arrive as direct traffic? Without a baseline, a paid program can simply pay for existing word of mouth.
• Which users and markets may participate? Are there age, geography, product, payment, or marketing-permission restrictions? Which delivery cells are near their capacity limit?
• Which rewards are allowed? Account credit, free delivery, a product voucher, or cash? Can both sides receive value, and what are the minimum basket and expiry rules?
• How should multiple referrals be attributed? If one person receives several links, signs up on another device, or sees an ad before entering a code, does first touch, last touch, or the first valid code win?
• What are the dominant abuse modes? Self-referral, multiple household accounts, virtual payment instruments, refund-after-reward, public coupon distribution, or referrer spam? Which signals exist today?
• Is fulfillment capacity available? Which areas, days, and time slots can absorb incremental orders? If on-time performance is already falling, repair supply or limit the pilot to cells with headroom.
• Which legal jurisdiction applies? Incentive disclosure, marketing consent, privacy, tax, and promotion terms require market-specific legal review. Guidance from one country is not a global rule.

30-Second Answer Framework

“I would optimize for incremental new customers who complete a second order within 30 days per 1,000 eligible referrers, not total shares or first orders. I would start with customers who recently completed two successful orders, are satisfied, and live in delivery areas with spare capacity. The app would present a system share sheet at a natural high-satisfaction moment without uploading contacts. A new customer gets a small first-order credit; the referrer receives credit only after that customer completes a second order and passes the refund window. Each referrer has a cap, household, device, payment, and address signals help detect duplicates, and the share copy clearly states that the sender may earn a reward. I would compare no reward, a new-customer-only reward, and a two-sided reward using intention-to-treat analysis, incremental retention, and net contribution, with fraud, complaint, cancellation, and on-time-delivery guardrails. I would scale by area only if the conservative incremental net contribution is positive and every guardrail passes.”

This framework covers objective, users, loop, incentive, risk, disclosure, experiment, and decision. The deeper answer should place the prompt's numbers in an incremental economic model instead of using $27 to choose a coupon value directly.

Step-by-Step Deep Dive

Step 1: Define an incremental retained outcome

Build a two-sided funnel from referrer to new customer: eligible and exposed → initiates share → recipient opens → signs up → completes a qualifying first order → completes a second order within 30 days. Retain counts, elapsed time, and failure reasons at every step, but set the primary metric to:

incremental 30-day second-order customers per 1,000 eligible referrers assigned to the program

“Incremental” is the treatment-control difference. “Assigned” includes people who saw the experience but did not share, preventing selection bias from an analysis limited to active sharers. A second order separates one-time coupon use from an initial repeat habit, but it is still not lifetime value. A longer retention window remains necessary before full scale.

Step 2: Select referrers, recipients, and moments

For the first version, let a referrer qualify after at least two completed orders in the previous 60 days, no unresolved refund or risk flag, and residence in a delivery area with capacity. Two orders and 60 days are pilot assumptions intended to anchor recommendations in recent real experience. Data should change them later.

Show the entry point after successful fulfillment, an unsolicited high rating, or a repeat purchase, not during a complaint, delay, or refund flow. Generate a personal link, code, and editable copy, then hand them to the operating system's share sheet. Do not require a full contact upload or send a message automatically on the user's behalf.

The recipient must have no prior completed order. Email or phone number is only one signal; household, device, payment instrument, delivery address, and account history also inform eligibility. Similar signals should not automatically reject roommates or family members. Hold higher-risk rewards for review and provide an appeal path.

Step 3: Close an explainable referral loop

A first version could work as follows:

1. An eligible customer sees “Invite a friend; both of you may earn account credit” after a successful order.
2. Copy adjacent to the link says the sender may receive credit after the friend completes the conditions, so the recipient understands the incentive before opening.
3. The deep-link page shows reward amount, minimum basket, service area, expiry, and conditions for both sides before registration.
4. The first valid referral code binds to the new customer; support cannot arbitrarily reassign it to a more favorable referrer.
5. The new customer receives or redeems $5 in account credit on a first order that meets the minimum basket.
6. The referrer receives $5 in account credit only after the new customer completes a second order within 30 days and the refund and chargeback window has passed.
7. Credit is non-cash, non-transferable, and expires on a disclosed date. A referrer can receive at most five successful rewards per month.

The $5 amounts, 30 days, and five monthly rewards are all interview pilot assumptions, not recommended industry values. Delaying the referrer reward reduces instant gratification but aligns spend with retention and makes registration or one-order abuse less attractive. If delay materially suppresses legitimate sharing, test a split reward after the first and second orders against full payment after the second.

Step 4: Calculate economics from incrementality, not face value

Define:

• I: incremental new customers who complete a second order within 30 days;
• CM30: their 30-day contribution margin before referral rewards;
• RN: reward cost actually redeemed by new customers;
• RR: reward cost actually redeemed by referrers;
• O: incremental support, payment, risk, and operating cost;
• F: fraud, refund, and chargeback loss.

Incremental net contribution is:

I × CM30 - RN - RR - O - F

The prompt gives $9 contribution margin per order. A customer who completes two orders produces at least $18 of 30-day contribution margin before referral rewards. If both $5 credits are redeemed, reward cost is $10, leaving $8 for operations, abuse, and uncertainty. That calculation only shows that a pilot has room to measure; it does not prove profitability. Some orders would happen anyway, rewards may not all redeem, a customer may place more than two orders, and orders may be refunded. Recalculate every term with experimental incrementality and actual redemption.

The $27 paid CAC is useful as an opportunity-cost comparison, but it does not imply that a $26 referral reward is acceptable. User quality, attribution, scale, and marginal cost can differ by channel. Compare incremental cost per retained 30-day customer, expected payback, and longer-term net contribution under one definition.

Step 5: Make abuse, trust, and fulfillment preconditions

Prevention includes new-customer eligibility, per-referrer caps, a minimum basket, delayed rewards, non-cash credit, and clear terms. Detection connects referrer, recipient, household, device, payment, address, IP, orders, and refunds to find self-referral loops, account farms, dense address patterns, and reward-then-refund behavior. Enforcement should be evidence-tiered: pay low-risk cases normally, delay and review medium-risk cases, and cancel rewards and restrict eligibility for high-confidence abuse while retaining appeal records.

The recommendation must tell recipients that the sender may benefit. Research may show that hiding a reward changes referral intent, but a product must not bypass applicable disclosure duties to improve an experiment. Market counsel should approve the exact language and placement. Track reports, opt-outs, and bulk distribution. If a referrer posts a personal code on a public coupon site, terms may stop future rewards without retroactively taking a reasonable discount from a genuine new customer.

Open the growth pilot only in delivery cells with headroom. If on-time performance, substitution, or cancellation deteriorates, pause new exposure. A referral program has not succeeded if it merely converts acquisition expense into fulfillment incidents.

Step 6: Run an experiment that can detect cannibalization

Pre-randomize eligible referrers into three arms: no-reward control, $5 for the new customer only, and $5 for each side. All three arms see the same share surface at the same moment. Control users receive an ordinary trackable link with no incentive copy or reward, separating interface exposure from the incentive and estimating natural advocacy. Analyze intention to treat and retain non-sharers.

The observation period must cover the first order, the 30-day second order, and the refund window. Pre-commit to the first valid code for attribution and count each new customer once. If people in one household or close social network can invite across arms, randomize by household or an identifiable network cluster, or at least record cross-arm contamination and state how it biases the estimate. If delivery capacity is shared within an area, use geographic clusters or time switchbacks so one arm's orders do not crowd out another arm.

The primary comparison is incremental 30-day second-order customers among the three arms. Diagnostic metrics include exposure, share, open, registration, first-order, second-order rates, and time at each step. Economic metrics include actual reward per incremental second-order customer, net contribution, and payback. Guardrails include duplicate identity, reward holds, refunds, chargebacks, complaints, opt-outs, on-time delivery, substitution, cancellation, support tickets, and referrer retention.

Step 7: Predefine scale, iterate, and stop decisions

Scale requires four conditions together: incremental second orders exceed control; the conservative interval for incremental net contribution is positive; referred-customer quality is no worse than a comparable alternative channel; and fraud, spam, and fulfillment guardrails all pass. Expand gradually by area and capacity, retaining a long-term no-reward or lower-reward holdout to measure diminishing marginal impact.

High sharing with low first purchase points to landing-page, service-area, or first-order value problems. High first purchase with low second purchase suggests poor targeting or a weak first experience. Good conversion with negative net contribution calls for lower rewards, a later trigger, or a one-sided scheme. If incentivized users share less than control, test whether the reward damaged trust and preserve an organic path. If abuse or complaints cross a stop threshold, close the affected audience or channel before adding more complex rules.

High-Quality Sample Answer

“I would first confirm the goal. The prompt gives paid CAC, but I would not define success as first purchases or a cost below $27. My primary outcome is incremental new customers completing a second order within 30 days per 1,000 eligible referrers, paired with net contribution, fraud, complaints, and fulfillment quality.

The first version targets customers with two completed orders in the last 60 days, no unresolved risk, and delivery capacity in their area. I would show the entry point after successful fulfillment or an unsolicited high rating, then use the system share sheet to generate a personal link without uploading contacts. A new customer must have no prior completed order. Share copy and the landing page clearly explain both sides' conditions and that the referrer may receive credit.

I would test three randomized arms: no reward, $5 for the new customer only, and $5 for both sides. The new customer redeems the offer on a qualifying first order. The referrer waits until the friend completes a second order within 30 days and passes the refund window. Rewards are capped at five per month. Household, device, payment, and address signals identify possible self-referrals; a shared address alone does not prove fraud, so medium-risk rewards are reviewed.

Two completed orders produce at least $18 in contribution margin before referral rewards under the prompt. If both $5 credits redeem, $8 remains for operations and risk, but that is only a pilot hypothesis. Actual net contribution uses incremental second-order customers relative to control and deducts redemption, support, refunds, and abuse. The $27 paid CAC only compares alternatives under the same definition.

I would use intention-to-treat analysis, bind attribution to the first valid code, and observe the 30-day second order plus the refund window. If delivery capacity is shared, I would randomize by area or use switchbacks to avoid crowd-out. I would expand area by area only when incremental second orders rise, the conservative net contribution is positive, longer-term quality is not worse, and fraud, complaint, cancellation, and on-time-delivery guardrails pass. Otherwise, I would diagnose the funnel and iterate; crossing a risk stop line pauses the program.”

The answer turns “give coupons for acquisition” into a falsifiable incremental growth system. Referrers, recipients, rewards, and fulfillment each have explicit eligibility, triggers, costs, and stop conditions.

Common Mistakes

• Using shares, registrations, or first orders as the north-star metric → the campaign can create low-quality traffic → optimize for incremental second orders or longer retention and use the share funnel only for diagnosis.
• Using $27 directly to choose the reward → channel quality, attribution, and marginal cost differ → compare incremental retained-customer cost, net contribution, and payback under one definition.
• Paying both sides cash at registration → self-referrals and account farms get immediate arbitrage → tie value to completed orders, the refund window, and retention, with caps.
• Targeting only top spenders → spend does not prove willingness to advocate, a relevant social moment, or a credible experience → combine recent satisfaction, repeat fulfillment, capacity, and risk.
• Analyzing only people who shared → participants were already more inclined to refer → use intention-to-treat analysis from randomized assignment and retain non-sharers.
• Ignoring organic advocacy and spillover → the program pays for natural orders and cross-arm exposure dilutes the estimate → keep a no-reward control, freeze attribution, and record network interference.
• Hiding the referrer's benefit to improve clicks → recipients cannot evaluate motivation and applicable disclosure rules may be breached → state the material relationship next to the message and obtain legal review.
• Uploading contacts or sending on a user's behalf by default → acquisition creates privacy and spam risk → let users choose channel and recipient, with reporting, opt-out, and caps.
• Launching nationwide at once → incrementality becomes unidentifiable and local fulfillment may break → start in areas with capacity and set experimental and operating stop lines.
• Adding opaque rules whenever fraud appears → false positives and appeals can cost more than abuse → tier risk, delay high-risk rewards, and retain explainable evidence and appeals.

Follow-Up Questions and Responses

Follow-up 1: Why not use first-order completion as the primary metric?

A discount directly drives the first order, which is also closest to one-time arbitrage. A second order requires the customer to choose the product again after the first experience, so it is a better early-retention proxy. It remains a proxy; after scale, inspect 60- or 90-day retention, frequency, and net contribution. For a naturally low-frequency business, choose a retention window that matches the purchase cycle instead of forcing 30 days.

Follow-up 2: Why reward both sides?

A two-sided reward is a hypothesis to test. The recipient offer lowers trial cost, while referrer credit motivates participation. Published research shows that reward scheme and product characteristics change referral intent, so two-sided cannot be assumed best. Keep no-reward and recipient-only arms, and clearly disclose every incentivized relationship.

Follow-up 3: How do you prove the orders are incremental?

Use a randomized control among eligible referrers to estimate natural sharing and purchase without the campaign, then compare second-order customers per 1,000 assigned referrers. Do not rely only on self-reported discovery source or count every code user as incremental. Freeze the attribution window, exclude prior purchasers, and report cross-arm contamination.

Follow-up 4: Are roommates or family at the same address fraudulent?

Not automatically. A shared address is a risk signal, not a verdict. Combine device, payment, identity, order timing, and refund behavior. Independent payment and genuine fulfillment in a normal household can pass. Delay medium-risk rewards for review, explain the rule in terms, and provide an appeal path.

Follow-up 5: What if rewards increase sharing but reduce trust?

Compare the no-reward, one-sided, and two-sided arms across the full share, open, first-order, second-order, and complaint path. More shares with worse recipient conversion or quality suggests low-intent distribution or damaged trust. Reduce the incentive, test product-related credit, narrow eligibility, or preserve organic sharing. Hiding the incentive is not the remedy because disclosure requirements take priority.

Follow-up 6: When should the program pause?

Predefine stop lines: materially worse on-time delivery or cancellations, fraud loss above budget, abnormal complaint or opt-out growth, a persistently negative conservative net-contribution interval, or identity false positives that cannot be resolved within the service target. Pause the affected market, audience, or channel, preserve data and reward liabilities, diagnose the cause, and then resume, redesign, or terminate.