Referral program examples that worked, and the mechanics behind them

Three referral program examples that worked are Dropbox, PayPal and Morning Brew, and each paid a reward close to the product only after the new person showed real use. Dropbox gives 500 megabytes (MB) per friend on Basic plans and grew from 100,000 to 4,000,000 users in 15 months. PayPal paid $20 to both sides, and Morning Brew tied branded rewards to referral milestones.

What the referral program examples that worked have in common

The three programs share one design choice. In our reading, each paid a reward that was the product itself, or close to it, and each paid only after the new person did something that showed real use.

Dropbox pays in storage. A Basic user earns 500 MB of space for each friend they refer, and paid plans earn 1 GB. Both sides get the bonus, but only after the friend installs the desktop or mobile app, logs in from it and verifies their email. That trigger is close to activation: the friend has to touch the product, not just create an account.

PayPal paid in cash, which was its product. It started by offering $20 to both the referrer and the new user per successful sign-up, then cut the reward to $10 and later $5. Referral Rocket reports the company grew from 1 million to over 5 million users in a few months.

Morning Brew paid in branded rewards on a milestone ladder, from stickers up to a full work-from-home setup. In 2019 it began requiring a double opt-in so only legitimate referrals counted. Referral Rock reports that 30% of growth came from referrals, and as much as 80% in the early days. It also puts the cost at about 25 cents per referral, against $4-5 per subscriber on its largest paid channels.

Company	Reward	Trigger	Who gets paid	Reported result or limit
Dropbox	500 MB per friend (Basic), 1 GB (paid plans)	Friend installs app, logs in, verifies email	Both sides	Basic capped at 16 GB
PayPal	$20, later $10, then $5	Successful sign-up	Both sides	1 million to over 5 million users in a few months
Morning Brew	Tiered: stickers up to a work-from-home setup	Double opt-in from 2019	Referrer	30% of growth, up to 80% early

What to copy this week

In our reading, the lesson is the pairing, not the size of the reward. A reward that is more of the product brings back people who will use it. A trigger tied to real use filters out fake accounts before you pay.

Write down three things before you build:

1. The reward unit. Pick something your product already sells: seats, credits, storage, months of a paid tier.
2. The qualifying action. Choose the first step that shows time to value, such as a first upload or a first invite accepted.
3. The cost at scale. As an example, not a benchmark: if 1000 referrals qualify and each costs you 10 dollars in reward value, the program costs 10,000 dollars.

Then track how many referred users go on to refer others, so you can estimate your viral coefficient once the trigger is in place.

How did Dropbox's referral program work?

Dropbox's referral program has four parts: a reward made of the product, a reward for both sides, a ceiling on what one account can earn, and a set of conditions the friend must meet before anyone is paid. The per-friend amounts and the Basic cap appear in the first section. The paid-plan ceiling is higher: paid accounts can earn up to 32 GB through referrals.

The conditions before the reward is released

Step the friend completes	What it protects against (our reading)
Accepts the invitation and signs up	Rewards for invites nobody acted on
Installs the app (covered earlier)	Accounts that never touch the product
Logs in from the Dropbox app	Installs that are never opened
Verifies their email address	Throwaway or fake addresses

We read this list as an activation gate. The reward costs storage only after the friend has done the things a retained user does.

Why a storage reward made sense: the worked numbers

One account of the early years reports that Dropbox had to spend $233 to $388 on search ads to acquire a new paying customer, for a product priced at $99 a year.

Run the payback period:

• At $233 per customer and $99 a year, $233 minus $99 leaves $134 of ad cost not covered by the first year.
• At $388 per customer and $99 a year, $388 minus $99 leaves $289 uncovered.
• At $388 per customer and $99 a year, payback takes about 3.9 years.

Spare storage has a marginal cost close to zero, so in our reading the reward could be generous without adding to that gap. A cash reward of similar perceived value would have widened it.

The same account reports that the program launched with 250 MB each for referrer and referee, and that referral sign-ups rose 60%. The program formalized word of mouth that already existed.

By early 2010, users were sending more than 2.8 million invites per month, and the company had grown from 100,000 users to 4,000,000 in 15 months.

Company context

Drew Houston co-founded Dropbox in June 2007.

To apply this week, write down your own paid cost per paying customer and your first-year revenue per customer. If the gap is large and you have a product unit that costs little to give away, test that unit as the reward. Release it only after an activation event, and track it as a growth loop.

How PayPal paid cash for referrals

PayPal's early referral offer worked because the reward and the product were the same thing. A payments service that pays you in cash shows you how it works the moment the reward lands. The case study from Referral Rocket describes the program as one that showcased the service by putting money directly into users' accounts. The new user's first experience was a balance they could spend.

The reported results were large. The same case study cites a 7-10% daily growth rate during the period when the user base grew several times over, as covered earlier in this piece. Treat that figure as a reported number from a referral software vendor's write-up, not an audited one.

What PayPal pays for now

PayPal's current referral program for developers works differently. It pays between $500 and $2,500 in commission for a qualified merchant referral. A sign-up alone earns nothing. The merchant's transaction volume is monitored for 60 days, starting at its first PayPal payments transaction, and the payout is based on the annualized volume in that window.

Annualized volume in first 60 days	One-time payout per merchant
Up to $1,000,000	$500.00
$1,000,000 to $3,000,000	$1,000.00
$1,000,000 to $5,000,000	$2,000.00
$5,000,000 and greater	$2,500.00

In our reading, payment is slow on purpose. Developers receive commission within 90 days after the end of each calendar quarter, and the program is limited to individuals and businesses in the US and Canada.

The lesson: pay on verified value

In our reading, the move from a flat reward per sign-up to a volume-tiered reward per merchant is the main thing to copy. The reward follows proof that the referral is worth something to the business.

Here is a worked example, using round inputs rather than benchmarks. Suppose 1,000 referred accounts sign up and 10 of them reach your qualifying threshold, so you pay out on 10, which is 1% of 1,000. At a $500 floor, those 10 payouts cost $5,000, and you paid nothing for the other accounts.

To set this up this week:

1. Pick the value event. Choose one action that predicts revenue, such as a first transaction or a paid invoice. See activation.
2. Set a monitoring window. Give the account long enough to show real usage before you pay.
3. Tier the reward by value. Scale the payout to the volume or plan the referred account reaches.
4. Batch the payouts. Pay on a fixed schedule so you can catch fraud and refunds first.
5. Check payback. Compare the reward to the referred customer's margin over your payback period.

How Morning Brew's milestone rewards worked

Share the Brew had an audience behind it before it launched. The referral program launched in 2017, when the list already stood at 100,000 subscribers, and Referral Rock reports it went on to see some days averaging over 1,000 referrals.

Why milestones suit a free product

The newsletter is also free, and the company earns most of its revenue through sponsored content. There was no subscription to discount and no credit to hand out, unlike the storage and cash rewards covered earlier. In our reading, that is why physical goods at referral counts made sense: the reward sits outside the product, and each new reader adds to the audience sponsors pay to reach.

The mechanics around the tiers

Mechanic	What Morning Brew did	Why it matters (our reading)
Placement	Share the Brew appeared in every single newsletter	Readers see the ask daily, so it does not depend on one launch email
Contests	Ran referral-based contests at times to re-engage past referrers	Gives people who already hit a tier a reason to keep going
Testing	Ran A/B tests on the email and referral page copy	Copy changes are cheap compared with changing the rewards

Reading the funnel

The 2019 numbers show how a ladder thins out. Referral Rock reports that 225,000 subscribers had shared at least one referral and 80,000 had shared three or more, so 80,000 of 225,000 is about 36 percent of referrers reaching that third step.

That makes 500 of 10,000 a 5 percent first step, and 25 of 500 another 5 percent.

Here is a worked example with round inputs, not a benchmark. If 5 percent of a 1,000 reader list starts referring, that is 50 referrers. If those 50 referrers bring in 10 friends each, that is 500 sign-ups. Then check what share of those referrers reached your second tier, against the 36 percent above.

This week: set your first tier at one referral so the first step is easy, put the share block in every send, and test one copy change on the referral page using our A/B test sample size calculator.

Should a referral program be one-sided or double-sided?

Most programs we looked at earlier rewarded both sides. That habit deserves a test, because field research on who gets the reward points somewhere else.

They found that recipient-focused rewards "not only increases sign-ups but can also cut program costs nearly in half."

The mechanism sits in the two steps of a referral. On the first step, customers were equally likely to make referrals whether the reward went to them or to their friend. The second step is harder. The friend has to act, sign up and change a habit. As Gershon put it: "Because the recipients have that harder job, the incentive is more important on their side."

One working paper went further. Against shared rewards that cost twice as much, recipient-only rewards performed about as well.

Double-sided is still the common default. Referral Factory reports that over 90% of programs run on its platform are double-sided. Read that as a count of what its customers chose, not as evidence of what converts. Some double-sided designs are lopsided on purpose: Airtable gave the referrer $10 in credit while the friend got a 20% discount.

Design	Who is rewarded	Where it fits (our reading)
Referrer-only	The existing customer	Affiliate-style programs where sharing is the bottleneck
Friend-only	The new customer	Products where the friend faces real effort or cost to start
Double-sided	Both	Financial products and high-ticket offers, where a referrer may balk at a friend getting a large reward while they get nothing

How to test the split on your own traffic

1. Randomize by referrer, not by visit. Each referrer sees one design, so their friends all see the matching offer.
2. Run three arms: referrer-only, friend-only and double-sided. Hold either the reward value or the total cost per conversion constant across arms, and decide which before launch.
3. Track three numbers per arm: share rate, friend conversion rate and reward cost per acquired customer.
4. Follow each referred cohort past sign-up. Gershon's team saw most referrals arrive in a customer's first weeks, and those were not the most valuable referrals, so check retention before you pick a winner.
5. Size each arm with the A/B test sample size calculator before you start.

As an example, not a benchmark: send 1000 referred visitors into each arm and set every reward at $10. Divide total reward spend by converted friends in each arm. The arm with the lowest cost per retained customer wins, even if its share rate is lower.

Which reward fits your product

Start with the reward menu, then pick the row that matches what you sell. Salesforce lists four common options: discounts as a percentage off the next purchase, cash-back on every successful referral, free items or premium upgrades, and VIP perks such as early access or premium support.

The economics give you room to pay. Yotpo reports that referred customers have a lower customer acquisition cost (CAC) and a higher customer lifetime value (LTV). That margin only holds if you pay for referrals that become real users, so the trigger matters as much as the reward.

The reward-matching template

Copy this into your planning doc and fill in your own trigger event.

Product type	Reward type	Qualifying trigger	When it fits (our reading)
Storage or usage-based software	Product credit or extra capacity	Friend completes the first key action, not just sign-up	The reward costs you capacity rather than cash and pulls the referrer further into the product
Transactional (payments, marketplaces)	Cash	Friend's first transaction is confirmed	Users already think in money, and a credit for something used now and then holds little value
Free media (newsletters, communities)	Milestone perks: merchandise, early access, status	Friend subscribes and confirms	There is no price to discount, so tiers reward volume instead
Subscription	Discount on the next bill or an extended trial	Friend converts to paid	A discount costs a slice of revenue you already earn, and a longer reverse trial costs only time

How to set the trigger

Write down what counts before launch. Salesforce advises that you identify what constitutes a successful referral, the reward system, and any limitations. In Yotpo's description of the loop, the reward fires once the friend's purchase is confirmed. For software, we think the equivalent is your activation event.

Check the reward against buying frequency. Referral Factory points out that a discount on the next purchase of a seasonal product won't be too attractive. If your customers buy rarely, move toward the cash or perk rows.

For milestone rows, set the threshold in referrals. Salesforce gives the example of early access for someone who brings in three new customers.

A worked example

Take 1000 referred sign-ups as an example, not a benchmark. If you pay on sign-up, every one earns a reward. If you pay on conversion and 10 of them convert at a $10 credit each, you owe $100. Run that math against your payback period before you choose a row.

How to measure a referral program, with a worked example

A referral program is a funnel with five stages. Measure each one separately, because a single blended number hides which stage is broken. CloudSponge defines participation rate as the number of people you’ve reached with the program, who have also signed up to tell others about your brand. Signing up is not sharing, so track the two apart.

Metric	How we calculate it	What a weak number points to (our reading)
Participation rate	Users who join the program ÷ active users	The invite surface is hidden
Invites per referrer	Invites sent ÷ users who shared	The reward is too small to act on
Invite conversion	Friends who reach the success event ÷ invites sent	Weak landing page or friend offer
Reward cost per referred customer	Total rewards paid ÷ referred customers	Rewards paid out before activation
Referred-cohort retention	Share of a referred cohort still active at day 30, 60, 90	Low-fit friends

A worked example

These are example inputs, not benchmarks. You have 1,000 active users. CloudSponge reports a share rate of 15% as good performance, and 1,000 users at a 15% share rate gives 150 referrers.

Those 150 referrers sending 10 invites each send 1,500 invites. At an example 10% conversion rate, 1,500 invites bring in 150 new users. That gives 150 new users from 1,000 active users, a k-factor of 0.15.

Which lever moves the number

The same source treats 15% or more as an optimal conversion result. Raise conversion to 15% and the same 1,500 invites bring in 225 new users. Share rate has room to move too: 30% is optimal, while 7% or lower is very poor.

We think conversion is usually the cheaper lever to test first. A landing page is easier to change than user behavior. Size the test with the A/B test sample size calculator before you run it.

Where programs stall

• Low-value reward. A weak share rate can mean your reward fails to engage and compel advocates to act.
• Trigger set before activation. Ask for referrals after the user has reached value, not at sign-up. See activation.
• Hidden invite surface. Low participation may mean your website fails to inform visitors about it.
• Untracked or undelivered rewards. Jeff Epstein, founder of Ambassador, put it this way: "Without the right systems in place, tracking and rewarding referrals can be a nightmare." Yotpo makes the same point: manually tracking referrals isn’t scalable.

This week, pull the five numbers for last month, find the lowest stage against the figures above and put one test for it in your experiment backlog.

How to build a referral program in six steps over eight weeks

Each step has one owner and one output, so a small team can run it alongside other work. The metrics are the ones from the previous section.

Step	Owner	What it produces	Signal it worked (our reading)
Pick the activation event as the trigger	Product lead	A written definition of the moment a user first gets value	The team can query that event for every user
Choose the reward from the template	Growth marketer	One reward from the earlier template, with its cost per referral	Finance signs off on the cost
Decide which side gets paid	Founder	A one-sided or double-sided rule	The rule fits in one line of invite copy
Place the invite where users already get value	Designer and engineer	An invite prompt shown right after the activation event	Users reach the prompt without searching for it
Instrument the funnel	Engineer	Events for join, share, friend visit and friend sign-up	Every stage reports a count
Run a two-arm test before scaling	Growth marketer	A control and a variant that differ in one thing	One arm wins on friend conversion

The eight-week timeline

• Weeks 1 and 2: steps one to three, written up as a one-page brief.
• Weeks 3 and 4: build the invite placement.
• Week 5: instrument the funnel and check every event fires.
• Weeks 6 to 8: run the test, then decide whether to scale.

Log each idea in the growth experiment backlog so the variant you test is the one you chose on purpose.

How to instrument step five

Report the two populations separately. CloudSponge's metrics guide says a program's numbers relate to two distinct groups, Advocates and Friends. Advocates join and share, and Friends act on the link. Blending them hides which side is stuck.

Tag every share event with its channel. The same guide suggests you analyze which sharing channel advocates prefer, since performance varies between channels.

Social reach is the hard part to see. The guide suggests a dedicated hashtag in the message advocates send through your widget, and dark social tracking for views of the referral link.

Reading the two-arm test

Friend conversion is the number to watch. Friendbuy data in the same guide marks a conversion rate of less than 4% as a poor result.

Here is a worked example, not a benchmark. If 1000 friends reach the landing page in one arm and 10 sign up, that arm converts at 1%. In our reading, a result that far under the floor points at the landing page or the friend offer, not at reach. Fix it before you send more traffic.

Size each arm with the sample size calculator before week 6, so the test can end on schedule with a clear result.

What this page does not cover

This page does not cover the tax or legal rules that apply to cash and gift rewards, fraud detection setups, or comparisons of referral software. It is not legal or financial advice, so check reward terms and payout rules with a qualified adviser before launch.

Related on Race to Scale: Waitlist launch strategy: how the referral queue and the gated waitlist work, what converts, and when a waitlist backf; A/B test sample size and duration calculator, with the formula; What is a good activation rate? Benchmarks by product type and how to define yours.