CLOSETHELOOP

A note before you start

Most brands buy every customer they have. The few that compound get each customer to bring the next one.

PayPal paid people cash to join and cash to invite friends. It worked, and it was ruinous: in 2000 the company spent $21.0 million on promotion and marketing against $14.5 million of revenue. Then the product took over. Every payment sent to an email address was an invitation the recipient had a reason to accept, and eBay sellers did the recruiting. Within eighteen months the cost of each new account fell by more than 80%, and the account base kept climbing. Chapters 5 and 6 take that apart.

That is the whole book in one company. A funnel ends at the sale. A loop starts there: something the customer does after buying produces the next customer, and the next customer does it again.

Dropbox paid in storage it made itself and cut its sales and marketing spend while growing past 500 million registered users. Whatnot's sellers fund the giveaways that fill their own shows, and the marketplace sold more than $8 billion in 2025. Binance pays referrers out of fees their friends already paid; I worked on the retention underneath that loop. OVRLND, a vehicle-sweepstakes brand whose retention strategy I ran in 2026, pays in entries it can print for free. Five companies, five currencies, one pattern.

A funnel ends at the sale. A loop starts there.

The Second Order was about one link: getting a first-time buyer to buy again. This book is about the link after it: getting that buyer to bring someone. It stands on the best writing about loops, from Eric Ries and Andrew Chen to Donella Meadows (see What the Books Got Right and The Shelf), and adds what those books mostly skip: physical goods with margins, rewards that fall under sweepstakes and consumer law, and proof that doesn't come from the referral dashboard.

How to read it

Start with The Loop Audit; your score names the part to read first. Or follow a path:

• Founder: Loop Math, PayPal, Pick the Currency, When Loops Break.
• Operator: Pay on Completed Value, Put the Invite Inside the Use, The Game Is the Loop, Thirty Days to a Loop.
• Marketplace, creator or community brand: Whatnot, Seed the Hard Side.
• Drops, launches or giveaways: The Game Is the Loop, The Loop Between Cycles.
• Analyst: Loop Math, Count the Loop, Not the Link, For Your Analyst.

Every chapter ends on a task. Seven tools run in the page; nothing you type leaves your browser.

What's proven and what isn't

• Filed: numbers from securities filings: PayPal's 2001 and 2002 registration statements and Dropbox's 2018 S-1.
• Published: what a company or founder published (Drew Houston's 2010 slides, Jeff Raider's account of Harry's launch, help-center terms, Binance announcements, Whatnot's own report) or what a named outlet reported, with the date it was true.
• Derived: my arithmetic on filed or published numbers, shown with its inputs.
• Reported: figures from my own work as the business or platform recorded them. They come from OVRLND, as Klaviyo or Recharge recorded them, and none is a holdout result. Where I can't re-derive a figure, as with Binance, it isn't here.
• Forecast: plans, labeled as plans.

Examples that open with Say or Picture use made-up round numbers. Every source is listed in Appendix C.

OVRLND appears by name. Its figures come from plans and pulls I wrote in 2026; they show how a loop gets diagnosed, not what it earned after I left.

Before anyone opens a report, ask three people one question. The founder, the growth lead and whoever runs retention each guess what share of last quarter's new customers came through an existing customer: a referral, a gift, a shared link, a friend at the checkout. Write the answers down, then pull the number. The spread between the guesses is your first finding.

Most teams guess high. A refer-a-friend page exists, a few codes get redeemed, and the room assumes word of mouth is doing real work. Then the pull says 3%, and half of those would have bought anyway.

Then open your store, your referral app and your email platform, and score each check 0 to 2: 0 if it failed or you can't answer it, 1 if partly true, 2 if clean. Don't read the bands until all ten are scored.

A referral page is not a loop. A loop is a reason to bring someone, built into the product.

The ten checks

1. Loop share, from memory · 4 minLook at: Nothing. Ask three people what share of last quarter's new customers came through an existing customer.
   Good: Three answers within five points of each other and of the pull.
   Cost if wrong: A team that believes word of mouth carries a third of growth stops funding the loop exactly when it needs work.
   Read next: Funnels End, Loops Return
2. The loop fits on one line · 3 minLook at: A blank page. Finish this sentence: a customer ___, which gives them a reason to involve ___, who then ___.
   Good: A specific moment in the product's use, a specific other person, a specific action. PayPal's version: a customer sends money to an email address, the recipient has to open an account to collect it.
   Cost if wrong: If the sentence needs the word "share," there's no trigger, only a button.
   Read next: Funnels End, Loops Return
3. The reward fires on completed value · 4 minLook at: Your referral terms. When exactly does the inviter get paid: on the invite, the signup, the first order, or the first order delivered and past its return window?
   Good: Paid on a completed, delivered first order. Whatnot pays the inviter after the friend's first purchase is delivered.
   Cost if wrong: A reward paid on signup buys fake signups. Fraud rings find it within weeks.
   Read next: Pay on Completed Value
4. You know what the reward costs you · 4 minLook at: The reward's face value beside your cost per dollar of it, after redemption rate.
   Good: A number per referred customer, beside your paid acquisition cost.
   Cost if wrong: A $20 cash reward and a $20 credit look the same on the landing page and differ by half on the P&L.
   Read next: Pick the Currency
5. The invite lives inside the use · 4 minLook at: Every place a customer can invite someone, and when each appears.
   Good: At least one placement at a moment of use (delivery, first use, a win, a gift) and not only a footer link or a quarterly email.
   Cost if wrong: The invite reaches people when they have no one in mind.
   Read next: Put the Invite Inside the Use
6. Referred customers are a cohort · 5 minLook at: Second-order rate, refund rate and average order for referred customers, beside paid customers acquired the same months.
   Good: The table exists, with customer counts, and someone read it this quarter.
   Cost if wrong: You can't tell a loop that brings your best customers from one that brings reward hunters.
   Read next: Retention Is the Fuel
7. Cycle time is measured · 4 minLook at: Median days from a customer's first order to the first order of someone they referred.
   Good: A number in days, by entry product.
   Cost if wrong: Without it you can't say whether the loop pays back inside this quarter or next year.
   Read next: Loop Math
8. The fraud brake exists · 3 minLook at: Caps per inviter, phone or payment verification, self-referral blocks, rewards held until after the return window.
   Good: All four, and a monthly count of voided rewards.
   Cost if wrong: The loop scales its own abuse. PayPal set aside $11.0M for transaction losses in 2000, 0.87% of payment volume.
   Read next: When Loops Break
9. Counsel has read the mechanic · 3 minLook at: Any reward with a chance element, any bonus tied to a purchase, any text or email the program sends on a customer's behalf.
   Good: Reviewed in writing, with a free method of entry wherever a prize and a purchase meet.
   Cost if wrong: Prize, chance and a required purchase together can make an illegal lottery.
   Read next: The Game Is the Loop
10. The loop is counted on total inflow · 3 minLook at: How the program is judged: referral-attributed signups, or total new customers against a group that never saw the prompt.
    Good: A randomized holdout, or at least a before-and-after on all new customers with the referral count beside it.
    Cost if wrong: Codes get used by people who were coming anyway, and the dashboard counts them all.
    Read next: Count the Loop, Not the Link

Score as you go; your band appears when all ten are in.

Read your score

Fix zeros before ones, in check order. The early checks feed the later ones: you can't price a reward until you know when it fires, and you can't count a loop you haven't drawn.

Set your two lowest beside the answers to check 1. If the room guessed 20% and the pull said 3%, the loop is a story the team tells itself. The rest of this book turns it into a system.

Do this

• Score all ten. Write the total, the three guesses from check 1 and the real number on one page, and hand it to whoever owns growth before Monday.

Draw your growth the way you'd explain it to a new hire. Most DTC brands draw a funnel: ad spend at the top, visits, carts, a sale at the bottom. Then the diagram stops. The customer has paid, the attribution window closes, and next month the team buys another batch of strangers.

A loop is a different drawing. The sale is the middle, not the end. Something the customer does after buying puts the product in front of someone else, and that person's purchase starts the same turn again.

Brian Balfour, Casey Winters, Kevin Kwok and Andrew Chen gave the idea its current name in a 2018 Reforge essay, "Growth Loops are the New Funnels." Eric Ries had already described three engines of growth in The Lean Startup (2011): sticky, viral and paid. The idea is older than either. Tupperware parties ran on it.

What those writers mostly studied was software, where the next user costs nothing to serve. A DTC brand ships a box. Every customer the loop brings carries cost of goods, shipping and a reward. So the question here is narrower: what does it cost to make a customer bring a customer, and when does it pay?

The four parts

1. A trigger inside the useA moment when using the product involves another person. Sending money involves the recipient. Sharing a folder involves a colleague. Going live involves an audience. Without a trigger, a referral program is a button waiting for someone to feel generous.
2. A currencyWhat the inviter gets for bringing someone, and what it costs you to print. Cash, credit, product, status, a share of revenue, a chance at a prize. The currency decides whether the loop pays for itself.
3. Completed valueThe event that counts. An invite isn't a customer, and a signup isn't either. The loop turns when the new person does the thing that earns you money, and the reward waits for it.
4. Retention underneathOnly active customers trigger anything. A customer who churns after one order invites no one after that order. The Second Order was about this part.

An invite isn't a customer. The loop turns on completed value.

Five loops, one table

Parts two to four take four famous loops apart, and OVRLND runs through part five. Here they are side by side.

Read down the currency column. Only PayPal paid in cash, and PayPal is the one that had to stop. Everyone else pays in something that costs them less than it's worth to the customer, and three of them pay only after the new customer has produced revenue.

Why most DTC referral programs aren't loops

Most brands install a referral app, set "give $15, get $15," put a link in the footer and a block in the post-purchase email, and call it a loop. It has a currency. It usually lacks the other three parts.

There's no trigger: nothing about using a moisturizer or a hoodie involves another person. The reward fires on the friend's first order, often before the return window closes. And nobody tracks whether the inviters are still customers in ninety days.

So the program collects codes from people who were buying anyway, shared into group chats full of people who were buying anyway. The dashboard shows referral revenue. Total new customers don't move.

That's not a reason to skip the loop. It's a reason to build all four parts.

Draw yours

Do this

• Fill in the loop canvas above with the four parts named: trigger, currency, what counts, and who has to stay active for it to turn.
• Circle the part you can't name. That's the chapter to read next.

Every loop comes down to two numbers. K is how many new customers each customer brings. Cycle time is how many days pass between a customer's first order and the first order of the person they brought. Get both from your own file before you believe any growth story, including the ones in this book.

K, measured honestly

K is invites sent per customer times the share of invites that become customers. Two invites each and a 10% conversion gives a K of 0.2: every ten customers bring two more.

Count the second half strictly. A customer is someone who completed a first order that wasn't refunded, not a signup and not a code redemption on a cancelled cart. Signups inflate K the same way opens inflated email.

Viral software can run above 1, where every user brings more than one more and growth feeds itself. A consumer brand shipping boxes almost never does. Assume yours is somewhere between 0.05 and 0.4 until the file says otherwise.

Below one, a loop is a discount

When K is under 1, each customer you buy brings a chain that ends: K customers in the first turn, K×K in the second, and so on. The chain adds up to 1 ÷ (1 − K) customers per customer bought. At K = 0.2 that's 1.25. At 0.5 it's 2.

That changes what your acquisition really costs. Spread the paid cost and the rewards over everyone the chain brings, and the blended cost per customer is:

blended cost = acquisition cost × (1 − K) + reward per referred customer × K

Say acquisition costs $40, K is 0.2 and each referred customer costs $20 in rewards across both sides. Blended cost is $40 × 0.8 + $20 × 0.2: $32 plus $4, or $36. The loop took 10% off every customer you buy.

Below one, a loop is a discount on every customer you buy.

That framing keeps the loop honest in a budget meeting. It won't replace paid acquisition. It lowers its price, and it tells you exactly how much the reward can grow before the discount disappears: at K = 0.2, a reward above the acquisition cost itself makes the blend worse, not better.

Cycle time decides when it pays

David Skok, a venture investor, published the standard formula on his For Entrepreneurs blog. Customers after t days, starting from a seed, with K and a cycle time ct:

customers(t) = seed × (K^(t ÷ ct + 1) − 1) ÷ (K − 1)

Run it on 1,000 seed customers:

Two things jump out. At K = 0.5, halving the cycle time gets you to the same 2,000 customers sooner, and no further. Below 1, cycle time decides when the loop pays, not how much. Above 1, the same halving turns 25,075 customers into 101,845 by day 365. That's why software founders obsess over cycle time and DTC operators should obsess over K first.

Cycle time still matters to you for one reason: cash. A loop whose referred customers arrive in 20 days pays back inside the month you spent on the seed. One that takes 200 days pays back in a different fiscal year, after some of the inviters have churned.

Retention is inside K

Invites per customer isn't a fixed trait. It's a rate times a lifetime. A customer who stays active for a year has twelve months of moments to bring someone; one who churns after the first order has a week.

That's why the same referral program shows a K of 0.05 at one brand and 0.3 at another. The second brand usually doesn't have a better referral page. It has customers who stick around long enough to use it. Retention Is the Fuel works through this.

Run it on your numbers

Do this

• Compute K and cycle time from last quarter's orders, counting completed first orders only.
• Put the blended cost on the same slide as acquisition cost at the next budget meeting.

Every network starts empty, and a payments network starts emptier than most: you can't pay someone who isn't on it. In late 1999 PayPal's answer was to pay people directly. $10 to open an account, and $10 more for every friend who opened one. It was the most expensive currency a loop can use, and for a few months it was the right one.

What the bonus bought

By Max Chafkin's account in The Contrarian, his 2021 biography of Peter Thiel, the bonuses came to about $20 for every new user. PayPal had a few thousand users in November 1999, 100,000 in January 2000 and a million about four months after that. By then it had spent something like $20 million of the $28 million it had raised on referral payments.

That's Andrew Chen's cold start problem, solved with money. A network has to reach a density where it's useful before anyone joins for the product itself, and paying people is the fastest way to get there. Thiel's own book, Zero to One (2014), argues that distribution is part of the product, not something you add after. PayPal treated it that way from the first month.

PayPal's S-1, its registration statement for the 2002 stock offering, picks up the count from March 2000:

What it cost

The income statement has one line for this, "Promotional and marketing." It ran $0.9 million from inception through 1999 and $21.0 million in 2000. PayPal's revenue in 2000 was $14.5 million, and its net loss was $169.5 million. The company spent more buying accounts than it earned from all of them together.

Cash is the one currency with no gap between what the customer sees and what you pay. A $10 bonus costs $10, every time, whether the new account is a real customer or a second account the same person opened. And cash is exactly what fraudsters want. The next chapter shows what that did to PayPal's losses.

Cash is the one reward that costs exactly what it's worth.

How fast the bill came down

Divide each period's marketing spend by the accounts it added and you can watch the loop take over from the money:

DerivedThe first two rows divide the "Promotional and marketing" line by the change in account count; they cover all promotion, not only bonuses. The third is the 2002 filing's own figure: average total marketing expense, including promotional bonuses, per new account.

Accounts kept arriving at the same pace or faster: about 18,000 a day in the second quarter of 2001 and about 28,000 a day in the first quarter of 2002. The money stopped doing the work because something else had started.

What a DTC brand can take from it

Most brands run PayPal's first phase forever. A discount for signing up, a discount for referring, paid in the currency closest to cash, with no end date and no second phase behind it. The bonus becomes the price.

• Treat a cash-like reward as ignition. It can get a loop started in a dense group. Write its end date before you launch it.
• Watch cost per new customer, not cost per referral. PayPal's win shows up in the per-account line, not in the bonus count.
• Plan the second phase first. If nothing in your product brings people in once the money stops, the money is just a slower paid channel.

Do this

• Write an end date on any cash-like referral reward you run today, and what replaces it.
• Chart cost per new customer by quarter beside your referral reward amount.

Dropbox's early growth is usually told as a referral story. Drew Houston's own slides tell it in a better order: first a product people wanted badly enough to wait for, then a paid channel that failed, and only then the loop. The order matters, because most brands try to build the loop before anyone wants the product.

Houston gave the talk, "Dropbox: Startup Lessons Learned," at the Startup Lessons Learned conference in April 2010. The slides are public, and they're still the best primary source on how Dropbox grew.

The waitlist

In March 2008, before launch, Dropbox posted a short demo video of the product working. It was voted up on Digg about 12,000 times, and the beta waitlist went from 5,000 people to 75,000 overnight.

A demo shows the product doing its job instead of describing it, and for a product like Dropbox that moment sells itself. Sean Ellis and Morgan Brown's Hacking Growth calls that the aha moment, and argues you should find it before you spend a dollar on growth. Dropbox found it, filmed it and let the audience share it.

A waitlist is also Robert Cialdini's commitment principle at work. Seventy-five thousand people had asked for something, in public, and were waiting for it.

The paid channel that failed

Dropbox launched publicly in September 2008 with 100,000 registered users and reached a million about seven months later. In early 2009 it tried search ads. By Houston's slides, it paid $233 to $388 to acquire a customer for a $99 product.

The problem wasn't the ads. Nobody was searching for a product category that didn't exist yet. People didn't know to look for file syncing; they had to see it, usually on a friend's computer.

You can't buy search demand for a thing nobody knows to search for.

What it means for a brand

If your product solves a problem people already search for, paid search works and the loop is a discount on it. If your product is new, different or easier to show than to describe, the loop isn't a nice-to-have. It's the channel. Most interesting consumer products are in the second group for longer than their founders expect.

• Film the moment it clicks. Not the brand, not the founder story. The first result.
• Build a waitlist before a launch, and give people on it something to do while they wait. The Loop Gallery shows what Harry's did with one.
• Price your paid channel against your product, not against your benchmark. If a customer from ads costs more than their first year is worth, the loop is the plan.

Do this

• Write down the moment your product clicks, in one sentence, and when a new customer reaches it.
• Put first-year contribution beside paid cost per customer for each paid channel.

Whatnot's first live show sold about $5,000 of Funko Pops in roughly two and a half hours, most of them above market price. Grant LaFontaine and Logan Head, the founders, were collectors themselves. They had pitched around a hundred investors before anyone funded them. In 2025 the marketplace they built sold more than $8 billion of goods, more than double the year before, and in August 2026 it raised $545 million at a $20 billion valuation.

This part takes Whatnot's loop apart in three chapters: how it started (this one), how a single show compounds (chapter 10), and what the loop costs the people who run it (chapter 11).

The timeline

PublishedSales and account figures are Whatnot's own. Funding figures are as reported.

Why supply came first

LaFontaine has described Whatnot as mostly supply driven. Andrew Chen's The Cold Start Problem explains why that's usually right for a marketplace: one side is harder to win and does most of the work. On Whatnot that's the seller, who has to show up on camera, stream for hours, and pack and ship everything they sell. Buyers follow sellers. Sellers don't follow buyers until there are a lot of them.

The founders' own first show was the recruiting pitch. It said, in dollars, what a night on Whatnot could earn. The early playbook, as LaFontaine told NEA in 2023, was deliberately slow: one live auction a day, then two, then three, and only then could sellers schedule whenever they wanted. Each show had to be full before there was another.

Fill one room before you open a second.

One category at a time

When Andreessen Horowitz led the Series A in March 2021, Whatnot sold Pokémon cards, Funko Pops, pins and sports cards, and LaFontaine told TechCrunch that about 95% of the team's effort went into the livestream experience. The a16z announcement described the format as a group treasure hunt, with hosts curating items and keeping the room lively.

LaFontaine has said category expansion followed what users asked for rather than a master plan. By 2026 the marketplace covered more than 250 categories, and the fastest growth in 2025 came from beauty, electronics, jewelry and women's fashion. Each new category repeated the start: a few sellers who already had audiences, and shows full enough to prove the format.

What a DTC brand can take from it

• Find your hard side. For a brand it's rarely suppliers. It's the creators, stylists, coaches, retail staff or superfans who already have an audience and a reason to grow it.
• Show them the money, literally. Whatnot's pitch to sellers was a $5,000 night. Yours is what a post, a class or a drop earned the last person who did it.
• Pace the launch. One full event beats five half-empty ones. A loop that starts in an empty room teaches everyone it doesn't work.

Do this

• Name the ten people outside your company with the biggest audiences who already buy from you.
• Write what one event with you would earn them, in dollars, from a real example.

The show loop brings buyers to sellers. Two more pieces keep the marketplace growing: a referral program that only pays for real customers, and a trust layer that makes a stranger's live auction safe to bid on. And one piece threatens it: the cost the loop puts on the people running it.

The referral program

Whatnot's buyer referral program pays the new user a credit when they sign up and pays the inviter only after the friend's first eligible purchase is delivered. Credits cover the order subtotal, never shipping, fees or tax. They expire after 90 days and cap at $10,000 per person. New users must verify a phone number, each friend can use one link, and mass email, messaging strangers and bots are banned.

One version is randomized: a credit between $10 and $200, with the odds published in the terms.

PublishedWhatnot Help Center, read September 2026. The minimum credit is $10.

That's Eyal's variable reward with the fine print written out. The chance of a big credit makes the invite more exciting than a flat $10, and publishing the distribution keeps it honest.

Trust is part of the loop

Buying a $400 card from a stranger on a live stream takes trust. Whatnot vets sellers: by Contrary Research's account, fewer than half of applicants were approved. It offers buyer protection and emphasized authenticated collectibles from its early days. a16z's investment note leaned on the founders being genuine collectors who understood sellers' problems with older marketplaces.

Trust isn't a separate program from the loop. Every bad transaction breaks it at the step where a viewer becomes a buyer, and every good one makes the next follow more likely.

The bill

The same 2026 report that showed the cadence multiples also carried the other side. An analysis of it by Value Added Resource, a marketplace-seller publication, noted that sellers find the real costs of giveaways and logistics surprisingly high, that some report net earnings modest or even below minimum wage after fees, and that buyer overspending tied to gamified features is a live trust concern.

PublishedValue Added Resource, 2026, on Whatnot's report and seller accounts. These are seller reports, not platform-wide figures.

That's the balancing loop Peter Senge would look for. A loop funded by its participants holds only while they come out ahead. When the giveaways, shipping and fees eat the margin, sellers stream less, and the cadence that made the loop compound slows. Whatnot's seller fee, 8% plus payment processing of 2.9% and $0.30 per order by Contrary Research's 2023 count, is the platform's side of that bargain.

A loop that runs on someone else's money has to keep paying them.

What a DTC brand can take from it

• Copy the referral terms. Pay after delivery, subtotal only, cap, expiry, phone verification, one link per friend, no bots or mass messages.
• If you randomize, publish the odds. A variable reward with hidden odds is a trust problem and, depending on the mechanics, a legal one.
• Price your partners' side. If creators or affiliates fund giveaways or samples for you, know what each event costs them and what it earns them.
• Design against overspending. Limits, cooling-off messages and clear totals cost you a little revenue and protect the trust the loop runs on.

Do this

• Read your referral terms against Whatnot's, line by line, and fix any that pay earlier than delivery.
• Write down what your best partner spends to promote you, and whether they'd say it's worth it.

I worked at Binance from 2019 to 2020, on contract, as a retention marketing manager. I didn't build the referral program or the fee tiers; they predate me. I worked on the part they depend on. A referrer earns nothing from a friend who stops trading. A fee tier means nothing to a trader who leaves. The acquisition loops are only as strong as the retention loop inside them.

What I worked on

• Loyalty tiers for high-volume traders, designed so the best customers had more reasons to stay than to leave.
• Twenty-plus customer journeys with fifty-plus conditional-logic paths, triggered by behavior across email, SMS and in-app messages.
• A gamified path from a new user's first trade to their first volume milestones, so every new trader had a visible next step instead of an empty dashboard.

Five principles I still use

1. Map the path to a habit as milestonesSignup isn't the goal and neither is the first transaction. Write the four to six steps between a new customer and a customer with a habit, and name each one so the customer can see it.
2. Branch on behavior, not on who people areFifty conditional paths sounds like complexity. It's the opposite: each customer gets the one message that fits what they just did or didn't do, and nothing else.
3. Pay the top of the file in status and costA better tier rewards the behavior you want and costs nothing on trades or orders that don't happen. A discount to your best customers marks down orders that were coming anyway.
4. Treat quiet as a signalA heavy user who goes quiet for a week is a bigger risk than a light user who goes quiet for a month. Set the trigger on their own rhythm.
5. Suppress when someone is already movingA customer climbing the ladder doesn't need a nudge. Every message you don't send to them is one less reason to unsubscribe.

A new customer needs a next step, not a welcome.

How it fed the referral loop

Run the numbers from chapter 3. A referrer's lifetime value to the loop is how many friends they bring times how long those friends keep trading. Every point of retention raises both: the referrer stays active long enough to bring more people, and the people they bring keep generating the fees that pay the referrer. Retention work wasn't a separate program from the referral loop. It was the loop's fuel line.

The same moves, for a consumer brand

The Second Order covers the timing and branching in detail. The point here is where it sits: underneath every acquisition loop you build.

Do this

• Write your ladder: four to six milestones from first order to habit, with the day each should happen.
• Pull your top 5% of customers by 12-month revenue and check how many are still active this quarter.

Every loop pays the inviter in something. Pick it on three numbers: what it costs you per dollar of value the customer sees, how many rewards actually get used, and how many referred customers would have bought anyway. The best currency costs you a fraction of what it's worth to the customer and makes them use the product more.

Part two's companies spread across the whole range. PayPal paid cash and had to stop. Dropbox paid in storage it made itself. Whatnot's sellers fund the giveaways. Binance pays a share of fees already collected. OVRLND pays in entries it can print for free, inside rules that make the printing legal.

Seven currencies

Pay in a currency you can print.

The credit that isn't cheap

Store credit looks like the product row and behaves like the cash row. A $15 credit costs you your cost of goods on the order it's spent on, if it's spent. That part is cheap. The expensive part is the orders that were coming anyway: the friend who was going to buy, and the inviter who would have reordered at full price and now pays $15 less.

That second cost never shows on the referral dashboard, which is why credit programs look better than they are. The calculator below asks for it directly.

Entries: the currency with a legal price

OVRLND sells entries into vehicle giveaways: $1 buys 10. It can also grant entries for free, and does, like the 150 free entries in its signup offer. Printing an entry costs the company almost nothing, and to a customer who wants the truck it's worth real money. That makes entries close to the perfect loop currency.

The price is legal, not financial. In a paid sweepstakes every grant of entries is part of a legal structure, and free entries have to be available on equal terms to people who don't buy. A referral bonus paid in entries needs a clause in the official rules and a free path to the same grant. My 2026 plan for OVRLND put it bluntly: entries cost nothing; whale trust doesn't.

Price your reward

With the example numbers, a $15-each credit costs about $6.30 per referred customer and $9.40 per incremental one, about a fifth of a $45 acquisition cost. The same face value in cash would cost about $45 per incremental customer, the same as buying one. That's PayPal's problem in one line.

Do this

• Run the calculator on your current program, with your real redemption rate and a third for anyway until you have a holdout.
• Price one product, status or access reward you could offer instead, and bring both numbers to the next growth meeting.

PayPal's invite was a payment. Dropbox's was a shared folder: shared folders and similar features brought in a fifth of its signups in January 2010 without any reward at all. Whatnot's is a show announcement. None of them asks the customer to go looking for a friend. The friend is already part of what the customer is doing.

Most DTC products aren't used with other people. But almost every one has moments where another person is present, and those are where the invite belongs.

Where the other person already is

Put the ask where another person already is.

Time it to the use, not the order

Most post-purchase referral asks go out the day the order ships, because that's when the flow starts. The customer hasn't touched the product. They have nothing to say to a friend except that they bought something.

Move the ask to the first moment the product has done its job: the day after delivery for apparel, the first result for skincare, the first match for a sports product, the first draw for a sweepstakes. It shortens cycle time too, because the invite happens at the earliest moment the customer has a reason to make it, not weeks later in a quarterly referral campaign.

Let the customer send it

OVRLND's referral design made one decision that removed a whole class of risk: the customer shares their own link, and OVRLND never sends the invite. A brand that emails or texts a friend's contact on a customer's behalf is sending marketing to someone who never consented, which puts the email and text rules squarely on the brand. A customer texting their own friend is just a person talking to a friend.

It also makes the invite better. A link from a friend, in the friend's words, beats any template you could write.

The receipt as the invite

OVRLND's best-performing email wasn't a promotion. It was the post-purchase confirmation headed "ENTRIES LOCKED," which told buyers how many entries their order had just added. It earned $1.59 to $1.90 per recipient, Klaviyo-attributed, among the highest in the account. People forward and screenshot good news about themselves. A receipt that shows a customer what they just won, earned or unlocked is the most natural place in the program for a share link.

Do this

• Write down the first moment your product has done its job, in days from delivery, and move the referral ask there.
• Add a recipient offer to every gift order, printed in the box, not emailed to the buyer.

OVRLND gives away vehicles: a restored Bronco, a Tacoma, a 1985 Toyota pickup. Customers buy gear and entry packs, and every dollar buys ten entries. When I rebuilt its retention plan in July 2026, the diagnosis took one sentence. The company had already built a game, and its email program wasn't playing it.

The game had four parts. A currency: $1 buys 10 entries. A status ladder: membership tiers from Bronze to Platinum. A recurring event: weekly draws for members. And proof people win: real winners, drawn live. That's a complete loop engine, the same one Whatnot runs with auctions and giveaways and Binance ran with tiers and token sales.

The welcome flow, meanwhile, ended in a coupon ladder: 25% off membership, then 50%.

ReportedDifferent flows reach different people at different stages, so read the order, not the ratios. The Tacoma figure was earned when orders averaged $35 to $45, not the $22 to $24 of mid-2026.

The coupon tail earned $237 on 27,326 sends. The receipt that simply told buyers how many entries they had just locked earned more per person than anything else in the account. It had no discount in it and, at first, not even a number.

A discount ends the game. A scoreboard keeps it going.

Give the game a scoreboard

The plan was one idea executed everywhere: show players their score. Three rules made the scoreboard work for the business and stay honest with the customer.

1. Show progress, not claims about oddsShow the customer's stack, how much it moved and their rank: "+620 entries locked, you passed 1,200 players." Every one of those numbers can be checked against the customer's own orders. Odds and total-entry counts live in the official rules, where the law puts them, and nowhere in marketing, because every number in a sweepstakes email is a representation that has to be exactly true.
2. The score appears only when it movesA balance that just changed is news. The same balance re-sent every week is wallpaper. Counts earn a subject line only on movement.
3. The lifetime score never resetsEach giveaway ends, and most players lose. Lifetime entries survive the loss. They turn a draw into a chapter instead of an ending, and they're what the next giveaway builds on.

Nir Eyal's Hooked names the last step of a habit loop investment: something the user puts in that makes the next trigger more likely. Lifetime entries are exactly that. A player with 1,860 lifetime entries has something to lose by not coming back.

Every mechanic is a legal structure

In the US, a prize, a chance and a required purchase together make a lottery, which private companies can't run. Sweepstakes stay legal by giving a free alternative method of entry with equal standing. That's why Whatnot's Buyer Appreciation Giveaways carry a free web form, and why every OVRLND mechanic went to counsel before it shipped.

The review turned that into a design rule: in a paid sweepstakes, every mechanic is a legal structure and every displayed number is a representation. Compliance is a design input, not a sign-off at the end. Three examples from the plan:

• The Sealed Envelope, a welcome surprise, is one per customer for life, a fixed amount identical for everyone, disclosed in the official rules with a free path to the same grant. The mystery lives in the copy, never in the mechanics. A random bonus on a purchase is a second lottery.
• The referral bonus ships only when the next giveaway's rules include it, with a free way to earn the same entries.
• No rules change mid-cycle, ever. What isn't in the rules before entries open ships as a non-entry perk, or waits.

What not to add

The plan's list of cuts was as long as its list of plays. No entry multipliers in flows. No bolt-on points, badges or spin-wheels: entries already are the points. One device per moment: the entries receipt stays in post-purchase, where it's true, instead of being pasted into every flow until it means nothing.

You don't need a sweepstakes to use this. Any brand with repeat purchase can find its currency (points that buy something real, progress toward a product), its ladder, its recurring event (a monthly drop, a weekly restock) and its proof (real customers, named with permission). Most already have three of the four and bury them under discounts.

Do this

• Write down your currency, ladder, recurring event and proof. Mark any that don't exist yet.
• Build one receipt that shows the customer's score moving after each purchase, and put the share link on it.

Some businesses run in cycles: a giveaway with a draw date, a drop that sells out, a season that ends. Each cycle has its own loop, and one more loop connects the cycles. Whether the next launch starts from zero or from a crowd depends on what you do on the day the last one ended.

The shape of a cycle

OVRLND's plan for its 1985 Toyota SR5 giveaway ran 56 days, from July 24 to September 17, 2026. It forecast $500,000 in revenue on $250,000 of ad spend, with $60,000 of giveaway costs, and a daily curve shaped like a U.

ForecastA plan, not a result. The shape matches the company's past closes, which ran $210,000, $244,000 and $277,000 in total store sales in their close months.

Launch week and close week earn most of the money. The 38 days between earn about a quarter of it, at $3,200 a day. And after day 56, the draw: one household wins and every other buyer loses on the same day.

Loss Day

The Bronco giveaway closed on July 23, 2026, with 9,488 buyers entered. The first version of the plan built them a visible entry balance, then had no design for the day that balance lost. That's the biggest single moment of disappointment the brand creates, and it arrives on a schedule.

The fix was a Loss Day sequence, sent the same day as the winner reveal. It leads with the number that survives the loss, lifetime entries, and with the next action already open. Big buyers get a plain-text note and a text from the shop. Anything the sequence promises about the next giveaway, like a head start for veterans, waits until counsel has cleared it and the rules include it.

The draw is the biggest crowd of the cycle. Don't let it land on nothing.

The Cover List

The winner announcement draws more attention than anything else in the cycle, and in the first plan it ended with nowhere to go. The Cover List gave it a destination: a one-click waitlist for the next vehicle, in the announcement, in every send between cycles and on the site. List members get a head start when entries open and a launch-day bonus revealed then.

Launch day then runs the close in reverse. Instead of two or three texts pushing a deadline to strangers, two or three texts go to a list of people who asked to be told.

Don't let the loop forget

The cycle structure hid a slower leak. Each new giveaway got new flows, and the old ones were switched off. The Tacoma welcome flow earned $0.52 per recipient; the rebuilt Bronco welcome earned $0.12. Whatever the Tacoma version knew was lost when it went to draft.

• Keep evergreen flows evergreen. Swap the prize, not the structure.
• Key balances per cycle. One stored balance per giveaway, never a single "current giveaway" field that rolls over while last cycle's emails are still sending.
• Work the trough. Weekly member draws, bonus-entry weekends and drops give the 38 middle days a reason to buy.
• Bring back past players. A lapsed-entrant flow names the giveaway they played and their lifetime entries when the next one opens.

Do this

• Draw your cycle's daily revenue curve and mark the day it ends. Write what every buyer sees that day.
• Put a waitlist for the next cycle in your last-day and announcement messages.

Peter Senge's The Fifth Discipline (1990) describes a pattern he calls limits to growth: a reinforcing loop runs until it triggers a balancing loop that slows it down. Donella Meadows' Thinking in Systems (2008) makes the same point with stocks and flows. Every customer loop in this book hit at least one brake. The ones that survived had built it in advance.

Fraud

A loop that pays people attracts people who fake the thing you pay for. PayPal's S-1 shows a $11.0 million provision for transaction losses in 2000, 0.87% of payment volume, and one July-to-October episode that cost $5.7 million. By the second quarter of 2001 the rate was 0.33%.

For a DTC brand the fraud is smaller and the same shape: self-referrals under second email addresses, codes posted to coupon forums, bot signups for sweepstakes entries. The brake is in Pay on Completed Value: pay after delivery, verify the new customer, cap each inviter, void and ban.

The law

Binance's growth outran its compliance. On November 21, 2023, the company and its founder pleaded guilty to US federal charges in a resolution totaling about $4.3 billion. The Justice Department said Binance had prioritized growth and profits over compliance with US law. Among its findings: until August 2021, accounts could be opened with only an email address, and after announcing in 2019 that it would stop serving US customers, the company kept many, including high-volume VIP traders.

A DTC loop won't face the Bank Secrecy Act. It faces smaller versions of the same edge:

• Sweepstakes law, where a prize, a chance and a required purchase meet. A free method of entry, in the rules, before entries open.
• Email and text consent, when a program messages a friend on a customer's behalf. Let customers send their own links.
• Endorsement disclosure, when rewarded customers or creators post about you. The FTC expects the reward to be disclosed.

Legal is a design input, not a sign-off at the end.

Check the mechanic

Before any reward with a chance element ships, answer eight questions. The checker names the pattern and what to fix. It doesn't replace counsel; it tells you what to ask them.

Rewards that stop working

A rich reward brings people who came for the reward, and they leave when it shrinks. PayPal cut its bonuses and then relied on the product. Binance put its BNB fee discount on a published schedule, 50% in the first year and 25% from July 2018, and years later ended the link between BNB holdings and referral rates. A reward that steps down on a schedule announced in advance doesn't feel like a broken promise.

Running out of people

A loop starts in its densest group: eBay traders, card collectors, the brand's first thousand fans. As it spreads, each new customer is further from the core, knows fewer people who'd care, and K falls. Dropbox capped referral space at 16 GB and 32 GB partly so the reward stayed a bonus as the loop spread. For cycle businesses, each giveaway or drop starts from the audience the last one left behind, which is why The Loop Between Cycles matters.

Do this

• Write the brake for each of the four before your next change to the program: fraud, law, reward decay, saturation.
• Send counsel one package, covering every mechanic at once, before launch.

Whoever owns the loop, a hire, a promotion or you, needs five things on day one.

1. The order file, taggedEvery order by customer and date, with the referring customer attached where there is one.
2. Admin on the referral appIncluding the exact event that pays inviters, and the terms.
3. A name at counselSomeone who has read the sweepstakes, consent and endorsement rules for your mechanics.
4. The top 20 invitersNames, what they bought, whom they brought.
5. A holdout digit on every profileWritten once, never recomputed.

Do this

• Pull last year's new customers tonight and tag the ones who came through someone. Look at the share.

• The Lean StartupEric Ries, 2011The three engines of growth, sticky, viral and paid, and the advice to tune one at a time.
• "Growth Loops are the New Funnels"Brian Balfour, Casey Winters, Kevin Kwok, Andrew Chen, Reforge, 2018The case for drawing growth as loops, where each cycle's output becomes the next cycle's input.
• "Lessons Learned: Viral Marketing"David Skok, For EntrepreneursThe viral growth formula with cycle time, used in Loop Math.
• The Cold Start ProblemAndrew Chen, 2021Atomic networks, the hard side, and how networks tip, scale and stall.
• Hacking GrowthSean Ellis and Morgan Brown, 2017How growth teams find the moment a product clicks and test quickly around it; Ellis worked with Dropbox early.
• Viral LoopAdam L. Penenberg, 2009Early histories of companies that grew through their users, including PayPal and eBay.
• HookedNir Eyal, 2014Trigger, action, variable reward, investment: the habit loop inside a customer that feeds the loop between customers.
• Thinking in SystemsDonella Meadows, 2008Reinforcing and balancing loops, stocks, flows and delays.
• The Fifth DisciplinePeter Senge, 1990Limits to growth and the other patterns that stall a loop.
• Turning the FlywheelJim Collins, 2019How to write your own flywheel, and why the order of its parts matters.
• InfluenceRobert Cialdini, 1984Reciprocity, social proof and scarcity: the psychology inside every double-sided reward and giveaway.
• ContagiousJonah Berger, 2013Why some things get talked about, and how to design for it.
• Zero to OnePeter Thiel with Blake Masters, 2014Distribution as part of the product, from someone who ran PayPal's loop.
• BlitzscalingReid Hoffman and Chris Yeh, 2018Growth when speed matters most, with PayPal and eBay from the inside.
• The ContrarianMax Chafkin, 2021The source used here for PayPal's early bonus amounts and user counts.
• The PayPal WarsEric M. Jackson, 2004An insider's history of PayPal's fight with eBay and fraud.

What most of these books leave out is what this one tried to add: loops that ship physical goods with margins, rewards that fall under sweepstakes and consumer law, and proof from a holdout rather than a dashboard.

Andrew Lauchner runs Growth Legend, embedding three to twelve months inside consumer brands to own lifecycle, email and SMS, and revenue operations. He wrote The Second Order, on turning first-time buyers into second-time buyers.

In 2026 he ran retention strategy for OVRLND, a vehicle-sweepstakes brand, the work behind chapters 18 to 20. As a contract retention marketing manager at Binance (2019–20), he designed loyalty tiers for high-volume traders and more than twenty customer journeys, the work in chapter 13.

As Senior Director of Growth and Retention Marketing at Gallery Furniture, he rebuilt the customer journey and the sales playbooks together. At 3Commas he worked on growth and retention and co-led the Copy Trader launch. He has been Head of Growth and Retention at Greatness Wins and at Nexus Agriscience.

His results here are as the platforms and companies reported them. None is offered as a holdout result. Count the Loop, Not the Link sets a stricter standard for yours.

Andrew answers every note from people building a loop, including those looking for someone to own it. Write to andrew@growthlegend.com or message him on LinkedIn.

Every pull below runs on an order table and a customer table. Resolve identity first, the way The Second Order's appendix does: customer ID, then normalized email, then phone. A split identity turns one referred customer into a stranger and a self-referral into a win.

The loop's data contract

Six pulls

1. Loop shareNew customers in the period with a non-empty referral_source, over all new customers. Bad reading: a share that jumps the month a coupon forum finds your code.
2. K, trailing 90 daysReferred customers with a completed, unrefunded first order in the window, over customers active in the window. Bad reading: counting signups.
3. Cycle timeMedian days from each inviter's first order to their first referred customer's first order. Bad reading: using the invite date instead of the order date.
4. Referred cohort quality90-day second-order rate and refund rate, referred against paid, same acquisition months. Bad reading: comparing a referred cohort from last year with a paid cohort from last month.
5. ConcentrationShare of referred customers brought by the top tenth of inviters. Bad reading: ranking on invites sent instead of customers brought.
6. Known before referralShare of referred customers who were on your list, had visited, or had ordered under another identity before the referral. That's a floor on "would have come anyway."

K in one query

-- referred customers per active customer, trailing 90 days
WITH active AS (
  SELECT DISTINCT customer_id FROM orders
  WHERE status = 'fulfilled' AND ordered_at >= CURRENT_DATE - 90
),
referred AS (
  SELECT c.customer_id FROM customers c
  JOIN orders o ON o.customer_id = c.customer_id AND o.is_first_order
  WHERE c.referred_by IS NOT NULL
    AND o.status = 'fulfilled' AND o.refunded_at IS NULL
    AND o.ordered_at >= CURRENT_DATE - 90
)
SELECT (SELECT COUNT(*) FROM referred) * 1.0
     / NULLIF((SELECT COUNT(*) FROM active), 0) AS k_90d;

Adapt the names to your warehouse. The logic is the point: completed first orders over active customers, same window.

Small-file rules

• Under ten expected referred customers per arm, don't run the test. Decide another way and write down how.
• Report every rate with its count. "12% of referred customers reordered" means little without "(n = 41)."
• Fix the read date before the test starts, and only look early to check the groups are clean.

PayPal (chapters 5 and 6)

• PayPal, Inc. Form S-1, 2001. Operating data, selected financial data, risk factors.
• PayPal, Inc. Registration statement, 2002. Accounts, marketing cost per new account, auction share, losses.
• Max Chafkin. The Contrarian: Peter Thiel and Silicon Valley's Pursuit of Power. 2021, pp. 59–60.
• Peter Thiel with Blake Masters. Zero to One. 2014.

Dropbox (chapters 7 and 8)

• Drew Houston. "Dropbox: Startup Lessons Learned." April 2010.
• Dropbox, Inc. Form S-1/A. March 2018.
• Dropbox Help Center. "How much free referral space can I earn?"

Whatnot (chapters 9 to 11)

• Whatnot Help Center: Giveaways; Buyer Appreciation Giveaways; Buyer Referral Program.
• Whatnot. 2026 State of Live Selling report, as covered by Tubefilter (January 2026), cllct and Value Added Resource.
• TechCrunch, March 2021. Andreessen Horowitz, "Investing in Whatnot," March 2021.
• NEA, "Marketplace Building Spotlight," September 2023. Masters of Scale, November 2025.
• Contrary Research, Whatnot. Business of Fashion, October 2025. Quartz, August 2026. Fortune, August 2026.

Binance (chapters 12, 13 and 23)

• Bloomberg. "How a Billionaire Crypto King Built the No. 1 Exchange in Just 8 Months." March 28, 2018.
• Binance: tiered fees, July 2018; BitTorrent sale results, January 2019; referral program; Learn and Earn; Binance Angels; 250 million users.
• US Department of Justice. "Binance and CEO Plead Guilty to Federal Charges in $4B Resolution." November 21, 2023.

The gallery (chapter 21)

• Adam L. Penenberg. Viral Loop. 2009. Excerpt, TechCrunch, October 2009.
• Jeff Raider. "How to Gather 100,000 Emails in One Week." July 2014.
• Tyler Denk. "How Morning Brew's referral program built an audience of 1.5 million subscribers." Digiday.

Frameworks (chapters 3 and 4)

• Brian Balfour, Casey Winters, Kevin Kwok, Andrew Chen. "Growth Loops are the New Funnels." Reforge, 2018.
• David Skok. "Lessons Learned: Viral Marketing." For Entrepreneurs.
• The books on The Shelf.

OVRLND (chapters 14 to 20, 22)

• My own 2026 plans and pulls for OVRLND: the growth plan and scoreboard, the 45-day flow plan, the Bronco final-month strategy and KPI scorecard, and the SR5 giveaway plan. Figures are Klaviyo- or Recharge-recorded, or forecasts where labeled.