Two-sided referrals: Rewarding advocacy you already have
Every referral program is a bet on a prior fact: that some of your users already want to tell people about you.
If that is true, a reward removes the last bit of friction and you get more of something good. If it is false, the reward is the only reason anyone shares, and what you have built is a coupon. Coupons work, briefly, and they bring you people who wanted a coupon.
A referral program cannot manufacture advocacy. It can only lower the cost of advocacy that exists, which is why the first question is not what to offer but whether anybody is recommending you now, unpaid, without being asked.
What is a two-sided referral?
A two-sided referral rewards both people: the one who shares and the one who arrives. Single-sided programs reward only the sharer.
The difference is not the economics, it is the social cost of the ask. “Use my code and I get $20” uses up the referrer’s credibility for you; they are asking a favor and both parties know it. “Here is $20 for you, and I get $20 too” is an entirely different sentence, because the referrer arrives bearing a gift, not a request.
That is the whole mechanism, and it is real. It is also frequently the only true thing in a business case otherwise built on viral arithmetic.
The viral coefficient is usually fiction
The standard model is seductive: invites sent multiplied by conversion rate gives you K, and K above 1 means self-sustaining growth. It is arithmetically correct and almost always inapplicable.
Look at what a K of 6 requires: something like fifteen invitations per user converting at forty percent. Those are consumer-scale numbers, and the B2B products we have seen reported publicly are nowhere near them.
The canonical example is worth reading properly, not through the summaries. Drew Houston’s own 2010 presentation on Dropbox’s early growth reports that the referral program “permanently increased signups by 60%” and accounted for 35% of daily signups, with users sending 2.8 million invitations in a single month.1 Note what that is and is not: a large, durable lift, credited to a program on a product whose reward was more of itself. It is not a self-sustaining loop, and Houston does not claim one.
Worse, K is a rate, so it decays. Early adopters invite enthusiastically because they found you first and their networks are dense with people like them. As adoption spreads, each new cohort has fewer uninvited contacts, and the coefficient falls even when nothing about the product has changed. A program that hits K above 1 in month two and plans on it for a year is planning on the wrong number.
The same deck contains the more useful number, and it is the one nobody quotes. Dropbox tried paid search and measured a cost per acquisition of $233 to $388 for a $99 product, a slide Houston labels simply “Fail.”1 That is the actual case for referrals: not viral arithmetic, but that the alternative channel cost several times the price of the thing being sold.
What the reward selects for
Abuse resistance is the part that gets skipped, and it decides whether the program is a good idea.
A reward changes who refers. Without one, the people who recommend you are the ones who genuinely like the product enough to spend social capital. With one, you also get people motivated by the money, and they refer differently: broader, less carefully, to people who fit less well.
It also changes who arrives. Somebody who signed up for a $20 credit is a different customer from somebody who signed up because a colleague they trust said this was the tool to use. Both count as one signup, and they behave nothing alike afterwards.
Which produces the measurement rule for this whole pattern: the only number that matters is how referred users retain compared with everyone else. Referral volume, participation rate and K all measure activity. Retention tells you whether you bought customers or traffic, and it is not usually on the dashboard referral tooling gives you.
The reward has to fit the product, not the budget
Three shapes, and the differences matter more than the amounts.
Product currency. Extra storage, extra credits, an extended tier. The classic version of this granted both parties more of the thing they were already using. Its virtue is that it selects for people who want more of your product, and it costs you marginal capacity rather than cash.
Two refinements make product currency work harder. Let it accumulate: a reward that keeps paying for each additional referral gives your best advocates a reason to continue rather than stopping at one. And pay immediately, because a reward that arrives after a delay is a promise, and promises motivate far less than the thing itself.
Cash or gift cards. Universal, effective, and indiscriminate: it attracts people who want cash, which is everybody. Reach for it when the product has no meaningful internal currency, or when getting to critical mass is genuinely existential and you are willing to buy your way there, which is the situation the famous payments-era cash bonuses were actually addressing. Expect the selection problem above.
Status or access. Early access, a visible badge, a place in a community. Cheap, weak on its own, and strong in combination with something material.
The choice interacts with your pricing. A credit against a subscription discounts revenue you would otherwise have collected; product currency in a usage-based model is genuinely cheaper for you and genuinely valuable to them, which is the most favorable version of this trade available.
Products nobody can easily recommend
| The reward | Compounds | Paying for signups |
|---|---|---|
| Existing advocacy | People already recommend you unprompted | Nobody mentions you without being paid |
| Who benefits | The referrer’s contact gains something real | The reward only makes sense to the sharer |
| Conversation fit | The category already comes up between peers | Nothing in normal conversation invites the mention |
| Explainable | A colleague can describe what it does in a sentence | So complex that recommending it requires a demo |
| Network overlap | Your users know other potential users | Users are isolated, or their peers are competitors |
| Reward affordability | The reward costs less than acquiring that customer | It costs more than an ad and converts worse |
| Abuse resistance | You can tell a real signup from a manufactured one | Trivially gamed with disposable accounts |
Conversation fit and explainability are the two people skip. A referral needs a moment in an existing conversation to attach itself to (“how are you handling that?”), and it needs to survive being described secondhand by somebody who is not you. Products that fail either condition are not badly incentivized, they are hard to recommend, and that is a positioning problem, not a rewards problem.
Abuse resistance deserves more attention than it usually gets. Any reward with cash value will be farmed, and the farming is invisible in aggregate metrics until somebody looks. Requiring the referred account to reach a real usage milestone before either party is paid fixes most of it, and it also aligns the reward with a product-qualified signal rather than a form submission.
The number that decides it
Do referred users retain better or worse than the rest, at the same tenure? That is the whole question. Worse retention means the reward is buying signups, not customers, and every increase in volume is making the business shallower.
Three things qualify the answer. Compare total referred signups against the pre-program baseline, because some of those referrals were always going to happen and you are now paying for them. Check whether participation sits in a handful of accounts, since a program carried by twenty enthusiasts is a relationship rather than a channel and will not scale by raising the reward. And cost it per retained referred customer against your other channels, never per signup, which is the metric that makes referral programs look cheapest by ignoring whether anybody stayed.
Price your alternative first
- Price your alternative channel first. Houston’s case for referrals was not a growth coefficient, it was that paid acquisition cost multiples of the product’s price. Work out your own equivalent, because it tells you how much a referral is worth paying for.
- Count your unpaid referrals. Ask new signups how they heard about you, in one question, and read the answers before designing anything.
- Name the moment worth talking about. If you cannot, that is the work, and no reward structure substitutes for it.
- Pay in product currency if you have one, because it is cheaper for you and self-selecting for them.
- Gate the reward on real usage, not signup, which fixes both the abuse problem and the quality problem at once.
- Instrument referred-user retention from day one, separately. If you build only one report, build this one.
- Assume the coefficient decays, and check it quarterly rather than annually.
A reward makes recommending easier. It still requires someone to decide to recommend, which is a deliberate act most people never get round to. The stronger version of advocacy skips the decision entirely: the product produces something a user wants to show off for their own reasons, and your name travels attached to it. Achievement sharing makes the user the beneficiary of their own promotion.
Footnotes
-
Drew Houston, “Dropbox Startup Lessons Learned”, presented 23 April 2010: the referral program “permanently increased signups by 60%” and produced “35% of daily signups,” with 2.8 million invitations sent in the trailing 30 days; the paid-search experiment recorded a “cost per acquisition: $233-$388” against a “$99 product,” on a slide labeled “Fail.” A dated first-person account from the founder, read directly, and unusually credible because the numbers it reports include a failure. It describes one consumer product in 2010 and should not be treated as a benchmark for anything else. Other figures commonly attached to this pattern do not survive checking and are not used: the “3,900% growth” framing of the same Dropbox period, “7-10% daily growth” for an early payments company, and “two-sided rewards lift referral rates 3x” all circulate through referral-software vendors’ pages citing each other. The K formula is arithmetic rather than evidence: correct as stated, silent on whether any product achieves the inputs. The widely repeated split attributing most viral growth to product value rather than referral mechanics is a practitioner’s framing, often credited to Reforge, and we could not locate the underlying publication. ↩ ↩2