How did a structured form of customer referrals come into being? It’s said that the first documented referral program was created by Julius Caesar, who in 55 BC would paid his soldiers 300 sestertii (something like a third of their annual pay) to refer a friend to join the army. And thousands of years later, we still use, plus or minus, the same idea. It seems as though every consumer app has implemented some form of a referral program, though I argue it really kicked off in ~2008, which is when Dropbox’s innovative referral program was rolled out.

Yes, the most famous early implementation of referral programs came from Dropbox, which inspired a generation of startups — particularly YCombinator-backed startups — to experiment with similar ideas. Why did this make sense for them? CEO/cofounder Drew Houston’s made a very helpful presentation describing his journey towards referral programs, and the general trajectory was the following:

• First, do all the things you’re “supposed” to do
  • Big bang launch at a tech conference. Try some AdWords, hire a PR firm / VP of Marketing
  • Paid marketing programs created a CAC of $233-388 for a $99 product
  • Then trying affiliate programs, display ads, and many other things — which all failed
• … but then failing! And realizing none of it works that well
• Then realizing the key was to accelerate word-of-mouth and viral growth by offering a “give and get storage space” program
• Boom. 100,000 users to 4 million in just 15 months, with 35% of daily signups

The entire deck is wonderful — created roughly a decade ago, but still very relevant — and I highly encourage you to check it out here.

Referral programs work very well for certain kinds of products, particularly ones that are already spreading via word of mouth. In Dropbox’s case, there is a natural use case between friends and colleagues — shared folders — which naturally complement the referral channel. Referrals drive that forward, providing an economic incentive to tell friends. As another example, at Uber where I ran the referral programs for drivers and riders at various points (and spent >$300M/year on them), the program for driver referrals was naturally successful. Drivers were often from certain sub-communities, whether newly arrived immigrants or limo drivers, and people were naturally already talking about the earning opportunity. Referrals, sometimes as high as $500/signup, accelerated that in a big way.

And yet referral programs have their limits. Of course they don’t really work that well for products that have low LTV — that’s why we don’t see free social photo sharing apps reward their users for referrals. There’s no LTV to arbitrage against, and the referral amounts create a form of customer acquisition cost. They also tend to decline in importance over time. Years after the rollout of Dropbox’s referral program, I had the opportunity to join Dropbox as an advisor, where I got a first-hand look at the data. By then, the natural virality of their core product — just the process of people sharing their folders and files with others — had come to completely dominate user acquisition. This had become the primary method of spread, and the referral program became much less important. I’ll discuss why, later on, but this seems to be the natural pattern of things — referral programs are very helpful at the beginning of a market. Eventually it becomes less important, and that’s OK.

But we get ahead of ourselves. Let’s start first by looking at how a referral program is usually defined.

The headline best practice is that your referral program should target new users to refer their friends — this means prompting users during their initial onboarding flows, and adding emails as part of the onboarding, among other surface areas. This is in direct contradiction to folks who often argue to let users experience the product first, have a good experience, before they’re hit up to invite. Why focus on new users? First, mathematically, it’s easiest to make a big impact when you are hitting a cohort of 1000 new users when it’s as close to 1000 as possible, not in day 30 when the cohort will have churned and gotten down to 150. And in the math of the viral factor, you have a better chance to hit >1 when you have 1000 users invite 1000 users than to ask 150 to invite 1000. Second, new users generally have more friends who haven’t yet used the product, because they are new themselves. Once they have gone through the referral program a few times, then they will have naturally tapped out their networks.

And of course, the simplest thing to do is a “give $5, get $5” and give that offer to everyone, in an untargeted fashion. But a product leader soon realizes that this is inefficient — perhaps it’s best to give some users $15 and others $5, depending on their value. This is exactly what many marketplace companies have done, when it’s easy to segment their network into high-value cities like New York and SF versus, say, Memphis — you can set custom referral amounts in each place. But why stop at cities? Perhaps you do an analysis and figure out certain leading characteristics of high-value users as their account balance, or the types of other apps they use, or otherwise — once you think of this as personalizing an ephemeral offer to users, then you can run whatever promotions you want.