Nuances of the referral program

Limiting factors, optimization, fraudsters and multi-user features to drive up the referral motion in a product.

Back in 2017, Brain Balfour created arguably one of the best pieces of knowledge in the growth world. This article is on Channel Model Fit. According to it, there’s a certain group of companies that can survive only on referral engines and SEO. Now, if you extend the referral programs to the right side of the Channel-Model’s scale, those products can also benefit from the referral motion. With the PLG being on the rise, the invite flow becomes a core component for the growth of a lot of companies.

I won’t be covering the entire referral program setup, but rather my goal is to highlight a few crucial elements of a great program that often go unspoken. This is a summary of my experience of working on referrals at over ten companies, including Bird and Vimeo.

Disclaimer: I’m about to use viral, referral, invites, and similar words. All of those in a nutshell represent the motion when one person makes another one sign up.

Networks 

People talk about products, they invite others, or they simply share their feedback publicly. To derive referral growth from it, a few conditions must be true: 

• A person needs to have other people to talk about the product.

• There should be a context when it’s appropriate to talk about it.

• Those people should have enough credibility between each other to establish the referral behavior. 

Simply put, if I don’t know anyone who creates music, I won’t tell them about a synthesizer that I bought. I won’t tell people I ordered an essay to be written for me, as it creates a negative reputation. And I won’t believe that there’s a quick scheme on how to make a million bucks advertised on TikTok. 

When are all of those conditions true though? In a network. 

Similarly to social graphs, there are three circles we all belong to. 

The first one - the people you live with.

The second - people you communicate with on a (semi)daily basis.

The third - people you know, but don't necessarily interact with often.

Now, all products that use the referral motion tend to fall into one of those buckets. The odds of having multiple invites for something like Flo for Partners are close to zero. Notion, Coda, and Airtable live just fine on top of the corporate networks. Cashapp and Venmo do incredibly well while their referral program spins between friends. Uber gets the most out of it from people sharing their codes on social networks. And Bird (RIP, I guess?) did well with their on-campus promo during the early days.

In this amazing write-up by Lenny on why Virality is a myth he states: 

“When we see a product going “viral,” it’s very rarely driven by a many-to-many spread, but is instead the result of someone with a large audience broadcasting it (i.e. one-to-many).”

This is true. For products capitalizing on the third circle. And the majority of them are indeed capitalizing the third circle. But not all of them. 

The network your product exists on top of is the limiting factor. This is why Flo’s referral machine, from the example above, won’t sustain only on their Partners feature. This is also why Lenny said that virality is a short-term thing, with Ben Thompson echoing him and saying that people run out of people to talk to. 

If you want to surf the referral wave long enough, you need to really think about what networks are out there and which one you want to capitalize on, and how far can it take you?

For the first circle, it’s physically impossible to get one person to invite the other. There always will be people who get lost going through your funnel and therefore, K-factor* will never be more or equal to 1.

*K-factor (or virality factor) indicates how many people get invited by one user.

But by layering other multi-user features or adding new use cases, it’s possible indeed to tap into other circles. 

Rebalancing rewards

The often-used GxGx program - Get $x off, Give $x off - became a default approach for incentivized programs. From my experience analyzing the referral programs, it seems that for products capitalizing on the first circle, people care more about their friends than about themselves and we see invites increase if users’ friends get higher rewards. The dynamic is completely opposite for the third circle - I guess, we just don’t care so much about strangers.

As time goes on, it’d be beneficial to experiment with different reward balances for different segments of users - based on their value, their location, use cases, etc. Personalizing rewards might be one of the most powerful moves, for something like e-commerce or marketplaces. 

Fuel your program

While at Bird, it was common knowledge that we don’t make a lot of money out of paid ads. It’s just too expensive. But when the referral spins, that’s when the dollars are falling from above. The problem is that we analyzed each channel separately, talking about how unit-economics is bad in one and is beyond any expectations in others. 

It all changed one day. 

I analyzed the dependency between the K-factor and the number of riders within a given city. Once I finished, I couldn’t believe my eyes, so I analyzed another city, and another, and another. And 20 more. The fact remains the same. There’s a clear relationship between the two that follows the pattern (warning: dummy data!) below:

The more people join the app within the city, the higher the K-factor rises. It eventually hits a plateau and declines over time, the S-shaped law is here to stay. There are two main reasons for this pattern: 

• Before recommending something, people need to see that this thing is popular and there’s a social proof to that. No one wants to recommend something that they aren’t entirely sure will be welcomed by others. 

• The number of times someone gets exposed to the referral program increases the chances of conversion. Meaning - that the more people riding with Bird, the higher the chances a person will get pinged multiple times to accept the invite. Alex Schultz explains the importance of frequency here.

This discovery is a perfect representation of why Uber is pairing paid ads with their referral programs. Of course, in their case, there’s also a demand-supply issue that exists, but paid ads power the referral program nevertheless. 

Optimize the loop

When it comes down to optimization techniques, we already have a list of the major ones. And I want to add a few hot takes on top. The K-factor is a bad metric. Not in the  sense that it shows wrong data, but more so because it isn’t sufficient enough to describe the whole system. What we have here are these metrics to measure: 

• % of people eligible for the referral program - if we only show the feature to those who made a $100 transaction within the system, we lose those who have yet to get past the activation stage. Andrew Chen has a controversial take on this one.

• # of exposures - how many times a user sees the entry point to the referral significantly affects the conversion rate.

• K-factor - measured as the average number of invites sent by one user multiplied by conversion rate to sign up. Measured against the entire population.

• Adjusted K-factor - this is not a common thing; essentially, it’s the K-factor measured against those who sent at least one invite. The difference between the K-factor and the Adjusted K-factor shows the potential impact we can achieve only if we increase the number of people participating in the program. Usually leads to increasing the number of exposures or rebalancing the reward system. 

• Speed of the invite loop - how fast a user sends an invite after they signed up. There's a huge difference in the number of users we’ll acquire in, say, two years, if the average time for a user to send an invite is 10 days versus 30 days. The model below illustrates this concept. What’s crazy is that when the K-factor is more than 1, the difference keeps getting bigger and bigger

There is one other type of analysis I found to be particularly useful. It’s to visualize the referral graph. It helps identify those who brought either the highest number of users, or those who brought the most valuable users. It then would be beneficial to reach out to those people for an interview, or come up with an additional reward system to support their effort. The graph would look something like this:  

Fraud prevention

Every referral program will be abused. There are two things we can do about that - tightening our eligibility rules by decreasing the number of people who can participate in a program, hence,  targeting only high-value users, assuming they won’t bring bad actors into the system. Or building a flagging algorithm that will either block the reward attribution or send it for a manual review. There’s no one-size-fits-all approach, and it should be modeled by every company. However, I’d like to go over  a few most popular things, that help in identifying bad actors in a priority order: 

• Using an artificial email - there are quite a lot of services that help people create an email for 5 minutes. There are also quite a lot of open-source projects that block those emails from signing up. It might also be used by users concerned with privacy, however.

• Using VoIP - the same principle applies if you use phone numbers during sign-up. 

• Using the same delivery address - while different users with the same delivery address are suspicious, they can also be members of the same household. 

• Using the same card - now, that’s a red flag. Two users with different emails, and different delivery addresses, but using the same card? Alarm bells are going crazy. 

I’d be happy to speedball more ideas if you will. Reach out to me on Linkedin. While working at Braid, the fraudsters were one of the biggest headaches and we all became way more sophisticated in detecting fake users. One of my favorite projects was pulling the IP address from the CRM provider when an alleged fraudster opened up an email and then cross-checked their Amplitude ID to see how many devices were associated with this user. Can’t even say what percentage of really hard working fraudsters got burned this way. 

Multi-user play

Ultimately, and especially within the B2B market with its PLG motion, it is possible to extend the lifetime of a referral program by tapping into multi-user features. Collaboration capabilities at Figma, shared documents at Notion and Google Drive are all examples of the same idea. Even within the B2C space, Group Rides at Bird serves as a creative example of using the multi-user approach. 

I’d write another piece on it the other day, but the key is always around collaboration. And if there’s an existent group dynamic of people collaborating together elsewhere, the question becomes why would they do it within our platform. Usually, this is because each product has a unique context that's hardly transferable into something like Slack. This is why we discuss designs in Figma, instead of saying “Oh, Welcome_Screen_v3_final(final), that button on to the top right corner, let’s make it green”.

So, I usually would ask myself: 

• Is there a use case, where two or more people might use it? 

• And if so, does this dynamic exist in the real world now? 

Similar to why Facebook bought Instagram, it’s hard to invent a new social dynamic, therefore if there’s no collaboration happening in the real world, there’s less sense in trying to invent one. It fails, and Evernote's Shared Notes is a great example of a collaboration motion that didn’t yield any meaningful result, as per Sarah Tavel’s talk with Lenny Rachitsky at 00:17:03.

Conclusion

The most important takeaway is that when we think about the referral program, we should treat it not as a standalone feature, but rather like a whole product area. It implies that one product might have different referral mechanics, and there’s no reason to limit it to only the GxGx program. 

In a combination with other channels - like paid ads - it’d be possible to spin the referral engine as much as possible, yielding that type of growth similar to Uber and Airbnb.