Concept · Growth

Virality

Why sharing is not the same as spreading, and what it takes for users to grow a product on their own

A loop that repeats.

When Microsoft announced on December 31, 1997 that it had acquired Hotmail, the free Web-based email service came with a growth story told by its investors. Steve Jurvetson, whose firm Draper Fisher Jurvetson funded it, wrote that Hotmail had signed up more than 12 million subscribers in its first year and a half. It had spent less than $500,000 on marketing to reach them, he said, against more than $20 million of advertising and brand promotion by its closest competitor, Juno, which had a fraction of the users. The device behind the growth was a single line. Would users object to an addition to their private messages?

In one sentence

A product has self-sustaining viral acquisition when each active user generates more than one new active user on average through a repeatable sharing loop; a weaker loop can still reduce paid acquisition.

Each email demonstrated that the service existed and worked; the source presents user-as-salesperson as the investors’ interpretation, not a measured endorsement or conversion rate.

Sharing and virality are different things. Sharing is an act; virality is a loop that brings in new active users who can continue it. A network effect is different again: it makes the product more valuable to participants already there. A product can spread virally without becoming more useful as more people join, and a useful network can grow through paid acquisition. David Skok’s viral coefficient, K, is the average number of new activated users generated by an active user in a defined cycle. It can be estimated as eligible unique invitations per user multiplied by the share that activate, but the cohort and time window must be fixed. In a simple no-churn model, K above 1 grows the next cohort. With churn, referrals must also replace users lost. K below 1 can still offset some paid acquisition; it cannot sustain growth by itself.

The second number is the cycle time: how long one full turn of the loop takes, from a user joining to the people they invite joining. Skok compares YouTube with Tabblo, two viral products built at about the same time. On YouTube a visitor saw a funny video and sent the link on at once. On Tabblo a user posted photos and invited friends, who might not take photos of their own to share for months. Both had loops, and the short one exploded.

Forms of virality

Products spread through users in different ways, and the way decides how large K can get and how fast the loop turns.

Embedded in use

The product’s normal output carries an invitation. Hotmail’s footer on every outgoing email is the classic case: the user does nothing extra and the recipient sees the offer.

Collaboration

Using the product with others draws them in. Dropbox’s 2018 filing says many people use it for work and spread the platform by collaborating on projects or sharing content with people who have no account.

Personal invitation

A user picks friends and invites them, often with a note. Sinan Aral and Dylan Walker found that such messages work better per message than automatic broadcasts, but users send them less often.

A continuum, not a switch

Virality depends on how many invitations each user generates, how many recipients convert, and how quickly the loop repeats. Between the two ends sit products that share often but not enough to replace their users, where sharing lowers acquisition cost without producing growth on its own.

LowFew users expose new users to the productHighReferrals replace churn and add retained users

“When users bring users.”
— Execemy

Why it matters

Referral loops can lower acquisition cost, but they are not costless and do not guarantee paying customers. More than 90% of its revenue came through self-serve channels, so it did not need a significant outbound sales force. Hotmail’s backers put its marketing spend at under $500,000 for 12 million users.

Design, not just enthusiasm, decides whether a product spreads. Aral and Walker ran a randomized field experiment on Facebook with 1.4 million friends of 9,687 users. Passive-broadcast features, which announce what a user does automatically, produced a 246% increase in peer influence and social contagion. Adding active, personalized messages added only a further 98%. The personalized messages were more effective per message and linked to more sustained use, but people used them less often, so broadcasts generated more total adoption.

Cycle time turns a modest coefficient into a large result. A product with K near 1 and a cycle of hours grows differently from one with K near 1 and a cycle of weeks. Skok’s YouTube and Tabblo comparison makes the same point without any new arithmetic.

Finally, virality connects to the rest of the business. Dropbox’s filing shows that users who arrive this way are numerous and mostly do not pay: over 500 million registered users at the end of 2017 against 11 million paying ones. Spread and revenue are separate questions.

Real-world examples

The same concept shows up in different ways across industries.

HotmailA footer on every email

→

Hotmail appended a promotional line to every message its users sent, so that each email carried an invitation. Its investors reported more than 12…

DropboxUsers as the sales force

→

Dropbox’s 2018 IPO filing reported over 500 million registered users, more than 100 million of whom had signed up since the start of 2017.…

ZyngaA loop that ran through Facebook

→

Zynga’s 2012 annual report said it acquired most of its players through unpaid channels, using the viral and sharing features of social networks. It…

When it breaks

A viral loop can depend on a channel someone else controls. In December 2012, Facebook amended those terms to prohibit apps on its canvas from promoting games on other platforms, and Zynga said it would be barred from cross-promoting to games on other platforms. The filing attributes the fall in daily users in the fourth quarter to declines in web and mobile players. It does not say the loop failed, and it does not name virality as the cause.

Viral loops also deliver users, not revenue. Dropbox’s filing says a majority of its registered users may never convert to a paid plan, and it cautions that its word-of-mouth and referral model may not continue to be as successful as it expects. The company runs in-product prompts and time-limited trials to convert free users. That ratio, roughly one paying user for every 45 registered ones, is why an acquisition loop is only half of a growth model.

Finally, the arithmetic is unforgiving. Skok’s model shows that a coefficient below 1 does not compound: growth stops after a few cycles. A K below 1 can still lower the amount of paid acquisition a company needs, but it is not the self-sustaining loop that Hotmail’s backers described. A company claiming viral growth should be able to state its K, its cycle time and how many of the users it counts convert to revenue.

Key takeaways

01
What share of active users generate a unique invitation, what share of invitees activate, and how long is one cycle? Compare retained referrals with churn; K below 1 can assist acquisition but cannot sustain growth alone.
02
Which sharing action fits normal use, and does it bring in active users rather than impressions or invitations? Test passive sharing and personal invites against the same activation and retention measures.
03
Who controls the referral channel, and what share of referred users remain active or pay? Recheck the loop if platform rules or incentives change.

Sources

Viral Marketing (Netscape M-Files, May 1, 1997; edited version in Business 2.0, November 1998), text reproduced by the author · Steve Jurvetson and Tim Draper, via Steve Jurvetson’s Flickr, 1997. Hotmail’s Amazing Growth (12 million subscribers in 1.5 years; under $500K marketing; Juno over $20 million); Tim Draper’s suggestion and its contentious reception;
Microsoft Acquires Hotmail · Microsoft Corp., 1997-12-31. Announcement of acquisition, December 31, 1997; no financial terms stated
Lessons Learned – Viral Marketing · David Skok, For Entrepreneurs. Viral Coefficient (K = invites × conversion; must exceed 1; 5% conversion example); Viral Cycle Time (YouTube versus Tabblo)
Creating Social Contagion Through Viral Product Design: A Randomized Trial of Peer Influence in Networks · Management Science, Vol. 57, No. 9, 2011-09. Abstract: 1.4 million friends of 9,687 Facebook users; passive-broadcast +246%; active-personalized additional +98%
Dropbox, Inc. Form S-1 · U.S. Securities and Exchange Commission, 2018-02-23. Our Business Model (word-of-mouth, in-product referrals, sharing; collaboration; small proportion via paid marketing; 500 million registered, 100 million since start of 2017); Risk factors (11 million paying users; majority may never convert; referral model may not continue); 2017 revenue $1,106.8 million;
Zynga Inc. Form 10-K for fiscal year 2012 · U.S. Securities and Exchange Commission, 2013-02. Item 1 Marketing (unpaid channels, viral and sharing features); Risk factors (Facebook primary platform; 81% of bookings, 86% of revenue; terms changes); Item 1 Facebook agreements (December 2012 terms); Item 7 quarterly DAUs and explanation of Q4 2012 decline