Virality - Freemium Economics: Leveraging Analytics and User Segmentation to Drive Revenue (2014)

Freemium Economics: Leveraging Analytics and User Segmentation to Drive Revenue (2014)

Chapter 7. Virality

This chapter, “Virality,” discusses the means through which a product is adopted by existing users. Virality is difficult to instigate, but it can be facilitated through the specific product development strategies that are outlined in the chapter. The chapter begins with an introduction to the concept of virality, focusing on the benefits of compounded virality and the correlation between retention and virality. The chapter then advances into a description of the formal techniques that can be used to calculate virality, either through a high-level, “global” coefficient that describes the degree to which the entire user base grows through viral channels, or by an individual coefficient that describes how many users each user, on average, can be expected to introduce to the product. The chapter then proceeds into a practical guide to modeling virality in a spreadsheet and ends with various techniques for engineering virality into a product.


virality; k-factor; viral mechanics; viral channel; viral growth; saturation; product retention; viral networks

The viral product

Virality was introduced briefly in the discussion of the minimum viable metrics model as one aspect of a product’s user base growth. This description was intentionally left vague; the truth is that virality is a nebulous concept and is notoriously difficult to measure, despite being one of the critical sources of growth for most consumer Internet and mobile products.

What is virality?

At the user level, virality is achieved through two possible viral mechanics: interpersonal recommendations and in-product invitations. An interpersonal recommendation, as it relates to virality, is any invitation to join a product that takes place outside the scope of the product (i.e., it wasn’t initiated by a user within the product). One example of an interpersonal recommendation as a viral mechanic is a word-of-mouth recommendation, which can’t be counted or attributed to a specific user but which nonetheless contributes to the product’s user base in a viral manner (and is often the most effective means of viral growth).

An in-product invitation is a proposal, initiated by an existing user to a non-user, to join the product, using some mechanism the product offers. In-product invitations often take the form of email invitations or social networking alerts; a user in a product is encouraged to invite the user’s social circle into the product and is presented with a prefilled template for distributing information about the product to others. The number of in-product invitations sent should be measurable through in-product event tracking infrastructure; the number received and acted on may be difficult or impossible to measure on some platforms, such as mobile.

It is not feasible to count the number of interpersonal recommendations a product receives; however, recommendations are likely the best-converting sources of viral growth. Likewise, the number of times a product is adopted because a person was seen using it—perhaps on a tablet on a city bus, or on a laptop in a café—is similarly unknowable. But these sources of users are real, and they must be considered. This phenomenon serves as the greatest obstacle to product development as it relates to user base growth: how can virality be designed into a product if it can’t be measured?

As it turns out, virality can be measured—it just can’t be measured precisely. At a high level, a product is inherently viral or it isn’t; viral mechanics designed as an afterthought generally can’t create contrived growth trends (or, if they can, those growth trends are generally short-lived and insubstantial).

One of the basic properties of a virus is that it is self-replicating; that is, it grows, independent of any external catalysts. By definition, any agent that does not replicate autonomously is not a virus. The same general condition holds with viral growth: if a user was acquired into a product through a marketing initiative, and not from another product user, then that acquisition was not the result of virality.

Calculating virality

The principle metric associated with virality is the k-factor, or the average number of additional users introduced to the product by each user. For reasons discussed earlier, measuring the k-factor with precision, in many cases, is difficult: tracking all the means by which existing users may introduce new users to the product is problematic (if not impossible), meaning conversions cannot be reliably attributed to viral mechanics. When a product platform is surrounded by an opaque data moat, as are many mobile platforms, ascertaining how a user becomes aware of a product can be treacherously costly and laborious. On the other hand, some platforms—notably, web-based platforms—present simple and straightforward methods for tracking this information, rendering the conversion rate from product invitation mechanics easy to quantify. For instance, a web-based product can track invitations and conversions from end to end through URL parameters, meaning confusion can enter into the process only if an invited user doesn’t click on an invitation link but rather types the base URL of the website directly into a browser.

When considering attribution in calculating virality, new users must be separated into three groups: those acquired through paid channels (who are removed from the virality equation completely), those acquired virally (through any means other than direct discovery and paid acquisition), and those acquired organically (those who came across the product without any influence from an existing user). Virally and organically acquired users can be thought of as the two elements of a population ratio; knowing the size of the total user population and the size of one of these user groups is sufficient to determine the product’s virality.

The strategy through which virality is estimated therefore depends on the accuracy provisions of the platform. Two extremes exist on a spectrum of platform data availability: the eminently transparent extreme, at which everything is capable of being tracked and all data related to viral invitations and conversions is at the disposal of the product team, and the eminently opaque extreme, at which no data related to the user source is accessible.

Even on platforms sitting near the eminently transparent extreme, the k-factor is not wholly measurable through counts of invitations and conversions because these platform-centric metrics do not take into account the effects of interpersonal viral mechanics. (SeeFigure 7.1.) Counting conversions that result from in-product viral mechanics only provides one aspect of overall product virality; to capture a conceptual viral metric, or a global k-factor, all users who weren’t acquired organically or through paid channels should be classified as viral.


FIGURE 7.1 Platform transparency versus clarity of the k-factor.

In other words, a conversion rate for viral invitations doesn’t provide much insight into a product’s true virality. Instead, the global k-factor should be calculated as a function of user base growth, of which viral invitations are only one contributor, with the understanding that the metric is fundamentally rooted in broad assumptions about how product use spreads within social circles. One approach to measuring the global k-factor this way is to analyze user base growth as a time series, controlling for external factors (such as media coverage and platform featuring) and natural user base decay.

When the user base is examined as a time series and broken down into periods, the global k-factor can be calculated by dividing the growth of the user base from one period by the number of users in the period preceding it, as shown in Figure 7.2. This calculation distributes the user base’s decay-adjusted growth (the net of churned users), regardless of attributable source, across the entire period-one user base (and thus accounts for interpersonal viral mechanics). This k-factor reflects historically relevant total user base growth as a ratio of the number of users introduced to the product virally, as opposed to a local k-factor that relies solely on the calculation of product mechanics (like viral invitations) to grow the user base.


FIGURE 7.2 A two-period user base growth chart, illustrating the global k-factor calculation.

Calculating virality as a function of in-product viral mechanics is likewise valid but only addresses those sources of virality that are able to be measured. This approach to calculating the k-factor relies on the straightforward formula found in Figure 7.3. In the figure, I represents the average number of viral invitations a user sends and C represents the average conversion rate of those invitations. The result is the number of users converted to the product, on average, via a single user.


FIGURE 7.3 An equation to calculate k-factor.

While the k-factor is more limited in scope than a time-series measurement of the entire user base, the invitations approach to calculating it is not without merit, as it provides clear insight into the performance of in-product viral mechanics and is more immediately actionable with regard to product development. Where the time-series approach relates a high-level (if ambiguous) sense of overall product appeal, the invitations approach relates a very specific measurement of user enthusiasm for the product and the subsequent clarity and allure of product enticements captured therein.

The effects of compounding virality

Perhaps the most appreciable and visible benefit of virality as it pertains to user base growth (especially through paid marketing) is its compounding effect. Like interest on a financial instrument, the compounding effect of virality allows a user base to grow geometrically; such a growth pattern provides access to users that would otherwise be achievable only through organic discovery or paid acquisition, and sets the conditions for massive, profitable scale.

True virality exists at a k-factor at or above 1, meaning each user, on average, introduces more than one new user to the product. In theory, this means that a user base can grow infinitely, after being seeded with an initial set of users, as a function of its own momentum. In practice, a number of limitations exist that cap the absolute extent to which virality can grow a user base.

As with compounding interest, compounding virality is rooted in exponential math, or arithmetic that is driven by some base value being multiplied by itself. The formula for calculating one-period growth, given a per-period rate of growth, is shown in Figure 7.4, where X2 is the value of the quantity in period two, X1 is the value of the quantity in period one, and r is the rate of growth.


FIGURE 7.4 A formula to calculate growth from one period to the next.

Extending this logic, the formula for calculating n-period growth under conditions of a constant growth rate is shown in Figure 7.5, where n is the number of periods over which growth takes place, and X1 is the value at the start of the growth period.


FIGURE 7.5 A formula to calculate growth over n periods.

The rate of growth in these equations would be expressed in terms of per-period percentage increases; for instance, if a user base grew from 100 to 110 in one period, the variable r in Figure 7.4 would be replaced with .10, representing 10 percent growth. If the user base had grown from 100 to 200, or 100 percent growth, the variable r would be replaced with 1.0 (and thus the value within the parentheses would be 1+1, or 2).

A 100 percent growth rate represents a doubling of the user base, which is a significant virality threshold and a benchmark that is often cited in terms of user base growth. A user base doubling in size is a powerful psychological motivator for a product development team and is often used in the product planning process or as motivation for investment into further development. In any case, knowing the time required, given existing growth rates, for a user base to double in size is useful when making decisions about the future course of a product.

The doubling time of a user base, given a consistent growth rate, can be calculated with exponential arithmetic, which allows for equations containing exponents to be rearranged using natural logarithms. To illustrate this concept, consider that the equations inFigure 7.6 are equivalent.


FIGURE 7.6 These two figures are equivalent.

The doubling of a user base over n periods, where Xn is twice as large as X1, is represented in Figure 7.7.


FIGURE 7.7 Calculating the growth rate needed to double a user base in n periods, given that Xn is twice as large as X1.

When a user base doubles, Xn is twice as large as X1, thus resolving to a multiple of 2. Therefore, rearranging the equation in Figure 7.7 using a natural logarithm to isolate the exponent for a doubling of the user base yields the results shown in Figure 7.8.


FIGURE 7.8 Re-arranging Figure 7.7 using a natural logarithm yields Figure 7.8.

Dividing the natural log of 2 by the natural log of the growth factor solves for n, as shown in Figure 7.9.


FIGURE 7.9 Solving Figure 7.8 for n yields Figure 7.9.

So, given a consistent growth rate of 10 percent, a product team can calculate the number of n periods over which a product’s user base will double by using the formula shown in Figure 7.10.


FIGURE 7.10 Assuming a growth rate of 10%, a user base will double in 7.27 periods.

In this example, the user base will double after 7.27 periods (which would be rounded up to eight, because periods are measured only in whole denominations). The general logic of this arithmetic isn’t unique to the doubling of a value; using the same format, thetripling time, quadrupling time, and so on can be calculated by replacing the 2 in Figure 7.10 with the multiple of interest.

This property of exponential arithmetic is often employed using a heuristic known as the rule of 72, which stipulates that the doubling time of any value experiencing compounding growth can be roughly calculated by dividing the growth rate expressed as a percentage (e.g., 10%, not .10) into 72. Using the same values from the example in Figure 7.10, a rough approximation of the doubling time is calculated in Figure 7.11.


FIGURE 7.11 A rough approximation of doubling time.

This heuristic is often used to generate estimates of doubling time in the planning stages of product development. The relationship between QUOTE and QUOTE should underline the notion that higher virality reduces the amount of time required for a user base to double. This is the chief benefit of viral mechanics and viral tendencies: they can create feedback loops that quickly increase the size of a user base.

Understanding the relationship between these variables and taking informed, proactive steps to increase r, thus decreasing n, are fundamental tactics in strategic product development. Over the course of a product’s lifetime, but especially in the earliest stages, user base growth is a product team’s principal concern. A firm grasp of the dynamics driving that growth efficiently and quantifiably is necessary to control it.

Virality and retention

Virality and retention exist on opposite sides of the acquisition threshold: virality describes how users are introduced to a product, and retention describes how long users remain with a product. But in essence, both sets of metrics measure the same general sense of delight users feel for a product, manifested in different ways. To that end, virality and retention generally exhibit a positively correlated relationship: products that users are inclined to return to over a long period of time are also likely to be products that users invite others to join.

At a conceptual level, the logic driving the one-directional relationship between virality and retention (i.e., strong retention increases the potential for strong virality) is fairly straightforward; users for whom a product exhibits overwhelming appeal retain with that product longer than users who don’t, and thus they are presented with more opportunities to invite others into the product than non-retaining users.

In practice, this relationship may not manifest; some products are designed for individual, private use and are not conducive to viral growth no matter how strongly users feel about them, and some products do not competently implement the means for viral dissemination. But at the core of any product’s use profile sits the single characteristic—delight—that contributes to both virality and retention metrics.

When these two metrics do not correlate positively—when a product exhibits strong virality but weak retention or vice versa—the positive effects are often neutralized (or even become liabilities). For instance, present strong virality but low retention, a product’s user base may grow rapidly but just as quickly diminish as users churn out of the service. This scenario is problematic because churned users must be reacquired, which is difficult when they have already abandoned a service.

In the opposite situation—a product exhibits strong retention but low virality—the growth costs are not offset by the accumulation of viral users, which in some cases may completely reduce the product’s ability to expand. A product cannot capitalize on either strong virality or retention unless the other metrics group exists at some minimum level.

The product development cycle should accommodate this relationship by focusing on the core experience that drives delight, rather than addressing an individual metrics group. Tactics used to boost either retention or virality—but especially virality—often detract from the product experience and appear contrived.

If a product doesn’t exhibit virality, users fundamentally oppose revealing their use of the product because the product isn’t useful to them, the product doesn’t meet the quality threshold required to make a recommendation, or because they simply don’t want people to know they use the product. None of these reasons can be attenuated by an additional viral mechanic; they must be addressed with deep core changes to the product experience.

Likewise, retention metrics underscore fundamental product delight; addressing a specific retention-day drop with a new feature or gimmick won’t fix the structural shortcomings that cause users to leave a product and not return. The underlying reasoning users employ to justify leaving a product (either for another product or because the product’s use case is simply irrelevant to them) won’t be dismantled with artificial auxiliary product features.

It is true that retention and virality can be improved incrementally through product features that don’t augment the product’s core use case. But for these metrics to contribute to the user base meaningfully and sustainably, both must exist at a minimum threshold that lets a virtuous cycle develop, where retention allows for greater virality and greater virality introduces more users to the product. If one of the groups of metrics doesn’t reach the minimum threshold, this symbiosis can’t occur and, as pointed out, the relationship could actually have negative effects on the long-term viability of the product.

Until both the retention and virality metrics groups reach a minimum level, product development shouldn’t be directed at features that address either one specifically; rather, the fundamental use case should be addressed and iterated upon, with the goal of improving all of the minimum viable metrics. Feature development undertaken to specifically address virality or retention, before the product has achieved a minimum level of performance, will likely achieve only short-term, negligible gains.

Signal versus noise

In-product viral invitations are often seen as a well that can be infinitely drawn from in pursuing virality: invitation mechanics are easy to implement, present massive potential for virally acquiring new users, and are easy to instrument and track. But invitation mechanics can be forced onto users to an extreme that not only negates their effectiveness but also produces unfavorable results through overexposure and a tone reminiscent of spam. The extent to which viral invitation mechanics are relied on should be considered (and, ideally, tested) within the full scope of their possible consequences.

One reason viral invitation mechanics should be implemented cautiously is psychological; given that a product’s value proposition should be self-evident, products that frequently encourage users to make use of invitation mechanics may appear adversely selected for (because products of obviously high quality don’t need to remind users to tell others about them). This realization could discourage users from sharing their use of the product at all, because they could feel like marketing agents rather than customers and because they might not want to risk their reputations on promoting a seemingly substandard product.

Copious and aggressive invitation mechanics render a strong signaling effect on the user base that contributes significantly to user sentiment for the product. But perhaps more importantly, they have a strong signaling effect on potential future users—those on the receiving end of the invitations. Numerous invitations to a service, especially from the same user, are understandably interpreted as spam and are ignored. Belligerently pestering would-be users with invitations to join a product can create permanent non-users or, worse yet, negative messengers and product denouncers.

These psychological effects are difficult, if not impossible, to offset. Users who have made determinations about a product, especially a product they perceive to have spammed them, will not likely change their minds about that product without significant input from existing users. And, given the presence of such enthusiasm, the product would almost certainly be better off if that enthusiasm were directed either toward potential users who have never been exposed to the product or inward to the existing user base, thereby increasing engagement. The effort required to convince potential users that a product is worth investigating after they have mentally dismissed it is monumental.

The second reason in-product invitation mechanics should be implemented methodically and with deliberate restraint is that users can engage with only a limited number of contacts at once in any service. When users exercise a viral invitation mechanic to lure a large number of new users into a product, the existing users’ capacity for communicating with their contacts strains with each new user who accepts the invitation. The users who accept the invitation, join the service, and are then seemingly ignored by the person who invited them may feel neglected or believe that the product isn’t functional; in either case, this effect increases the likelihood that they churn.

A low number of highly engaged and socially active viral users can be worth more than a high number of viral users who feel isolated. A limit on the number of viral invitations that a single user can send serves to ensure that users acquired by those invitations are given the fullest attention possible by the users who invited them, especially in social products or products where users are aware of their contacts’ actions.

Additionally, when a person knows an invitation was one of a limited number that could be sent—he or she was allocated a scarce resource—that person is more likely to attach value to the invitation than if he or she knows invitations are limitless. The same is true on the current user’s part when selecting invitation recipients: having only a limited number of invitations, the user is more likely to select the most appropriate contacts, personalize the invitation (to the extent made possible by the product), and potentially follow up on the invitation. This mentality substantially increases the likelihood that an invitation converts to an acquisition.

Viral invitations are well structured as mechanics that exist as passive, permanent fixtures of the user interface in terms of product implementation; aggressive pop-ups and entreaties to invite friends appear desperate and unfavorable. Invitation limits may be either explicit (e.g., the user is directly told that the in-product mechanic may be used to send a specific number of invitations) or enforced through the layout of the mechanic (e.g., the mechanic has only three fields for inputting contact email addresses).

When the user must proactively seek out an invitation mechanic, that mechanic is more likely to convert than mechanics that have been forced on the user. Given the negative effects of a rejected invitation—appearing as spam, the impression of adverse selection, etc. —the number of viral, in-product invitations should be kept to a minimum, filtering out invitations from all but the most enthusiastic and dedicated users.

Quantified virality

As noted earlier, virality is notoriously difficult to measure, much less model for predictions. A global k-factor may be easily calculated, but it is not necessarily forward-facing or predictive; the growth in one period of the non-paid user base can be influenced by any number of factors and may not presage future growth trends. And calculating the effects of viral mechanics may be impossible without an accurate mechanism for attributing a new user to a specific viral invitation.

That said, freemium product virality is important enough, in light of the cost of user acquisition and the scale needed to fully exploit the business model, that an informed attempt at modeling virality provides some value, even when it isn’t entirely accurate. One reason for this value is that understanding the dynamics of virality can help improve its effects, even if that improvement can’t be accurately measured. In an environment where user base growth is dependent on either viral reach or paid acquisition, modeling virality under imperfect information and assumptions can still help the product team achieve non-paid user base growth.

A second reason even imperfect virality models are valuable is that they can help draw attention to shortcomings in retention that otherwise might be overlooked. When virality is high but retention is low, the user base experiences wild swings in size as users enter and churn quickly. This phenomenon is better to witness under the auspices of virality, which provides free users, than under paid acquisition. Viral models can serve as warning signals that the product is not ready to effectively utilize paid acquisition, thus sidestepping potentially ineffective expenditure.

Virality models are predicated on a limited number of variables, all of which are conceptually simplistic. Likewise, they can be built in standard desktop spreadsheet software and don’t require sophisticated statistical packages to deploy. That said, the concept of virality can be hard to grasp because of its compounding effect: it is a challenge to explain how viral growth takes shape over multiple periods. For this reason, virality models may be the most cumbersome to use in distributed decisions because, although they might exist in spreadsheet format, they’re seen as a form of black box argumentation. The goal of this section is to outline the proper structure of a virality model and provide a framework for discussing virality in a way that can be easily digested.

Viral periods

The effects of virality are more or less fluid; users don’t invite other users into a product by a prescribed schedule, but they do whenever it is convenient and sensible for them to do so. However, for practical purposes when calculating virality, it is analyzed in terms of periods. A period is any amount of time between two endpoints over which user base growth is investigated; the denomination of the period (days, months, weeks, etc.) depends only on the purpose of the analysis.

Viral periods are generally described in the abstract, as opposed to a predefined amount of time, because virality is a key factor in any robust revenue model, and revenue models can be necessitated over any number of arbitrary timelines (such as quarterly, for financial reporting). But in practice, the virality period is often designated in days: that is, one period is equal to one day.

As in calculating interest on a principal sum of money, calculating compounded virality is more precise the more granular the measurement period, which is another reason why days are preferable to longer periods. Additionally, calculating compounded viral user base growth is straightforward and more or less easy to explain on a daily basis.

Virality is expressed as the growth of the user base relative to an initial group of users who adopt the product in the same period (where period length is flexible and could be one day, one week, one month, etc.). Cohort analysis is frequently used throughout freemium product analysis; the calculation of virality is one such analytical implementation. When calculating the effects of virality from period to period, user base growth is attributed to the start of the second period; that is, new users adopted virally from an initial cohort of users in period one are considered to have joined the product at the start of period two.

The compounding effect of virality—the phenomenon of users virally acquired themselves inviting other users into a product through viral channels—creates the need for identifying viral cohorts by their order. The cohort acquisition order relates to how far away, in periods, a cohort is from the origin cohort (which is the cohort that spawned the viral growth). Each subsequent order of cohort—second order, third order, fourth order, etc.—is a product of the k-factor and the number of users in the cohort preceding it.

For example, if the origin cohort contains 100 users and the global k-factor is 20 percent, then the first-order cohort (in period two) is composed of 20 users. The second-order cohort, or 20 users producing a new viral cohort at a global k-factor of 20 percent, would be four users; the third-order cohort, or four users exhibiting virality at a rate of 20 percent, would be 0.8 (truncated to zero). The total size of the user base at the end of the fourth period would be 124, representing growth of 24 users through four periods of virality.

When modeling virality in a spreadsheet, the origin cohort usually sits in the top left-most cell, representing the number of users at period one. The y-axis of the spreadsheet represents ordered cohorts, and the x-axis represents periods; as ordered cohorts are added, they are inserted into the model downward and diagonally to the right. Figure 7.12 demonstrates the example discussed here, with an origin cohort of 100 users and a global k-factor of 20 percent growing through period four, as modeled in a spreadsheet.


FIGURE 7.12 Viral growth for an origin cohort of 100 users and a global k-factor of 20%.

The growth of 24 users (cell B6 through cell F10, rounded to a whole number) is larger than the global k-factor times the origin cohort (20, or 20 percent times 100). This is a function of the multi-order effect of virality (in other words, that virally acquired users can produce virally acquired users).

In practice, growth as modeled in Figure 7.12 is improbable; this model depicts a situation wherein users capture total virality in their first period of interaction with the product (i.e., the k-factor is manifested in the period after adoption). Delayed virality over a certain length of time after the user’s initial adoption of the product is more realistic, because users must become acquainted with the product before feeling comfortable inviting others into it and because viral invitations cannot always be immediately accepted. This delay is implemented into the virality model via a concept called the virality timeline.

The virality timeline

The timing of acquiring users through viral channels, whether they are acquired in-product or through interpersonal recommendations, is an important consideration when modeling the effects of virality. One reason for this relates to the discussion of the time value of money in Chapter 5: the sooner users adopt a product, the sooner they may begin contributing product revenue, which can be reinvested in further product development, marketing, or savings. Whatever the case, revenue generated sooner is preferable to revenue generated later.

But a second element adds complexity to the timing of virally acquired users; in highly viral systems, when virality is “front-loaded,” or takes place primarily at the beginning of a user’s tenure, the compounded effects of virality manifest through user base growth near the start of the original user’s tenure. This accelerated growth curve can be a boon to certain products—especially those that rely on network effects for viability, such as social networks—but it can also present scaling and user base consistency issues.

A rapidly growing user base may present technological and operational challenges to product support structures in terms of scaling, which could prove detrimental to the user experience and thus undermine the value of accelerated growth. Furthermore, a poor early user experience can repel users who may otherwise find the product useful and appealing under circumstances of slower growth, and those users may not return.

And with respect to user base consistency, a user base that grows quickly is likely to experience volatility as user cohorts churn out in large numbers at the end of their lifetimes, even with a gently declining retention curve. This presents problems with network efficacy and reliability for users, but it also renders user data difficult to interpret, as instability in the size of the user base skews the effects of real phenomena. This volatility also poses problems for forecasting future performance: when a massive cohort falls across a broad retention profile and churns out in numerous and appreciably large blocks, estimating the lifetimes of those disparate user segments is cumbersome.

Taking these factors into account, the virality timeline, which is the weighted-average number of virality periods over which virality manifests, must contribute to any model of virality. A virality timeline is calculated by assigning weights to each period’s contribution to total virality relative to total virality and reducing those weights to a single number. The formula to calculate a weighted average timeline is expressed in Figure 7.13.


FIGURE 7.13 The calculation for a weighted-average timeline as used in the virality timeline.

The virality timeline is denominated in periods and describes the average length of time over which all virally acquired users who are invited into the product by a single user adopt the product. For example, if one user virally acquires six users over three days through in-product viral invitations and interpersonal recommendations (where three users are acquired on day 1, two users are acquired on day 2, and one user is acquired on day 3), then the users’ virality timeline is 1.67 days, as illustrated in Figure 7.14.


FIGURE 7.14 The virality timeline should be weighted to more accurately reflect when users adopt the service after being invited.

The virality timeline should be calculated for whatever user segment is being analyzed; that is, the virally acquired users for a given cohort should be aggregated on a timeline, as opposed to averaging the individual virality timelines for the users in that cohort (which would produce an average of averages). Once calculated, the virality timeline should inform a model of virality by serving as the period into which all virality is divided.

The model explored in Figure 7.12—a product with a global k-factor of 20 percent and an origin cohort of 100 users—can be extended by incorporating a virality timeline of three periods, meaning the global k-factor is distributed evenly across the three periods following product adoption. Since the global k-factor is applied to the origin cohort as a universal total and not a per-period amount of growth, the first-order virality can be expressed across the three succeeding periods, with the rest of the ordered virality cascading in a similar manner, as shown in Figure 7.15.


FIGURE 7.15 Viral growth in a spreadsheet format, incorporating the virality timeline.

Note that virality is distributed across three periods for the first-order cohort (row 8) but five periods for the second-order cohort (row 9); this is because, as the cohorts progress, virality timelines for successive cohorts overlap. In other words, in period 5, viral users are being created for periods 2, 3, and 4 across the first-order cohort virality timelines. This is manifested in cell F9 from Figure 7.15 by using the formula in Figure 7.16.


FIGURE 7.16 The formula for calculating viral users acquired for overlapping virality timelines.

As the cohorts in Figure 7.15 advance, the total length of each row increases by 2: the length of the virality timeline, which is 3, minus 1. The logic of this structure is illustrated in Figure 7.17.


FIGURE 7.17 A visual depiction of viral growth across cohorts.


Up to this point, virality has been described as a powerful force that, once achieved, is propelled indefinitely by its own momentum. But the reality is that an upper bound exists on the number of users to which any product can offer a meaningful experience. Even the most viral product has a threshold past which adoption will slow as a result of increasing exposure within the demographic matrix for which it is most appealing. This threshold is known as saturation, and it represents the maximum reach of a product through viral means.

In terms of virality, saturation is defined as the scope of market penetration a product can accomplish before no new users can be expected to adopt the product from a viral campaign; it quantifies the boundary of growth that a product could expect to experience virality. And while this number may be difficult to estimate with any accuracy, it is an important component of any virality model because it serves to ground viral assumptions in reality.

Projections of virality don’t contribute to a realistic picture of potential product growth if they are not constrained by any limits. Even products with universal consumer appeal slow in growth as they reach large levels of market penetration. A saturation metric tempers a model of virality by serving as a sort of gravitational force exerted on viral expectations.

The easiest way to implement saturation into a model of virality is to consider it a limiting factor based on the percentage of the total user base already reached through viral channels. That is, at any specific point, the potential viral growth rate is reduced by the extent to which the user base has progressed toward the saturation point. As a concrete example, given the conditions laid out earlier, of a product with an origin cohort of 100 users, a global k-factor of 20 percent, and a virality timeline of three periods, a saturation level of 100,000 users would have the effect on first-order viral growth as described in the equation in Figure 7.18.


FIGURE 7.18 As the user base grows, virality should be adjusted by diminishing potential future growth given a saturation point.

In the figure, the ratio of users within the saturation threshold yet to be reached is expressed for periods 2, 3, and 4 in the third element in brackets as image. This ratio is used to reduce the extent to which virality can still occur. As the number of users already reached grows, the ratio increases, and when it reaches 1, the saturation equation resolves to 0, resulting in a viral growth value of 0.

The practical effect of saturation on virality is that growth slows as the number of potential new recruits into the product decreases. This growth limitation forms a sigmoid shape to the user base curve over time: as virality takes root, the curve will inflect upward, and as the product reaches saturation, the curve will inflect downward, as in Figure 7.19.


FIGURE 7.19 A graph of user base growth assumes a sigmoid shape as it reaches its saturation point.

Estimating saturation is a valuable exercise even outside the context of virality. Another way to describe saturation is market size, and it is a relevant metric at the planning stage of any product. Saturation represents a product’s total, realistic potential user base; it is a fundamental component of a revenue projection or viability study. Quantifying saturation involves estimating the size of the demographics to which the product is likely to appeal. When a product is location-dependent—a city-centric directory service, for example—saturation starts at a concrete value, which is the population of the geography under consideration. In such cases, estimating saturation as a “top-down” approach is sensible: the starting value is the relevant population metric, which is decremented as specific demographics are discounted as being inappropriate for the service.

In other cases, where a product’s relevant users aren’t defined by a discrete population metric but rather by how well the product fits with an existing need, a “bottom-up” approach to estimating saturation may be more appropriate. This involves estimating the size of various demographics for which the product is likely to hold appeal and adding those population sizes together to form a total potential user base.

Estimating saturation may appear as an impractical intellectual exercise—and to some degree, it is—but virality itself is amorphous and difficult to concretely predict. While the estimation for saturation is not exact, it can be used in a methodical process that at least places restraints on a force that otherwise can be modeled to illogical extremes. Endless virality is not a realistic assumption, but given its significance, a firm must attempt to incorporate it into a model of user base growth. Saturation provides a parameter to any such estimation that serves to contain its aspirational nature.

Modeling viral growth rates

Building a reasonable model of virality first requires incorporating the component elements of viral growth discussed so far into a model of viral growth rates that can be used to drive a model of absolute user numbers. This viral rates model contains ordered viral growth values, which are the rates of accumulation of the user base per period. The viral growth rates model is a middle layer of logic that preempts the virality model to ease interpretation and to allow for making modifications to the viral values without having to edit the formulas in the viral rate worksheet.

Developing the viral rates model in a spreadsheet begins with recognizing the virality inputs—the global k-factor and the virality timeline—as well as determining the number of ordered cohorts over which virality should be tracked. The idea behind tracking ordered virality is that the members of the first-order cohort of virality, or the cohort invited to the product directly by the origin cohort, will themselves introduce users to the product through viral channels, and so on.

The purpose of the viral rates sheet is to answer the question, “When a user joins the product through a non-viral channel, how does virality cascade from that user over a finite number of cohorts to contribute to the overall growth of the user base?” The answer is a discrete number representing the total number of users over whatever number of cohorts are tracked, whom any one user virally introduces to the user base. This number is different from the k-factor, which represents the number of users any one user will directlyintroduce to the product. The viral rates sheet output is the total number of users a given user indirectly or directly introduces to the product as a result of compounding virality.

Tracking virality over a predetermined number of ordered cohorts isn’t strictly necessary; when the global k-factor is less than 1, the total number can be calculated using a converging infinite geometric series, as expressed in Figure 7.20. In other words, the formula in the figure gives the total number of users, aggregated over an infinite number of ordered cohorts, whom any one user will introduce to the product, given a global k-factor of less than 1. (When the k-factor is greater than 1, the values will not converge, since they approach infinity.)


FIGURE 7.20 Calculating the total number of viral users when k-factor is greater than 1.

But the purpose of a model is to see the progression of a metric over time and to use those time-based values to estimate something else: typically, revenue. Therefore, the total number of eventual users isn’t as important as the growth of the user base on a per-period basis. The viral rates sheet, then, should be laid out as a matrix, formatted as shown in Figure 7.21, with the number of periods being tracked running across the x-axis at the top and the ordered cohorts running down the y-axis at the bottom.


FIGURE 7.21 The virality rates worksheet.

Cell B6 represents the origin cohort, or the cohort of users introduced to the product via non-viral means. This cohort is expressed as 100 percent because it symbolizes the entirety of the origin cohort; by expressing the values in the spreadsheets as proportions of the origin cohort, the end result can be multiplied by an absolute cohort size to produce a value denominated in users.

Row 7 represents the first-order cohort, or the first group of users introduced to the product through virality by the origin cohort. The 6.7 percent values are derived from the local k-factor of 20 percent multiplied by the origin cohort of 100 percent, then divided by 3 (because virality is manifested over the three-period virality timeline: image).

Row 8, the second-order cohort, follows the same pattern: 6.7 percent is multiplied by the local k-factor of 20 percent then divided by the virality timeline of 3 to illustrate the percentage of users (relative to the origin cohort) virally introduced to the product by the first-order cohort. This row extends for five cells, as opposed to three, for reasons discussed earlier: the virality timelines for the first-order cohort overlap. The third-order cohort, in row 9, progresses similar to the second-order cohort.

Row 10 is where the virality percentages are aggregated vertically by period; cell L10 is the sum of these aggregates and represents the total virality coefficient, or the number of subsequent users virally introduced to the product by each user. Subtracting 1 from the value in L10 (in other words, backing out the origin cohort) produces a value of 24.8 percent, which represents the local k-factor compounded over two ordered cohorts.

The numbers in Figure 7.21 are similar to those in Figure 7.15, with the key distinction that, in Figure 7.21, the values are percentages. This is an important point; the virality rates spreadsheet is an abstraction, meaning it represents user growth in percentage. The virality coefficient in cell L10 can be multiplied by any number of users joining the product in order to produce the total number of resultant users who are introduced to the product over 10 periods across three viral cohorts.

Increasing the complexity of the model by extending the y-axis with additional cohorts and the x-axis with additional periods increases the virality coefficient but by rapidly diminishing amounts.

Building the viral model

The model of viral growth rates provides a timeline over which viral users are adopted relative to an origin cohort. The viral model should provide a timeline of total users in the user base, with each period in the model introducing a new origin cohort as well as viral users originating from previous origin cohorts.

The virality model is structured in a spreadsheet in the same general format as the virality rates model: the y-axis represents ordered cohorts and the x-axis represents periods (again, usually days). The conceptual difference in the models is that the viral growth rates model represents the viral growth rates from a single cohort over the course of a predetermined number of ordered cohorts; the virality model aggregates these growth rates per-period into absolute user numbers. In other words, the virality model outputs the actual number of users who can be expected to join the product through viral channels per period.

The virality model does not take the downward-sloping diagonal line shape of the virality rates model because it is not following the viral growth of a single cohort. Instead, it takes the shape of an upside-down plateau, as the number of users representing each ordered cohort is stacked in reverse order from the origin cohort. The first row of the virality model is thus the origin cohort for that day, or the number of new product users acquired through non-viral channels.

The number of periods covered in the virality model should span whatever length of time is being considered for projecting the size of the user base. In many cases, this will be one quarter, two quarters, or a year. When the period length is one day and the timespan of the model is one year, the width of the matrix should be set to 365 columns. The height of the matrix should match the height of the virality rates model matrix. The first row should be filled with the number of users in each origin cohort on each day. In some models, this number might be constant, representing an expectation of a constant number of new users from organic and paid channels. In other models, the number might decline over time, spike intermittently to reflect media coverage, or assume any number of shapes that new user values can follow over the lifetime of a product.

On the viral growth worksheet in Figure 7.22, each row after the origin row reflects the number of users from the nth-order cohort being added on that day. The value in each cell is derived from the growth rate in the viral growth rates worksheet, based on the ordered cohort and period of that cell; the ordered cohort number should be taken from the current row and looked up in the viral growth rates lookup table.


FIGURE 7.22 Reconciling the viral growth rates sheet with the virality sheet.

The value of cell C4 in Figure 7.23, which represents the first-order viral cohort invited to the product from the origin cohort in period 1, is the product of two numbers: the size of the origin cohort and a viral growth rate. The first number is looked up in the viral growth worksheet and found in cell B3. The second number is looked up in the virality rates worksheet and is based on period and cohort number; in this case, the value is found in cell C7 on the virality rates worksheet, representing period 2 for the first-order cohort.


FIGURE 7.23 The formula for cell D4 in Figure 7.21.

Similar to the way the virality rates worksheet is organized, as cells progress to the right in the viral growth worksheet, the virality timelines of preceding periods overlap and must be summed. For example, the formula in Figure 7.23 that produces the value in cell D4 from the viral growth worksheet in Figure 7.22 adds two values together: the period 2 virality for the origin cohort in cell B3, and the period 1 virality for the origin cohort in cell C3. This pattern continues along the row; because the virality rates sheet extends to 10 periods, the number of terms added together for each ordered cohort in the virality worksheet increases until it reaches 10. (The number in the right-hand-most column represents the overlapping virality for the previous 10 periods.)

At this point, the viral growth worksheet represents the absolute number of users acquired through viral channels, by ordered cohort, per period, as illustrated in Figure 7.24. The bottom row of the matrix contains the sum of each column, representing the overall per-period count of new users, or the origin cohort plus the users acquired virally in the ordered cohorts tracked. But as discussed, these per-period sums must be reduced by the saturation threshold, given the cumulative users recruited into the product up to that point. As this cumulative number increases, the proportion of potential product users who have yet to be reached decreases.


FIGURE 7.24 A complete five-period viral growth worksheet.

To implement saturation, the sums at the bottom of the columns must be reduced. The formula in cell F7 from Figure 7.24, which holds the total number of saturation-adjusted users acquired in period 5, is given in Figure 7.25, with a saturation level of 100,000 users.


FIGURE 7.25 The formula for cell F7 in Figure 7.24.

Note that reducing the sums, as opposed to the component rows, may lead to confusion. Reducing each row of the period columns with the formula exhibited in Figure 7.25 would produce a more straightforward and intuitive sum at the foot of each column, but it would also render the formulas within the virality matrix much harder to read. The updated virality worksheet, accommodating for saturating, is depicted in Figure 7.26.


FIGURE 7.26 The updated virality worksheet, with per-period sums reduced by a saturation ratio.

The summed columns in Figure 7.26 represent absolute, per-period numbers of users acquired virally. These values can be incorporated into a broader model of the user base (which would need to include returning users to be complete) to present a considered, if somewhat static (and divorced from exogenous effects) picture of the product’s user base over time.

Engineering virality

Although it is an inexact science, measuring and quantifying virality may be the easiest aspect of managing a product’s virality. Virality is a fickle beast, and often, overt, contrived efforts at instilling virality in a product are fruitless. Moreover, they may alienate the user base, instilling in it a sense that, rather than being served by the product, the product exists to be served by them. A company should not assume its users to be marketing channels; when a user contributes to a product’s marketing efforts, the user does so as an enthusiast, not a customer.

But, given how fundamental virality is to the freemium model (short of a substantial marketing budget, it is the only viable path to massive scale), it is essential to control, or at the very least create, the opportunity for virality. While virality can’t be expertly and predictably willed into existence, a broken virality framework—overwrought viral features, aggressive invitation mechanics, or an isolated user base—won’t contribute to growth. Over the course of product development, a company must strike a balance in pursuit of virality; it must invest sufficient time in developing a virality architecture with the potential to provide value, while not over-investing time into viral features for which return cannot be reliably estimated.

To that end, engineering freemium virality is accomplished with the same conceptual logic as engineering monetization: provide users with the tools and latitude to engage, and the most zealous and enthusiastic users will do so willingly.

The viral product

At the heart of any viral growth pattern is a product that inherently invokes and rewards connectivity. Products that are useful only at the individual level can be shared and enthused upon, but long-term, sustained viral growth is a function of the fundamental necessity and advantageous benefits of a product network.

One means of achieving sustained growth is through in-product social features that add value to the core of a product experience. These features generally fall into three categories: collaborative, which allows for shared product use; competitive, which gives users opportunities to rank themselves against other users; and communicative, which allows for users to share and discuss ideas. These three feature categories broadly form a sense of community within a product’s user base. When users believe that their membership to a community augments the product experience, they are more likely to recruit for a product than if overwhelming delight is the requirement for such proactive, enthusiastic sharing.

The operational concern in implementing features within this scope, however, is experience enhancement: if the features don’t truly improve the experience but merely facilitate potential virality, then they aren’t likely to engender significant user base growth and could instead alienate users by diverting attention away from the product’s fundamental use case. Ill-conceived social functionality, often developed late in the design process to address a specific lack of virality, may potentially cause more harm than good.

The success of social features is often directly and positively correlated to how early in the development process they are conceived. A product is either social in nature or it is not, and the presence of social features does not form the exclusive basis of this distinction. A product is likewise either viral or it is not; that is, it either supports sustained, continued virality through its fundamental use case, or it benefits only from short-term, tenuous viral growth through superficial invitation mechanics.

The viral product is a product with a fundamental use case designed with the objective of collaboration, competition, or communication in mind. Without at least one of these objectives driving the user experience, a product cannot be viral; rather, it can merely benefit from specific viral mechanics and user excitement. But when the core experience is designed with one of the three objectives as the backdrop, users are compelled to recruit additional users into the product to enhance their own experience.

The risk with such a tack, of course, is that the user doesn’t recognize the value in the product from the outset, or that the user simply has no network from which to draw. A viral product, or a social product, is only as effective as the user’s network is available. If the product does not offer an experience that can be undertaken at the individual level, then it won’t inspire virality and won’t capture the interest of users acquired organically or through paid channels.

The prototypical categorical example of such a product is the social network, which is useless to a user without a minimal, relevant group of people with whom the user can connect. A product may be viral at its core, but when network effects are a product’s basic experience—the product is merely a tool that can extract value from a preexisting network—then the product will fail to gain traction from users without those networks.

Such a design decision—whether to architect the product as inherently viral or as an individually relevant product with attendant virality features—requires careful consideration of the demographics the product will appeal to. This design decision may also require a multi-staged approach to roll out at launch: while users may not have preexisting networks relevant to the product to begin with, a product might facilitate the creation of new networks through discovery and interaction.

In such a case, the product may be designed with two launch stages in mind: the first introducing a personal experience and establishing the fundamentals of the use case, and the second reconciling those use cases into a network that enhances the experience for all users. Such an approach sets the stage for collaboration, competition, and communication from the initial launch, but it manifests the product as an isolated experience; in the second stage, social features are added as an experience layer on top of the core functionality of the product. In essence, this strategy lays the foundation for a network to form around a notional nucleus in the first stage and then introduces the in-product infrastructure for that network to take shape in the second stage. But this essentially requires the creation of two products, and the development of a product predicated on the adoption and success of a subsequent product invites a high degree of risk.

At the conceptual level, the viral product is subject to a harsher set of preconditions to success than is a product that can be used by a user in a vacuum; the viral product’s use case must not only meet a real need, but a preexisting network must also serve it and manifest that need. And while the viral product benefits from inherent virality (and thus fairly effortless user base growth), it also risks a structural lack of potential for momentum.

Viral networks

The network connecting people is the propagation medium for any viral campaign. Networks are sometimes based on physical boundaries, such as location-based services, but often they’re built around loose affiliations of commonality, such as online communities or forums. And while many people belong to a number of networks and can thus propagate viral products on a number of fronts, most networks are fairly isolated and limited; notions and sentiments cannot bridge the void between one network and another without an impetus.

While saturation represents the total potential market for a product based on demographics and tastes, the pace and extent of virality is set by the networks through which virality can be transmitted. Understanding which viral networks potential users participate in and how those networks can be penetrated is an important part of developing and incorporating viral mechanics. Likewise, for interpersonal product recommendations to take root and produce appreciable virality, a product’s users must have interpersonal networks.

Viral networks effectively act as restraints on the propagation routes of viral campaigns; the natural silos potential users are segmented into prevent viral campaigns from being transmitted evenly throughout the entire population of users a product might appeal to (its saturation population). This restriction manifests in more money needed to construct virality features—each network may have to be accommodated—and in diminished returns on the effects of compounding virality. In other words, since each viral network must be seeded with an initial set of users the virality will propagate from, more effort is required to manifest virality than if the entire saturation population could be reached within one network. This limitation requires consideration during the product development cycle and the product’s management after its launch; small networks may provide less value than larger ones do because compounded virality is not as likely to occur. If the saturation population is seen not as a monolith but as a collection of networks through which virality can be propagated, then the largest and most interconnected networks should be given priority within the product.

Additionally, seeding a small network with users through paid acquisition may cost as much as seeding a large network, yet the reach of a smaller network is, by definition, more limited. Thus, the commitment to pursuing virality on a specific network should take into consideration the total size of that network; viral mechanics require development time and must be implemented sensibly, with an eye toward return on investment.

Similarly, networks must be evaluated by the potential revenue quality of their users, or the demographic attributes of users on a network related to likelihood to monetize. Some networks may provide incredible viral growth and distribution at a large scale but represent a fundamental, measurable unlikelihood to deliver revenue. Networks exhibiting a homogeneous specific demographic attribute, such as age, are the most likely ones to sit at an extreme of revenue quality. The degree to which such networks can contribute users with disposable income should be evaluated before viral mechanics tailored to those networks are developed.

That said, the opposite may also be true: a network’s characteristics may provide favorable conditions for virality that serve to reduce the effective cost of a marketing campaign. Networks composed of highly connected early adopters, although generally small, are fertile ground for the propagation of new products and services through viral channels. These networks also tend to feature users with large amounts of disposable income and lower barriers to monetization (such as the lack of a stigma around purchasing virtual products or services). In such cases, however, impressive virality should not be mistaken for widespread, universal appeal; early adopters tend to exist on a separate plane of product appreciation than does the general public, and their behavior with respect to new products can just as easily expose a passing fad as affirm a long-term trend.

Geographic networks may be the most rigidly isolated and therefore the least cost-effective to penetrate, especially at the city level. Products related to a city, such as city-level directories or social networks oriented toward local activity, suffer from strict virality barriers that can vastly increase the costs of marketing. Products of this nature require extreme virality to offset the effort required in seeding a user base at the local level. Geographic boundaries are perhaps the most cumbersome constraints on viral growth; local products may face the highest marketing costs, as each limited market contributes to marketing overhead.

Increasing viral invitations

One strategy for increasing virality is to simply increase the way that virality throughput is achieved, which is the transmission of viral invitations. As discussed earlier, the effects of increasing the volume of invitations sent are not entirely beneficial: highly prolific viral mechanics may be indistinguishable from spam, despite the best of intentions on the part of the developer. And potential users may interpret spam from a product as a signal that the product is not legitimate or is of poor quality.

The appearance of spam can be mitigated even when viral invitations are made frictionless. One method of doing so is to simply control the viral invitations any given user is permitted to send, by channel, recipient, or otherwise. By preventing a user from sending the same viral invitation to the same user or group of users more than once, the product can still maintain a high number of viral invitation mechanics while dodging potential misuse or abuse. Diversifying the format and transmission channels of viral invitations diminishes the likelihood of those invitations being considered spam, while still providing very enthusiastic users with a platform for broadcasting their use of the product. In essence, broadening the invitation channels shifts the onus of responsible use from the product to the user.

One way to increase the number of viral invitations that can be sent is by incentivizing the mechanic with an explicit reward for the user upon sending a viral invitation. This is often done by providing premium functionality for free in exchange for sending the invitation; for example, a freemium product offering premium services with a subscription model might offer one free month of access in exchange for inviting someone to the service.

This approach is fairly simple to implement and carries with it the positive signal of the initial user actively attempting to unlock further functionality from the product. On the downside, users are not necessarily incentivized to choose appropriate recipients, as they are rewarded for the invitation, not the conversion.

Many products therefore reward the invitation only upon success; that is, the user collects the reward when recipients of the invitations adopt the product, which is often defined as registering with the product. This approach may reduce the total number of invitations sent, but it likely results in those invitations being more thoughtfully sent, thus yielding more new users. Many times these mechanics reward the invitation recipient, with the goal of inuring both users to the premium feature in order to encourage continued purchases.

Another approach to incentivizing the invitation mechanic is to gamify the use of the product. Gamification is an approach to product design that rewards users with designations, titles, and product-specific, trophy-like keepsakes for meeting specific engagement benchmarks. Gamification fosters virality by giving users an interesting, often entertaining focal point through which to broadcast their use of a product. Gamifying the product incentivizes users to announce the achievements they have unlocked, not just to other existing users of the product, but to outsiders who may feel compelled to investigate the product for fear of being excluded.

Increasing the velocity and volume of viral invitations sent is not, of course, a goal unto itself; it is rather the means by which virality can be increased, and if conversion dips substantially as the number of invitations sent increases, the net effect could be neutral or negative. Making incremental improvements to existing viral mechanics may produce better overall virality than does adding new mechanics alongside the existing viral toolkit, especially if the viral infrastructure is varied and not overly aggressive.

Once a product has launched, incremental, individual performance improvements to a large number of disparate viral mechanics is more likely to add to increased virality than a single new mechanic is. The most enthusiastic users (for whom the product should be selecting in crafting viral mechanics) probably already understand existing mechanics, so simply making the mechanics easier and more entertaining to use could produce instant results. Users may also feel that the addition of new viral mechanics reduces the users’ value to that of a marketing channel; post-launch, the product should take pains to ensure that enthusiastic users don’t feel exploited. Likewise, users are more committed to products they deem as being committed to them, and the use of development time and resources on features that do not enhance the user experience detracts from that sentiment.

Increasing viral conversions

The viral invitations conversion rate represents the second set of territories on which virality can increase. As opposed to increasing the volume of viral invitations sent, increasing the invitation conversion rate does not introduce attendant risk: by improving invitations so there are more conversions (given that the number of invitations remains constant), the product does not produce new tension by looking like spam or by changing the perception of user base exploitation.

That said, increasing the conversion rate of viral invitations may be less easily achieved than increasing the number of invitations sent. A/B testing is the principle approach to improving the likelihood of producing a new user with a viral invitation. As discussed earlier, A/B testing evaluates various nuanced versions of content based on their conversion rates. And while A/B testing does allow for specific invitation metrics to be optimized, the fertility of this source of increased virality erodes after a relatively limited period of time as the easiest and most obvious tests are conducted.

A second approach to increasing the conversion rate of viral invitations is incentivizing adoption by rewarding either the person sending the invitation, the person receiving the invitation, or both. When the sender is rewarded only upon the receiver’s conversion, the sender is incentivized to select the people most likely to convert. And when the receiver is rewarded for joining the service, that person is more likely to act on the invitation. As in the tactic employed when attempting to increase the number of viral invitations sent, the reward given upon conversion is usually free access to premium functionality.

Establishing an element of functionality that is premium in name but offered as a reward for conversion may effectively reduce the size of the product catalogue if the viral mechanic through which the reward is received is used frequently. This is why rewards are usually time-constrained for subscription services offering, for example, one free month of premium service. For non-subscription services, or services where the reward would be permanently applied to the user’s account, some consideration should be made to what, exactly, is rewarded to both the person sending and the person receiving the invitation when the invited person adopts the product. The benefits of virality must be weighed against the possible cannibalization of revenues.

If a viral mechanic is popular, and many people joining the product receive free functionality, then the structure of the product’s value proposition changes: a large portion of the user base sees the rewarded functionality as free, which may alienate those users who were not acquired through the viral mechanic and must pay for it. Likewise, the users who did receive the functionality as a free reward will come to expect that functionality, and perhaps similar functionality, for free in the future. The dynamics of a product catalogue can shift dramatically when large swaths of the user base are given access to some aspect of the product for free.

A third method to increase conversion rates in viral invitations is to instill in the sender the sense that invitations hold intrinsic value and thus should be sent strategically. This method may be conceptually similar to rewarding conversion, but it is applied, at a fundamental level, to the user’s understanding of the product’s use case, and it is often manifested by a restriction on the number of invitations a user can send. If the product’s value proposition is a function of the quality of the user’s network, then a limited number of invitations may inspire the user to qualify the people to whom invitations are sent on the basis of their likelihood to enjoy the product.

Such a perception from the user is powerful; the scarcity tactic attempts to prevent users from making impulse decisions on whom to invite into the product and to instead carefully and strategically evaluate each invitee. One effect of this approach may be an elongated virality timeline, as users consult people before inviting them, or simply deliberate more thoroughly on each invitation.

Artificial scarcity techniques need not be permanent, and they can be rolled out over time in accordance with product use to ensure that the most engaged users are given the latitude to serve as promoters for the product and that users carefully consider who will receive their viral invitations. The user can be granted additional invitations by continued use, or, if the size of a user’s network itself can be portrayed as product functionality, additional viral invitations can form part of the product’s monetization strategy (e.g., the user is given a limited number of free invitations but must purchase additional invitations). When invoked this way, artificial scarcity would achieve the same effect as the reward mechanism but approach it from the opposite position.

Unlike the strategies for increasing the number of viral invitations sent (which can produce negative effects and therefore must be undertaken with discipline), strategies for increasing conversion do not generally pose product perception risks and can therefore be undertaken in unison in pursuit of higher virality. But, generally speaking, the conversion rate for invitations and the number of invitation mechanics available are inversely correlated: the more viral invitations sent, the lower the proportion of invitations accepted.