How network thinking can help you hack growth

Introduction

My goal in this piece is to illustrate how networks complement established growth hacking practices. Network thinking, including concepts like centrality, periphery, and topology, can help hackers implement the experimentation and tinkering required to address current challenges in shaping consumer behavior. Growth hacking is experimentation and tinkering related to marketing questions. As Ryan Holliday put, “It is more a mindset than toolkits or buzzwords.” In this case, we start this conversation with some of the most tested growth hacking frameworks, add network thinking basics, and finish up with a set of principles consistent with the hacking mindset.

Understanding Sales funnels and Growth Hacking

The core of growth in the digital economy. Sales funnels are at the heart of any growth hacking, digital marketing, or similar digital world approach. They are a visual metaphor for the sales process covering the customer journey from the initial interest to the final product adoption. In this context, any sales funnel deals mostly with four distinct challenges in the process:

1. interest maximization at the entry point (top of the funnel);
2. demand qualification of the most promising leads;
3. conversion of qualified leads into customers (bottom of the funnel);
4. exploring additional revenue opportunities from current customers

Sales funnels are everywhere. There are many reasons for the adoption of sales funnels as the standard framework for marketing and sales. Sales funnels are visual and straightforward. In the short term, their format provides an intuitive but detailed view into the long and ambiguous purchase decision process experienced by different consumer groups. Over the long term, the repeated use of sales funnels helps companies build a healthy base of loyal customers. In addition, the widespread adoption of sales funnels has provided a broad library of benchmarking cases and a vast vocabulary in comparing similar challenges on marketing and sales decisions. Terms like top or bottom of the funnel are standard industry vocabulary for customer acquisition and retention challenges.

However, even after so much success, there are still significant gaps in understanding customer behavior. As with any framework, sales funnels leave many vital aspects of the conversion process untouched. Consumers’ journeys are neither clear-cut nor unidirectional, as presented in the sales funnels.

The context of growth hacking continues to change at an increasing speed. With the emergence of social media, technology has disaggregated physical and social spaces into many different relationship networks while also changing how consumers absorb and react to information.

Taking action in such a context becomes more difficult. We either run the risk of acting like bureaucrats, considering our simplified funnels are the de facto process (and not a visual representation to it), or we get smothered by the complexity in trying to come up with practical answers. The remaining part of this article aims to offer a middle ground, providing an analytical intuition about networks and how they are suitable for the progressive exploration on growth hacking questions.

It’s a world of networks!

We have two critical questions to address in this part. First, we have to map and understand the specific relationships shaping consumer behavior. Then, how might these relationships evolve.

A simple case, many perspectives. As an initial example, a single household has many different relationships occurring in a diverse set of environments. Zooming into those relationships, we can see that a 35-year-old mom living in central São Paulo, Brazil has i) tightly-knit relationships with her husband and children at home; ii) new connections with fellow members of her church or religious organization, iii) specific friendships on a highly clustered social club and iv) leadership position of 15–25 coworkers in her job as the marketing manager for a local tech startup. We can also understand whether these relationships are central or peripherical depending on each specific environment.

Networks analysis (or graphs) provides the conceptual map of an individual, meaningful relationships (or interactions) within a specific environment, and scale of observation.

Connections, Centrality, and Periphery. At its most basic level, a network represents real-life relationships amongst a set of specific agents (or individual customers). Under this definition, a network has two main structural elements: i) nodes, as the representation of agents or individuals; and ii) links or edges, representing relationships amongst the agents. As a result of the connections between the nodes, we can infer which notes are central, defining the most of the network connectivity and which nodes are peripherical, depending on the central nodes to access the remaining regions of the network. Furthermore, the centrality of a node or a path (sequence of nodes) can be measured by its degree. This quantity measures how many edges or paths connect to each node. Finally, the arrangement of nodes and edges defines its topology.

Patterns of network topology and their classes of behavior

Different topologies drive emergent behaviors. Besides that, every network is a continuous pattern, not a fixed structure. In reality, we might see a juxtaposition of different network structures depending on the perspective we take. Amongst the most frequent topologies, we can mention random networks (where nodes connect according to a specific probability distribution), small-world networks, and scale-free networks.

A small-world network topology reduces the distance between any nodes. Centrality lies in few long-range connections bridging otherwise distant neighbors’ clusters — these long-range connections concentrate the most significant number of possible paths on the network. Consequently, the presence of long-range links reduces the average distance between nodes (making the world “smaller,” providing the reason for its name). Stanley Milgram’s classical study of 6 degrees of separation is one of the best examples of small-world networks in real life.

A scale-free network arrangement presents a regular “winner takes all” structure at different scales. In this case, a small number of nodes drive most connections (also called super connectors). In contrast, the large remainder of nodes shares the residual number of links. The international system of airports is an excellent example of scale-free networks. Looking at either global (e.g., Atlanta, Chicago O’Hare, Frankfurt, or Dubai) or regional scale (e.g., Western Europe, Madrid, Paris, and Milan), you can see super connectors consolidating most of the connections at that level. In this type of topology, we should always consider i) the overall degree distribution of the network (not their averages) and ii) different possible collective trajectories (not individual estimations).

Even with the background presented above, analyzing the richness of networks can be a daunting exercise in all its detail. Therefore, in the next section, we will discuss some progressive ways of using network thinking in growth hacking decisions.

Looking at the threes without losing the big picture

We will touch just on some of the most fundamental concepts helping users understand networks’ power for sales funnels management decisions. Treat the steps below as a set of first principles in using network thinking, applying them as you see fit.

If this piece has caught your attention, I suggest looking at the reference section in this article, where you can find many sources related to networks’ formal mathematical treatment.

1. Choose the part you want to hack: Each sales funnel pattern presents specific challenges to be addressed. For instance, exposure at the funnel top deals with covering the broadest base possible across a range of different segments. On the other hand, conversion deals with processing the most likely or the most significant nodes on your target network. Selecting the part of the funnel you want to hack allows you to frame the right initial questions related to the network topology and the aggregated behavior you want to achieve.
2. Define a scale of analysis: This is the trickiest part. Zoom too much; you end up losing the big picture. Zoom too little, and your network becomes this illegible with a lot of noise and minimal relevant signal. If you are trying these waters for the first time, I suggest taking one level of zoom below your current analysis scale. For example, if you’re looking at regions, try looking at specific clusters of areas. Stop the zooming process when you feel you are getting too much input for very little return in understanding. For example, by targeting superspreader clusters (and not individuals), Japanese and South Korean policymakers fighting COVID addressed the growth of contaminations acting on two different fronts. First, by reducing the impact of already established superspreader clusters via backward tracing. Second, by avoiding future superspreader network formation via social distancing and large-scale vaccinations. As you see, selecting the right “zoom” level can make all the difference in strategy effectiveness.
3. Draw the first draft of the network according to your current understanding of its structure. This step aims to have a valuable and actionable picture, answering the most relevant structural questions. Where is the centrality on this network (on the link or nodes)? How does the access to the periphery change with changes on the centrality? Finally, focus your analysis on the areas that will help you answer the questions presented above.
4. Refine your understanding via incremental exploration: It is easy to get overloaded with the sheer amount of questions and options available in networks. And there is still a lot of ambiguity to be tackled and many unverified causal relationships. At this point, we have to create a portfolio of targeted experiments. As the British statistician George Box said: “All models are wrong, some are useful,” we should focus more on discrete incremental improvements rather than arguing if an answer is correct in every single context. As part of this process, understanding gaps dictate additional analytical concepts such as clustering coefficients, dyadic structures, motifs, etc. New data sources should follow the same reasoning for inclusion. With this thinking in place, we should understand the connection between local and simple agent behavior and other emergent phenomena (like the formation and development of communities) in growth hacking strategies. In the excellent article Seeing like an Algorithm, Eugene Wei describes how Tik Tok decodes user experience into tangible attributes, creates an algorithm reading these attributes, and establishes positive feedback loops for users based on the quality of the content. With this system in place, TikTok reconfigures the topology of its user network from a topology based on social connections to one based quality of content or similarity of interests. It is an excellent example of progressively granular interventions based on the wealth of data generated by its user base.

The road ahead. Finally, we should be humble. We have to make practical decisions in the absence of data and significant uncertainty. In this context, we should aim for improved understanding via exploration. The principles presented in the last section illustrate this type of thinking.

I hope this brief conversation has piqued your interest in the power of network thinking in tackling some of the most challenging problems in growth hacking. Networks are suitable to many different fields, and their body of knowledge grows with new applications and reference cases. However, the practical application of network thinking requires a deliberate approach of first principles progressively building the practitioner’s understanding about its network of focus and his/her ability to drive the desired change.

References

Caldarelli, Guido & Catanzaro, Michele — Networks A Very Short Introduction, Oxford Press

Tufekci, Zeynep, K: The Overlooked Variable That’s Driving the Pandemic — The Atlantic, https://www.theatlantic.com/health/archive/2020/09/k-overlooked-variable-driving-pandemic/616548/

Wei, Eugene, Seeking like an Algorithm,https://www.eugenewei.com/blog/2020/9/18/seeing-like-an-algorithm

Gal, Orit, Pattern Analysis — Making Sense of a Rival System — http://www.socialacupuncture.co.uk/pattern-analysis/