Growth Hacking & Data Science
Marketing has always been fast to change its practices and adopt the latest trends, but the rapid growth in online media and the tremendous amount of personal information accompanying it has radically transformed the marketing industry over the past decade or so (see figure below). In response, growth marketing or growth hacking, starting in the Silicon Valley, has quickly become an epidemic among startups and several pre-established organisations. At its core, growth marketing is a lean, data-driven approach aimed at helping organisations to quickly develop a deep understanding of their potential customers and how best to find, approach and retain them. To find out how data science is applied within the industry, Roy and I interviewed Bernardo Nunes, Growth Tribe’s Head Data Scientist.
Data science meets social science
Bernardo’s spent four years obtaining his PhD and worked in Finance/Asset Management. We asked him why he’s made the move to Growth Tribe:
My PhD is in behavioural science, a research field made up of economists and psychologists who tend to have a deeper understanding of economic behaviour. For psychologists, economics provides a good behavioural framework. For example, they look at the rational behaviour and then study the deviations from rationality. For economists it is fruitful to look at the psychological insights because this allows them to understand how cognitive biases affect economic behaviour. During my PhD I worked for six months with the Financial Conduct Authority, in the UK, and they have a behaviour and data science unit which is responsible for implementing large scale data analysis. During this period I realised that data science is a growing field.
“If you you have a smartphone, it’s like a questionnaire that you are filling out every time you use it”
I was interested in the Internet of Things and the idea that if you you have a smartphone, it’s like a questionnaire that you are filling out every time you use it. It was the next step, I believed. So I asked myself: Which companies are trying to combine insights gained from these devices in order to generate behavioural change? This is what led me to growth hacking. Growth Tribe Academy is a company where the combination of skills necessary for growth hacking and digital experimentation are combined with behavioural psychology and data science. And it’s a company where I have the opportunity to teach, which is something I really enjoy.
The new marketing
Marketing can have a somewhat negative reputation (think making people click on ads), so we asked for Bernardo’s take on this.
It’s partially true but it’s mostly the view of what a traditional marketer does. Traditional marketers focus on the first steps of a customer journey (a.k.a. the pirate funnel — see below): generating awareness and activation by nudging consumers. But many new business models are subscription-based, such as Netflix and Spotify, and demand a good understanding of customer retention. So the task of a growth marketer is to optimise all steps of the funnel (customer journey). In contrast to traditional marketers, growth marketers would also be concerned with users’ retention rate and may employ a recommender system in order to optimise app usage, for example. Finally, growth marketing is also concerned with optimising revenue.
We were curious what role growth marketing (or growth hacking) would play within organisations in the near future.
The market is converging the idea of growth hacking with growth marketing but the word “hacking” brings an idea of subversiveness. If you look at the definition of data science, you see some Venn diagrams (see below) which include hacking skills as well. Hacking represents the idea that you can deal with debugging, scraping and other challenges, which is useful but not the core of what a data scientist does. The core is to learn the tools, understand the technology and so on. So I prefer the term growth marketer.
Growth marketing will play a more prominent role within organisations in the future. You might see a shift in organisational structure where a growth department would replace the marketing department and you’d have a chief growth officer.
But “growth” should be understood in the right context. Organisations which need to maintain their position need a “growth” department just as much as an organisation looking to scale exponentially. So growth is not only about growing and generating more revenue but also about maintaining your position, which often implies a need to launch new products. So the growth team is responsible to ensure that the new products are adopted.
“If your product depends on a website or an application, you have to start running at least part of your company under a system which facilitates idea generation and testing”
We asked what common problems Growth Tribe has encountered in its collaboration with its different partners or clients.
In the last few years, incumbent companies have been forced to make many changes in terms of human resources. This is mostly because what used to be called a ‘one big decision day’, is being replaced by lots of small decisions which are tried out — some of them are failures and some are successes. This is the idea of rapid experimentation.
Obviously, in this ‘one big day’ scenario you have a relatively vertical hierarchy, so one of the C-level professionals (executive managers) is convinced by, say a consultant, that change A needs to be implemented. This change is announced to other departments which then start working on it and come up with prototypes. These are then presented to the board of the company, which in turn decides on the new norm. This is how traditional companies operated in the past and it still happens in many traditional companies today.
Some companies realise that they have to move towards rapid experimentation. Their websites and apps (and online presence) need to be adapted for a fast pace of change, and you can no longer work with an infrastructure designed for the ‘big decision day’. Such traditional companies typically don’t have a rapid experimentation mindset yet. So when we teach digital skills, they face problems to create a culture which complements this pace of doing business or solving problems — if your product depends on a website or an application, you have to start running at least part of your company under a system which facilitates idea generation and testing.
The best solution would be to hire X extraordinary people. These are the unicorns who know behavioural science, data science, marketing and product management but also understand your sector. This is of course impossible. We suggest creating growth teams made up of people with complementary skills but who also know each other’s jobs a bit so they can collaborate and implement these experiments.
Another problem encountered often in such organisations relates to the perception of risk. In some companies you’re able to build teams which are oriented to try new stuff internally, but they can be perceived as a threat by the rest of the organisation. This is especially the case with the lawyers or the parties which are concerned with the organisation’s risk. As a workaround, some companies buy a startup or invest in a new company outside the organic group. Within small companies it is easier to adopt, or start with a culture of rapid experimentation — but you still have to breach the skills gap. This leads us to the third most observed problem.
If you are an individual who wants to be at the top of your field, you have to continuously update your skills. If you are part of this small company you need to have knowledge of front-end, back end, A/B testing, applied machine learning, GDPR and so on. You don’t learn all of these things together at the university — and this is where we come in.
Marketing and data science
A typical organisation is made up of a few departments. We asked why the marketing department is the most obvious choice to adopt a data-driven mindset.
Marketing departments tend to have a better understanding of the customer’s journey and pains. Besides, the technical marketers already use tools which give them a descriptive idea about their audience. Going data-driven means going beyond descriptive statistics (given by Google Analytics, for example), and making use of machine learning and experimentation to make business decisions. So the aim is to take the same marketing processes, say persona building or churn indicators, and make them data-driven.
Assuming you have C-level support, that you know the risks the organisation is willing to take, and that you have a budget, the next step is to build a team with some people from the marketing department. You might want to build a team with some people from UX, some who are analytical and some who know more about product management, for example. So it’s best to start with a cross-functional team.
A typical case study is TransferWise. They operate in a very traditional business — international money transfer. But they started asking themselves how this could be done faster and cheaper than banks do it. Instead of working exactly as the banks do (one treasury department dealing with all the currencies), they created a growth team for each currency. These teams have the autonomy to implement all the steps necessary to open a new currency, allowing them to take ownership of the product and capture the business value.
We asked Bernardo to give us some concrete examples of how data science is applied in growth marketing.
If you go back to the customer journey, at the top of the funnel, the first touch point is the signals you get about the customer’s awareness of your existence. Once you attract people to your website, you can create a lead scoring based on the clicks new visitors make on a website so you can see which are the ones you can offer vouchers or discounts to in order to nudge them to purchase or to add a product to their chart or even to see a specific product. Giving such discounts to everyone would be costly as you’d be giving away money to customers who are willing to purchase anyway. So a lead scoring model would help you focus your efforts on top potential customers.
Next, say in the activation or retention step, the same thing applies. You create a score and you interact with those customers who are most likely to churn or who might need that extra bit of an incentive to signup or share their email with you — so you push them to go further into the customer journey.
For referrals you’d be following a similar procedure. You ask yourself who are the influencers in my database, for example. With segmentation also — you cluster based on personality traits and you create personalised messages which are tailored to the customer’s personality. Machine learning allows you to do that because it allows you to discover the personality trait and also score — how extroverted or conscious is this customer, for example.
Freelancing is common. We asked what role freelancers play within a growth marketing organisation.
If you hire a consultant or a freelancer, he or she will solve the problem for now and maybe teach you some of the tools that you can use, but the solution provided may no longer be suitable in three months’ time. So the risk with hiring freelancers is that the learning is lost. Without having the skills needed to adapt to the next best channel or tool, you’d have to hire a freelancer constantly. You would rather build the skills within the organisation and keep them working for you. Which is another challenge — how to keep them interested and satisfied?
On lifelong learning and updating one’s skills
On the one hand, universities focus on teaching proven methods: Growth Tribe is teaching the latest tools and tricks, on the other. We asked how they keep their curriculum relevant.
As a prospective learner of a new skill, you have three main options. The first option you have is to wait for universities to adopt it. This is very safe. The university adopted it because there are jobs requiring these skill sets. The other way is to do in-job education, which would allow you to start learning and mastering a skill as soon as possible, while having a moderate amount of risk. The third option is self-teaching. Some 15% of the population might have the self-control for self-paced learning, but most people are willing to pay for in-person education. The traineeship is aims at combining in-job training while giving you the opportunity to engage in side projects which are of interest to you.
On interpretability and academia
University courses place a strong emphasis on understanding the results, interpretability and direction. In contrast, some machine learning models are blackboxes. Given Bernardo’s academic background, we asked him how machine learning will impact academic research.
It depends. If your goal is to be precise or more accurate, then you can run a set of algorithms. In academia we are developing very good models to predict cancer, for example. The first algorithms to outperform radiologists and dermatologists have been created last year and this year, respectively. In such a situation we should aim for the best predictions. But we still need a lot of research on interpretability and I think a lot of academic research will focus on this aspect. In image recognition this is happening — how do you take those filters from the convolutional neural network and interpret them?
“Interpretability will be a big concern for academia and corporates due to the right to explanation.”
But interpretability impacts businesses also…
But academic researchers often collaborate with organisations also. Great examples are Andrew Ng (Stanford University and founder of Coursera) and Yann Lecun (New York University and Facebook’s director of AI research). So this is a very important practical problem for companies as well, especially because of GDPR, which has a very valuable point — the right for an explanation. If I deny you credit as a bank, I need to have an explanation. This forces organisations to take interpretability into consideration. Can I solve this by running three parallel models and use the insights from the regression to explain my decisions, even if I didn’t use the regression to make the decision? You have to balance that. So interpretability will be a big concern for academia and corporates due to the right to explanation.
*Want to know more about Data Science in academia and business?
Check out JADS!
