2020 will be the year we see further democratisation of machine learning tools and a lower point of entry for their usage. This will make this area of data science even more commonplace not only among top tech companies, but also small and medium-sized businesses across various verticals.
However, one aspect which is potentially underrated when looking at the big trends, in terms of the future of data science, is around language frameworks used to make the everyday data science tasks possible. Today, there are two major frameworks, R or Python (or in more pragmatic data science circles, both!). One is praised for having the most beautifully designed data wrangling syntax and plotting libraries, the other for its expressiveness and having the best deep learning libraries available today. However, both suffer from being relatively slow as they’re higher level languages. One additional disadvantage of R and Python is that neither were built with mobile devices in mind. And, if the last ten years are anything to go by, mobile, wearables, and IoT (internet of things) devices will only further advance their presence. What should data scientists do in that case? Various members of our data science team at Greenlight Digital shared their thoughts on the use of data science for the year ahead.
Swift will change the face of data science in 2020
There is a new saviour on the horizon – Swift. When you think about Swift, you usually think about the Apple app ecosystem and hours spent playing Candy Crush on your commute. Swift was developed by Apple to make it easier to build apps and, by default, grow its app ecosystem and increase customer retention. Swift was built with mobile devices in mind from the start; it natively supports apps for iPhones, iWatches, and iPads so developers can immediately develop apps for those devices. Combining mobile devices and the latest developments in machine learning, Swift creates a potent and promising mixture. The concept of using machine learning on mobiles is nothing new, of course; we often hear about how Apple or Google are bringing AI to mobile devices, for example with the introduction of Face ID algorithms that unlock your phone.
What’s different though is the fact that both Apple and Google have worked together to bring one of the most iconic deep learning frameworks, TensorFlow, to Swift. Also, their approach in doing so was different from the get-go; it wasn’t about building a wrapper in Swift for TensorFlow, it was all about making it the “first class citizen”, as they called it, meaning that deep learning algorithms would become commonplace when working on new apps. This is an interesting paradigm shift for data science. An additional advantage of Swift is that it’s not only limited to the iOS ecosystem anymore and was “open sourced” to work on Linux. There are also efforts to bring it to Windows environments, making Swift potentially a very interesting alternative to both R and Python – especially as we can easily call all the native libraries from Python into Swift effortlessly.
If I were to bet on changes for the year ahead, it’d be that 1) more advanced mobile technologies are on the horizon and 2) the apps that we use will become more powerful than ever. So, assuming a language which was built with these predictions in mind will get more traction in the data science circles in next few years is a fairly safe bet.
Joshua Hudson, junior data scientist
According to recent studies, 60% of content marketers identify personalisation as one of the key challenges to their organisation, while companies that have fully invested in all types of personalisation are predicted to outsell those who have not by 20% (Gartner). Customer segmentation has long been used to tailor marketing strategies to target each group of customers differently. One-to-one marketing takes this a step further, applying these concepts to the individual level, with each customer getting their own unique experience.
The problem with one-to-one marketing is its scalability; it would be impossible to manually develop an individual strategy for thousands of customers simultaneously. Artificial Intelligence (AI), however, could use the ever-growing amount of data on each user, learn the most effective way to target said user, and take the appropriate action. AI is already being implemented within various channels. For example, Google Ads offers automated bidding, which uses AI to produce an optimum strategy to maximise impressions, clicks or other metrics. Currently, these applications tend to be channel-specific, but 2020 could see the emergence of more omnichannel platforms. This would allow for a more comprehensive and truly personalised experience for each customer, delivering the content most relevant to them, via their preferred channel.
George German, junior data scientist
The (anticlimactic) rise of big data
Once the hottest topics in the marketing world, big data, has become a meaningless buzzword, serving as white noise to most in the industry. It once promised marketers untold increases in ROI but, as of yet, hasn’t quite lived up to the hype. On its own it might be “so last year”, but through developments in deep learning and the boom of data science, a new paradigm in digital marketing has developed - predictive marketing. Predictive marketing has already been branded with the potential to improve and streamline every aspect of the marketing pipeline. It’s still about extracting insights from huge amounts of data in order to predict future outcomes, but with greater efficiency and higher levels of automation.
What is predictive marketing?
Digging below the surface of these hysterical claims, predictive marketing is a technique which determines the probability that a marketing strategy will succeed based on previous outcomes. However, this process not only uses historical data to probe for the right marketing strategy, it can develop and improve its marketing decisions in a continuous (sometimes iterative) process. This dynamic form of “learning” is what really sets it apart from previous techniques, giving predictive marketing the potential to automate parts of the marketing pipeline, saving copious amount of time. The dynamic learning process which underpins every prediction stem from the subset of machine learning is called deep reinforcement learning. Both deep reinforcement learning and predictive marketing are in their infancy, with less than 25% of marketers using any form of predictive models, meaning the upcoming year could herald a race to master this new technology, according to Martech Zone.
The concept of reinforcement learning was born out of advancements in AI, animal psychology, and control theory. At its heart, it involves an autonomous agent, like a person, dog, or deep neural network attempting to navigate an uncertain environment with the goal of maximising a reward. Each action taken in the environment is associated with either a reward (if the action is beneficial) or a punishment (if the action is detrimental), so the agent learns which actions are worthwhile repeating in a particular scenario. A similar process to how you train your pets. In the case of marketing strategies, you can design a deep reinforcement model to use any KPI as the metric to maximise, therefore giving a marketer the ability to tailor the automated strategy to the business needs. Combining deep learning techniques with powerful computing infrastructure and greater accessibility to cloud technologies, in-house AI systems can be easily implemented. So, could machine-learning fuelled automation be the MarTech trend of 2020?
Wojciech Bednarz, data & insights senior strategy manager at Greenlight Digital.