Each year, hundreds of data scientists, analysts, product managers, and business leaders gather in London for the Big Data Week Conference. It’s a place to discuss the latest trends in automation, AI, machine learning and business analytics. And this year, I was fortunate enough to attend some of the keynote talks and sessions from trailblazers who are shaping the future of big data in business.
As a whole, it was a day of future-gazing and forward-thinking on the biggest questions facing the data industry today: how do we create more ethical algorithms? How can we automate systems in a responsible way? How do we create a data-driven culture from the top-down?
As a data strategist, these are the questions I could spend all day thinking about. But for everyone else, it’s useful to have some key points to keep an eye on for the future.
Here are five key takeaways that I walked away with from this year’s conference:
1.Ethics will be essential to creating the next wave of automated systems.
AI is known as the sexiest new technology these days, but with the advancement in tech will come huge responsibilities. Many of the speakers in the AI themed room touched on the role that ethics and decision making will play in the future.
Artificial intelligence and other automated systems, without restraints, can start to learn bad behaviours, and exhibit prejudices that exist within an algorithm. Microsoft experienced this challenge firsthand in 2016, to the shock of many. After launching a self-learning AI chatbot on Twitter, it took only 24 hours for the bot to be corrupted by trolls and start tweeting out racist and misogynistic tweets.
The problem? The chatbot had one task: to learn how people speak and try to imitate that in a natural way. But algorithms need to be more than singularly focussed. They need to be created with human-designed rules to guide learning in a responsible way.
2. We will develop highly intelligent, AI systems in the future: but humans will likely always be involved.
Developing more AI systems holds a huge amount of promise for businesses in the future. A report co-authored by Accenture Research and Frontier Economics claims that this growing field of technology will boost economic profitability by an average of 38% by 2035.
Satalia founder and CEO Daniel Hulme, a keynote speaker at the event, spoke about the importance of human involvement in creating AI systems. Deep learning algorithms make up some of the key components of AI by automating the self-teaching process through a series of inputs and outputs. But these algorithms alone can’t always solve a business problem.
“Deep learning is good at finding patterns. But it’s not good at solving problems”. – Daniel Hulme, Big Data Week London 2018
Even with the acceleration of technology, we are still in an age where people are the key drivers in solving problems. AI and deep learning is simply a more efficient way to solve these challenges.
But that’s not to say things will always stay this way; there is a possible future in which data scientists will create AI that is so intelligent it will surpass human ability to perform most tasks (as warned by the illustrious Elon Musk). Until then, data professionals will need to play a hands-on role in shaping the algorithms of the future.
3.Marketing teams don’t need a full-service data stack to start using analytics.
For businesses without a data strategy, starting an analytics programme from scratch can feel overwhelming. But it doesn’t have to be. Another key theme that emerged from this year’s conference was the emphasis on scalability: start small, and build on your progress.
Bernardo Nunes, Chief Data Scientist at Growth Tribe, used the marketing sector as a key example. Many businesses feel pressured to invest in a full data science team before starting with the basics. But Nunes said that for each business, there are usually only a few key big data “must-haves” to get started.
In marketing, these must-haves are predictive analytics and clustering algorithms. Predictive analytics empower marketers with data to predict customer churn, and then adds context to how they can prevent it with new features. Clustering allows marketers to better identify and understand their customers based on real user data, leading to more targeted user personas.
By starting small with only two models, marketing teams can start to experiment with what works best for their business goals.
4.Data products suffer from a lack of process and from teams being in silos.
Data science often suffers from the growing pains of trying to figure out how to spend more time on creative work, and less on maintenance. It’s currently estimated that on average, data scientists spend 80% of their time finding, cleaning and organising data, leaving only 20% doing actual data analysis.
In the specific field of machine learning, too many data scientists are writing models and creating code that never gets put to use. The best way to tackle both these challenges is through establishing a solid, reproducible data process that cuts down on maintenance and increases the number of data projects deployed.
Nick Jewell, Director of Product Strategy at Alteryx, advocated for a shift to a “deploy, or it didn’t happen” culture in data science. Instead of starting from scratch each time, he recommended building a process that hits four key stages:
A key stage in this process is “collaborate and reuse”. Data teams are too often isolated from other departments in a business. But by initiating a deployment procedure, teams can start talking and working together more closely in order to solve real business problems.
5.Building an analytics programme isn’t just about the numbers; it’s about creating change.
Decision making in business has started to shift more towards being data-driven rather than based on instinct. But that doesn’t mean that everyone in a business will immediately be on-board. When building an analytics programme, it’s essential to first create a cultural change before developing the actual tech behind it.
In order to get everyone in the business on board, you need to be able to clearly communicate data processes and benefits in layman’s terms. This skill is becoming so important that there’s even a new job title for it: the “analytics translator”.
The role of an analytics translator (sometimes called data translator) is to bridge the technical expertise of data scientists with the operational mindset of marketing, supply chain and other business functions.
Recent research by the McKinsey Global Institute went so far as to label this job as “the new must-have role” for businesses looking to invest in advanced analytics and AI. Their report estimates that demand in the US for this role will grow to between two and four million by 2026. As data becomes more accessible and cheaper to process, wider demand for data translators will surely continue.