The traditional companies took a century to grow their first 100 million customers. In contrast, the internet giants have evolved at a much faster pace. With companies like Google and Facebook ‘hyperscaling’ in 10 to 20 years, they have set an unparalleled pace for every industry.
So what’s behind the hyperscalable business model? It turns out that the algorithm is the Internet’s billion dollar unicorn, transforming and innovating faster than humans.
The New Ingredients Behind Successful AI Deployments
The companies succeeding with AI are taking advantage of today’s technology landscape to drive new developments and innovation. They are harnessing the 1000 fold data increases, access to new sources of data, in conjunction with the improvements to algorithmic design and accessibility. They are also taking advantage of more powerful compute and the 10x speedups when moving from traditional CPUs to GPU technology.
When companies begin to drive and capitalize on these elements, they are able to unlock new use cases, as well as realise improvements in age old industry challenges. These techniques support organisations to mature machine learning efforts to skilfully handle explosive growths in complex and time-series data.
Whereas traditional machine learning requires onerous time from a data engineering and wrangling perspective, AI and specifically deep learning, is enabling the algorithm itself to self-train, test-and-learn. With algorithms having the power to adapt through learning, they are able to find complex predictive patterns difficult for human analysts to detect. These algorithms mean organisations can accelerate at an unparalleled rate. In industries like automotive, these advancements are driving complete transformation, enabling self-driving cars and smart city initiatives.
Fraud detection: Danske Bank takes AI action
However, it is not just new applications and emerging businesses that benefit from these techniques. Take Danske Bank, who were struggling with identifying fraud given the exponential growth of digital transactions and the increasing complexity of fraud. With a low detection rate of 40%, false positives at an all time high of 99.5%, and tens of millions of Euros lost every month – the bank knew they had to take action. In implementing innovative AI models via an Analytic Ops framework, Dankse Bank was able to realise a 60% improvement in fraud detection rates, 50% reduction in false positives, and millions saved in fraud losses.
But how did the bank, and other companies who have seen similar success, learn to crawl and walk before they could really run with AI?
Automation Applied to Analytics
Successful analytics focused companies can make the value driven application of advanced analytics look easy. Yet, the truth of the matter is that the practical and profitable applications of AI, machine learning, and deep learning are complex. It is a multistep process that only a handful of companies have successfully deployed at scale.
The companies truly invested in AI are creating frameworks that enable the deployment of hundreds or thousands of models, at any time, for any analyst, written in any language, operating on a joined-up view of their business data. This emerging discipline of Analytic Operationalization, or ‘Analytic Ops’, is combining the well-established principles of DevOps from software engineering with the domain of analytics. Without automation applied to analytics, innovative models are destined to be left on the bench.
Personalisation through recommendation: Netflix balances accuracy versus cost
Take Netflix as an example, personalisation through recommendations is a lucrative business reported to drive savings of more than $1B per year. Through implementing best in class algorithms Netflix is able to realise subscriber growth and retention, as well as optimise the catalogue size. However, when Netflix paid out a million dollars for a state of the art recommendation algorithm, in a Kaggle competition, they could not productionalize the winning entry. The personalisation science and engineering team at Netflix reflected “we evaluated some of the new methods offline but the additional accuracy gains that we measured did not seem to justify the engineering effort needed to bring them into a production environment”.
This example, and many others, signify the importance of considering Analytic Ops at the outset of a project. Analytic Ops frameworks are essential to the success of AI endeavours, providing version control of data and models, automated workflows with continuous integration and automated testing. Without these environments, the most sophisticated models have no route to production.
Level up to Autonomous AI
Not every organization investing in AI has leaped to hyperscale stardom. Many are still in infancy, struggling with the challenges of adapting to a new data driven and algorithmic automation culture. A culture that requires a solid foundation before AI can be truly scaled to deliver value across the enterprise.
So, what must businesses remember in the journey to running with AI in a fluid, autonomous way?
Firstly, there is no single technology that can do it all, ecosystems are needed that balance agility with control: A Unified Data Ecosystem is essential, with easy access, integration and flexibility to work with data of different formats, sources and complexity. When it comes to the tools and technology within this ecosystem, it’s essential to balance agility with control. Completely locking down an IT environment is not possible with today’s rate of innovation and change – it is expected that new technologies will come and go in a timeframe of weeks and months, rather than years. In this age of change, consider Cloud to provide the necessary flexibility to keep up with the technology advancements.
Secondly, diversity of skills and talent is needed to be successful in AI R&D and deployment: To complete successful pilots and promote these past the prototype stage and into production, teams need to have architects, engineers, data scientists, dev ops and automation engineers, visualisation experts, as well as business experts working together. However, hiring talented individuals without well-designed functional environments minimizes the value they can bring over the long term. Hence, companies must also provide data science labs for analytic R&D, as well as automated environments for managing and monitoring analytic products, with governance and accountability playing a crucial role.
And finally: Don’t be left behind! Leveraging analytical capability, in the forms of AI and machine learning, is being used today to realise true business benefits such as improvements in customer experience, achieving cost reduction, building cross-sell capability and asset optimization. Businesses that start with a business use case and weave AI into the heart of operations are harnessing the AI trend to deliver value. Making AI a key component of company culture going forward will help more companies move through crawling, walking and running with AI to evolve as the hyperscale examples of our future.
Perhaps Harvard Business Review said it best: “Over the next decade, AI won’t replace managers, but managers who use AI will replace those who don’t.”