So, you’re all aboard the machine learning bandwagon, but what now? Mathias Golombek, CTO at Exasol outlines three steps to ensure your investment is optimised.
The age of machine learning (ML) is upon us. Last year we saw ML technology gain the recognition and practical applications that it deserved. Stepping out of a Hollywood-induced shadow that foretold an apocalypse and human enslavement, for me the practical reality of machine learning and AI now looks a lot more like better healthcare systems, more energy efficient facilities and our very own virtual PAs – Siri, Alexa, Google Now.
Businesses today want their data to do the heavy lifting for them, addressing the ever-present challenges of cutting costs, streamlining operations and increasing margins with new tricks. Applying ML to business processes helps to make granular analysis and improvements possible at scale – it’s a force multiplier. What’s more, your competitors are also paying attention: a Deloitte survey revealed that 57% of businesses increased spending on related technology in 2018.
Translating this impressive technology into useable, supportable applications can be daunting for many, and organisations should be cautious to avoid rushing into ill-defined projects. Tech giants Amazon, Facebook and Google have paved the way for new ML breakthroughs, making the latest and greatest frameworks more accessible than ever, but there is a lot more to successfully applying ML than plugging some data into TensorFlow. Disregarding the hype for a moment, I would like to share three steps that are crucial for organisations to successfully apply ML to their data.
Prove the concept
Some organisations jump on the ML bandwagon without a realistic strategy to create direct business value, or to rationalise the sizeable investment they need to make. In fact, according to the report Driving the rise of AI and ML with data, one in four business leaders found that their ML investments didn’t deliver the expected time savings or estimated cost savings, largely due to unrealistic expectations and a lack of clear objectives.
Organisations looking to begin their ML journey should start with a clear proof of concept. An unrealistic project will be dubbed a failure and set back progress throughout an organisation, so it is imperative to have a clear proof of concept in mind before investing. The journey to success is only as achievable as the road map used to chart the process from start to finish.
Such a clear, concise road map also helps secure vital C-suite and senior executive support. Moving ahead without this support is a very tangible risk – a quarter of organisations studied in the Moving the Enterprise to Data Analytics report reported a lack of board buy-in as the primary cause of failure in their data-driven initiatives.
Many executives still regard ML as a cost, and regard it as unlikely to drive revenue or improve current business operations. However, it is imperative to reverse these preconceived ideas in order to reap the huge potential business benefits of this advanced technology. To get buy-in, you need to show how ML projects will add value – the C-Suite need proof, in numbers they value, from a real project.
Reinforce the database
Data science combines ML with statistical models, algorithms and numerous processes to exploit data. But whether it is applied to supply chains and stock control or the automation of repetitive tasks, it relies entirely on a consistent feedstock of clean data. Each project requires a single source where data is collected, collated and manipulated to allow the algorithms to exploit its value.
Findings from the Driving the rise of AI and ML with data report revealed that nearly half of organisations recognise this reliance on data, and are already investing in data quality services to ensure their information is serviceable for ML.
The foundation of an organisation’s ML infrastructure is a fast analytical data repository, which provides access to multiple datasets and enables algorithms to process large volumes of data quickly. Any organisation that is applying data analytics across their business will already have a data warehouse and an in-memory analytical database in place to ensure this data retrieval process and their analyses are performant – for the rest, there is a lot of catching up to do.
For ML applications, specifically, businesses should ensure they provide a common data infrastructure that all data professionals can work from. For the BI team, this will typically be accessed through SQL, but data scientists will need an infrastructure that can run scripts on the data using their preferred languages – typically Python. This standardisation of data infrastructure allows businesses to be more flexible in their use of ML across all parts of the organisation.
Build skills and education
Employing and retaining the right skills is important to all areas of a business, but it is vital to success in ML investments. A quarter of UK business leaders cite a lack of employee skills for the failure of their ML initiatives, so it is critical to have both widespread understanding of data and specific skills in data science and relevant programming within an organisation.
In fact, according to LinkedIn’s 2019 annual skills report released earlier this year, the second most in-demand technical skills are artificial intelligence and machine learning. As AI and ML technologies become increasingly important for all industries, business leaders must recognise the importance of the employees who implement and maintain these systems.
There is no sure-fire way to guarantee the success of an ML or AI operation, but organisations can safeguard their investment by ensuring they properly understand the capabilities of machine learning, form a strategy with demonstrable value for the C-Suite and build the data infrastructure and skill base to make it possible. As machine learning is applied to richer data sets and real time data feeds, I hope to see organisations become more efficient and more innovative in providing better service to the people they serve.