The power of machine learning in business
ITWeb Cloud Summit 2018 will host experts from across industry in just under a week to uncover the evolution of cloud from a disruptive force to next-generation IT. Tim Andrews, director, Dotmodus, is one of them.
ITWeb: You are presenting on a very topical business concern at the moment: what is machine learning (ML) and how can you build an uncomplicated practice in your business using the cloud tech available today? Could you provide us with a summary of how ML can positively impact on business?
Andrews: ML is the ability to investigate large amounts of unstructured or structured data to find patterns in that data with the ultimate goal of producing valuable business outcomes. At the heart of ML is the ability to do all of this without explicitly programming the computer to find those patterns. Machines learn from their own errors and successes.
To get started, Google provides a number of out-of-the-box trained models that allow us to extract information from any image, determine sentiment from paragraphs and translate many different languages to many other different languages.
To build a custom ML model, business should partner with a company that employs good data scientists with focus on cloud-level scalability.
ITWeb: What, in your opinion, are the key benefits and challenges associated with ML and why is this relevant to the cloud as a whole?
Andrews: ML opens up an entirely new way of thinking in terms of solving our everyday business problems. Suddenly, we're able to automate menial but expensive processes like data capturing, image recognition and tagging, data analytics, analytical prediction and many more.
These benefits relate specifically to reduced cost: time to market for a solution is significantly reduced since we don't need to spend time thinking of every eventuality; staff used to process these tasks manually can be reskilled and shifted into more valuable parts of the business; ML models improve over time and as accuracy and efficiency improve, costs reduce.
Typically, these benefits are enough to drive an ML initiative inside your business. However, it's important to understand some of the challenges to make sure your strategy is successful: the skills required to build ML models are extremely specialised and so difficult to find and retain. The solution to this involves, more often than not, finding a partner who has a proven track record in sourcing and retaining this level of talent; ML problems are iterative and they require an attitude of learning and improvement over the short term. Business often approaches the problem from a traditional perspective of 100% accuracy out the box. It's important to understand that completely custom models require completely custom data. And where we use completely custom data we must recognise the margin for error on day one. However, it never stays at that level. It improves in a short amount of time as the model learns from its mistakes; ML models are computationally hungry. They are very heavy on CPU usage. Luckily, with the rise of cloud computing and the reduction of costs to using the cloud, this has become a non-issue.
To the last point, this is why ML can ONLY exist sustainably and cost-effectively on the cloud. In addition, ML requires a lot of data. Big data presents the prime problem of horizontal scalability which is a problem the cloud has long since solved.
ITWeb: If you had to provide a checklist to assist attendees of the 2018 Cloud Summit to help an organisation when making the decision to move to ML using cloud tech available today - what questions should they ask themselves and the provider?
Andrews: There are so many use cases for ML that it's hard to think of any company that can't benefit from implementing a ML model or consuming a pre-built model provided by Google. If a company is serious about leveraging the benefits that machine learning can offer, they can look at the following: Do you have a problem in your business that is repetitive and time consuming? Does your company have the collaborative appetite to implement a ML model to resolve the problem? Does the problem you're having have a lot of data that can be used to build a ML model? Will the automation of your problem, using ML, reduce the cost of operations or otherwise improve a key strategy? Always remember that we can automate a machine to do what a human does but ultimately, if a human would have trouble completing a task (for example, they can't read the handwriting on a form), then the machine is not going to be able to infer that information.
When approaching a provider, make sure they have proven industry experience. The process of building a model can turn out to be very expensive if a company doesn't know what they're doing. Partner with a company who has cloud first as a strategy when approaching the ML problem. Always ensure that your provider balances the implementation of an ML model with the anticipated gains. There is no point in building a R100k model to solve a R10k problem.
ITWeb: Why are you presenting at the Cloud Summit in February, and what outcomes/takeaways would you like attendees to leave the event with?
Andrews: In our company, we deal with corporates across the globe. We have a strong presence in Europe but our home base has been and will always be South Africa. There is a stark difference between the uptake of cloud and, more specifically, ML in South Africa and the uptake across Europe. Some of the most innovative solutions are being found to problems that, sometimes, companies didn't even know they had. Innovations that drive accuracy, profit and sustainability.
It's time for South Africa to join the 4th revolution and understand and embrace the impact that ML can have on the business landscape across all domains - finance, media, advertising, mining, education, health and lots more. The time is ripe for South Africa to catch the revolution wave and transform the way that we do business.
Companies need to leave the cloud summit feeling energised. The uptake of these technologies is only as ambitious as the will that decision-makers and thought leaders have. We need to start having the conversations the enable our creative thinking and propel us to a new landscape of ML-enabled automation.