Skip to main content

Making machine learning work for your business

(Image credit: Image source: Shutterstock/Sarah Holmlund)

David Mytton, founder and CEO of Server Density (opens in new tab) , looks beyond the hype to the useful applications of machine learning in practice 

Google ‘machine learning’ - I dare you. Machine learning is the ‘big data’ of our time, hopelessly hyped up with outlandish applications for any and every application and business sector under the sun. Machine learning is ranked as a buzzword (opens in new tab)by Gartner, at the very top of their peak of inflated expectations; in this excitable environment how can businesses cut through the noise and find practical applications for these novel technologies?   

The key is to realise that ‘machine learning’ by itself doesn’t actually mean very much: it’s a stand-in phrase for a diverse array of technologies which, broadly speaking, mean computer systems learn from feedback loops and improve themselves. Much ink has been spilled on predictions for how machine learning technologies can be applied, ranging from calculating the best time to serve advertisements to predicting the risk of heart failure, but this isn’t exactly helpful for the average business.     

In this post I’ll provide an overview of how machine learning can be leveraged for the average technology-driven business, and the technologies you should become familiar with - and hire for - in order to drive real machine learning innovation in your business.   

Applications for modern businesses 

Solutions like IBM Watson, and the constant media buzz around AI, machine learning and the like, have led too many people to believe that the only possible applications for machine learning technology are ones that have to be revolutionary or game-changing and only suitable for businesses that have vast sums of money to spend. However, as is often the case, truly innovative and important solutions are found in surprising places.     

Google provides an interesting example (opens in new tab): unsurprisingly one of Google’s primary cost outlays is the upkeep of their data centres, and reducing the energy demands of those data centres is a key cost reduction challenge. To tackle this problem - itself not the typical ‘sexy’ challenge machine learning is often brought in to solve - they began building a machine learning model to predict and improve data centre performance. After a standard phase of trial and error, their models can now predict data centre energy use and efficiency with 99.6% accuracy - very impressive. 

It’s all fine and well for a corporation like Google to experiment with these technologies - they can afford to invest in innovation and experiment with new technologies much more than smaller players. Or so you would think - previously it would indeed have been considered to be impossible for a small or medium sized business to keep up with the likes of Google or Amazon, however things are moving rapidly in this market. The technology needed for machine learning is maturing rapidly, becoming more commoditised and packaged into products that smaller businesses can pick up and use as they wish.   

Practical tech for practical purposes   

So what are the technologies most often used in machine learning implementations?  Emerging from the machine learning team at Google, TensorFlow is a framework for the implementation of machine learning at scale. Open-sourced in 2015, TensorFlow has changed the game for businesses looking to take advantage of machine learning, and is used by companies from cutting-edge startups to established businesses like Ocado.   

Characteristically, Amazon have also jumped into the competition with their own solution called Amazon Machine Learning (opens in new tab). AML differs from TensorFlow in a number of ways: with TensorFlow, you build your own models and can then execute them against your datasets wherever you like whereas AML requires you upload your dataset to Amazon then use their API to execute queries. AML is essentially a hosted, machine-learning-as-a-service product whereas TensorFlow is a framework to build your own models and control everything yourself.   

Of course Amazon aren’t the only people looking to provide machine learning “as-a-service” - Microsoft’s Azure Machine Learning Studio (opens in new tab)helps users visualise their algorithms then executes them in their Azure Cloud, and there are a host of technology start-ups including Algorthmia, BigML and MLJar aiming to offer machine learning through an API or SaaS application.   

The important point is that you don’t need to be an accomplished data scientist to leverage these technologies in your business. In the same way that public cloud has transformed how organisations manage their infrastructure, these powerful tools are making machine learning more accessible and affordable for everyone.   

Look at everything with your use case in mind   

Whatever approach you use, it’s of vital importance that you choose a platform that’s open, extensible and compatible with the best services. The machine learning API ecosystem has matured rapidly over the past two years, and there’s currently a huge range of potential applications available to businesses at little cost aside from development time. It’s important to not get overwhelmed, and look at all the options through the lens of the specific use case your business is considering.   

You must also decide whether you want a “product” which provides its functionality through machine learning e.g. tools like Amazon Lex (opens in new tab)(for chatbots), Amazon Polly (opens in new tab)(text to speech) and Amazon Rekognition (opens in new tab)(image recognition) which are easy to integrate and have a wide range of potential uses.   

You might want to build sophisticated features into your own products which are powered by APIs that are themselves powered by machine learning. For example the Google Cloud Video Intelligence API, (opens in new tab) currently in beta, is a recently announced advanced tool for recognising objects in videos. On the other hand, you might want to build your own machine learning models and train them yourself, such as with AML or TensorFlow. The key first step is to think critically about exactly what your problem is, the development resource you have, and really do your research and assess what kinds of solutions would help you get there. Machine learning may be a buzzword, but that doesn’t stop the technology from having a multitude of useful applications to businesses of all types. 

David Mytton, Founder and CEO, Server Density (opens in new tab)

Image Credit: Sarah Holmlund / Shutterstock

David Mytton is founder and CEO of Server Density, a scalable infrastructure monitoring software company. Server Density offers a SaaS product featuring the graphing, dashboards and low management overhead that modern businesses need. Server Density has more than 700 customers, including the NHS, Drupal, Firebox and Greenpeace, and has offices in London and New York.