Machine learning — sexy and it knows it. Post 3 of 3.
But I don’t want to win a board game, I want to win at business.
Having explored what Machine learning is and why it is flavour of the month, this final post will consider the potential benefits to organisations.
The million dollar question is can machine learning benefit more traditional organisations? Or is it reserved for the likes of Facebook and Google? There is a lot of hype here and care needs to be taken to ensure that a genuine business problem is being addressed. That said, machine leaning certainly can form part of a solution to a valid analytics problem, and in fact has been doing so for many years (we just didn’t call it machine learning back then).
Financial Services
Sexy tech always finds a home in Financial Services (I guess because they are the ones with the cash, and a whole industry of consultancies trying to flog them solutions to problems they didn’t even know they had). There are some very elegant examples of machine learning being applied to financial services in a digital setting to create sleek customer-centric products. Robo-Advisors is one of my favourites (mostly because I always picture a line of robots on the phone shouting “buy” or “sell”. Which probably doesn’t happen).
In reality, robo-advice refers to automated portfolio management where investments are allocated against investment classes based on the risk appetite of the individual investor and the real time market conditions. Limits are set based on a customer’s individual risk appetite, which is normally measured via a set of questions posed to the customer. Machine learning algorithms such as support vector machines and artificial neural networks are then used to model and predict breaches to risk lists and proactively shift funds between different asset classes to provide maximum returns within the individual risk tolerance of the investor. All in real (or near real) time. Snazzy.
An application of machine learning even closer to my heart is real time fraud detection. You know those automated calls and text messages you get from your credit card and banking providers when you’ve drank a bottle of wine and had a massive online shopping splurge and they assume that you must have been the victim of fraud because no sane person would ever spend £2000 on shoes in a single transaction? That’s machine learning. Algorithms are used to model your typical patterns of spending and then monitor your transactions in real time for behaviour that deviates from this. As you provide the model with feedback (by confirming or denying that fraud has taken place) the model learns and uses this to improve the accuracy of future alerts. (And, in my case, comes to understand that any online purchases that are preceded by a transaction at a bar are likely to be genuine).
From farming to fashion
What if you don’t work in the deeply fulfilling corporate sector? Well there are examples of machine learning being successfully applied to solve problems in industries as diverse as farming and fashion. In dairy farming machine learning algorithms are used to determine whether to retain cows in the herd or remove them based on their output. In arable farming machine learning is used to optimise decisions on crop selection by predicting yields; early crop disease detection and classification; and determine when and by how much to irrigate crops. In developing environments where sustainable agriculture is essential to meet the demands of a rapidly growing population, technologies such as these are crucial.
At the less worthy end of the scale, machine learning has recently begun to disrupt the fashion industry. Recommendation engines that learn from customer feedback on whether they like or dislike an outfit have been around for some time. Now apps that measure how fashionable your choice of outfit is using a combination of image recognition and machine learning are in development. In design, neural networks are being fed existing fabric print patterns and used to develop appealing new ones.
Not just for the unicorns
One of the common misconceptions about data analysis in general, and advanced techniques like machine learning in particular, is that they are costly to implement and require highly specialist expertise to wield. Now don’t get me wrong, if you are building a massively complex model that will underpin a corporate pricing strategy or determine drug discovery in the pharmaceutical industry, then you do absolutely need someone who knows their self organising maps from their learning vector quantisation, and probably some pretty robust architecture to handle the level of computation required.
However, one of the phenomenas that has accompanied the emergence of advanced techniques like machine learning is the overall democratisation of data analytics. Many of the tools wielded by data scientists are open source. Yes you need the requisite skill to use them, but enough knowledge and skill to build simple models can be gleaned from free online content from providers like Coursera, General Assembly and many others.
In 2015 Amazon launched a cloud based machine learning proposition which provides visualisation tools and wizards to enable non data scientists to gain insights from their data with minimal investment in infrastructure and skills. Thus small business owners and start ups should not feel precluded from using data analysis generally or machine learning specifically where it could be used to solve genuine business problems. For example, customer segmentation; predicting cash flow; brand sentiment analysis; and website content personalisation.
Conclusion
Machine learning can be a powerful force for good, but decision makers need to recognise it is not a strategy in and of itself. It (like data analysis more generally) can potentially form part of the solution to a (genuine) business problem. The challenge is clearly framing the problems and the desired outcomes in the first place. Otherwise, like so many of the recently hyped advancements in the field of data, it just becomes tech looking for a problem to solve (shudder). That said, the rapidly decreasing cost of computation, accessibility of cloud based wizard-driven solutions and media love affair with machine learning have laid the foundations for this sixty year old technology to play a key role in the continuing evolution of data analysis.
(PS — I have deliberately kept this series frothy and not touched on anything deep like the role of machine learning in the field of Artificial General Intelligence. Mostly because it scares the shit out of me. If you want to never sleep again, read this. Then this).
