When you think about machine learning and AI, you probably go to one of two places in your head. Either, the Terminator movie franchise or to the West coast U.S. tech giants (e.g. Google, Microsoft, Amazon et al).
Whilst both might strike fear into the hearts of your average citizen (for very different reasons!), the reality is that machine learning applications carry out very useful decision making based on (in most cases) pretty banal data-sets.
It is this ability to pattern match at scale by utilising evolving models that is creating such significant value for businesses. It may not quite be Skynet or DeepMind, but it is still incredibly exciting, so let’s find out more.
Global spending is predicted to reach $98 billion by 2023 on AI & ML, which is a growth of 275% from 2019.
From the automotive industry to financial services, healthcare to energy, the application of machine learning is driving everything from fraudulent payments identification to the diagnosis of illnesses from routine healthcare imagery.
The ability to take large scale data (numerical, behavioural, image, text, video & more), train it with scalable models and then make it efficient, debuggable & adaptable, is creating value, reducing cost and reducing risk.
Real world examples really help isolate just how and why machine learning applications are so beneficial in certain verticals, so let’s take a look at some examples:
By taking transactional data and training a model, the applied machine learning algorithms will have an understanding key identifiers by category within the dataset, for example, the average:
This means that you can build a real-time alert system based on any deviation from expected levels. Has there been unusually high ATM activity in Romania? Are debit cards transacting at a much higher level than they should be for a Sunday at 2am? Has the value of transactions skyrocketed in John Lewis?
The built-in pattern matching and identification capabilities in this instance are doing a job that a human would just not be able to achieve. They are also enabling banks to quickly identify fraud, pause accounts and re-issue cards without the potential for time lags, negative customer outcomes and financial losses.
Creating databases where there are none currently can be challenging, but the potential in doing so, especially for digital health is significant.
By taking a large data set made up of images of skin lesions and training a model to identify different diseases by size, shape, colour, position etc. of said lesion. The data amassed can be categorised and as the database grows, the algorithm learns not only about the characteristics within the image but also how to navigate image quality (i.e. zoom, angle, lighting etc.)
A machine learning algorithm has already been developed and tested by researchers at Stanford to do exactly this, and was proven to be as accurate as 21 dermatologists with an accuracy of 91%.
This kind of computer aided classification powered by machine learning may allow for potential patients to self-diagnose with the sensor filled, high power computing that sits within their smartphone, which will reduce the latency of referral and allow people to approach healthcare professionals with a clear understanding of their health challenges.
In ingesting, processing and categorising huge volumes of energy consumption data, machine learning models can not only address datasets that would be unimaginable for a human to contextualise, but also make data-driven recommendations that are objective and do not require key resource expertise.
Machine learning is already supporting the generation of incredibly accurate demand and consumption forecasts for the energy sector, where this data can be transformative is in setting the optimal pricing strategy based on the full understanding of supply and demand.
Training the algorithms to make recommendations based on previous data points, previous market responses and experienced complex scenarios, should mean that the recommendations are constantly refining and getting more precise over time.
This can help make counter-intuitive pricing decisions based on accurate, quantifiable data, allow pricing to be auditable, transparent and easy to monitor, alongside saving time and money by allowing pricing managers the freedom to focus on key high-level decisions.
As with software development, data development needs hypotheses to be generated, minimum viable products proved out and investment ring-fenced.
If you want to explore how machine learning applications could benefit your business operations, get in touch with our team to discover the untapped opportunities that you should explore.