For years, statistics were commonplace in business decision-making processes, but organisations are now shifting toward machine learning and predictive analytics.
While data science focuses on the extraction of information from data, predictive analytics focuses on using data to predict and prescribe behaviour. The more information gathered on past behaviour and occurrences, the better the causes and effects of these occurrences can be understood.
This, in turn, allows businesses to recognise when a similar situation may arise, and how best to handle it. The more information or details collected and analysed, the more specific and accurate models and predictions will be.
‘Machine learning has a wide range of applications,’ says Martin Moran, EMEA SVP and GM at InsideSales.com. ‘It can track disease outbreaks, help marketing teams determine the success of a campaign, or inform sales teams about how likely it is that a particular lead will buy from you.’
Machine learning and its cousin artificial intelligence (AI) have been romanticised since the 1970s, and have repeatedly failed to deliver on the dream of computer systems that can actually think for themselves.
Machine learning, perhaps, is the simplification of the grand AI dream, and bases its premise on the idea that computers can learn, which surely ought to be easier than actually thinking.
There are many approaches to machine learning, but many try to mirror nature and specifically the brain.
Indeed, sophisticated algorithms can learn – either by being trained or by automatically classifying data into relevant ‘groups’ of interesting data points. And the biggest improvements have come over the past decade.
‘Advancements have resulted from the application of huge amounts of computing power to the enormous amount of data available to the major web providers like Google, Amazon and Netflix,’ says Simon Crosby, CTO and co-founder of Bromium. ‘They use the technology to learn how better to serve you ads, offer you more relevant products to buy, and display movies you’re more likely to enjoy.’
While seeming like something dreamt up in 1990s sci-fi movies, the origins of machine learning actually stretch way back to the 1950s and early computer science research.
Neural networks, a machine learning technique that mimics simple structures in the human brain, started with the development of perceptron models. Back then, scientists used what would later be recognised as early machine learning methods to work out how to teach a computer to play draughts.
During the intervening 60 years, we have seen huge increases in processing power, allowing computer scientists to solve much more complex problems, much faster – and it’s only recently that much of this research has moved from academia to full-scale enterprise products.
‘It’s really the web that has been the game changer,’ says Jamie Turner, CTO at PCA Predict, ‘giving us access to vast amounts of data to train the models and help them learn.’
Fulfilling the promise
As well as the rapid increase in available computing power, recent advancements have mainly been driven by a sharp reduction in cost. ‘Alongside the availability of access to big data and labelled data sets,’ says Ben Taylor, CEO of Rainbird, ‘this has allowed these models to begin to live up to their potential.’
According to analyst firm Gartner, smart agents will facilitate 40% of smart interactions by 2020. In just a few years’ time, businesses systems will be able to respond to requests spoken in natural human language.
It’s hard to predict just how much time this will save employees, but it will certainly enhance workforce productivity.
‘With machines learning basic tasks, businesses will be freed up to have their employees perform more valuable roles, such as interacting directly with customers and planning for the future,’ says Antony Bourne, global industry sales director at IFS.
In fact, many people make use of machine learning technology every day without even realising it, through applications like Apple’s personal assistant Siri, Google Translate and Facebook’s facial recognition.
What’s different now, though, is that the frameworks developed by big innovators such as Google and Facebook have been open-sourced, so any organisation can build systems based on them.
As well as being a bellwether of a data landscape shifting towards machine learning, this move will help disrupt the status quo, where high-tech solutions have often been the preserve of big companies.
‘Start-ups in particular have struggled to benefit from big data, lacking expertise and the cash required,’ says Turner. ‘However, services such as this new platform will help bridge the gap and allow small businesses to become much more data savvy.’
Test of skill
When it comes to applying machine learning techniques to enterprise environments, a highly specialised skill set is required.
Typically, engineers in this space need a strong background in data science and mathematics, as well as potentially neuroscience and philosophy. As these people come together to solve problems and provide solutions, the field is advancing.
Machine learning encompasses a large set of different techniques, but broadly the principle is to solve problems without specifically programming a computer.
‘Because of the specialist nature of the task and the generalist nature of the solution, it makes sense for enterprise to turn to outside solutions,’ says Taylor.
It is possible to automate entire enterprise functions, but it might require significant changes to the underlying architecture for some applications, so it’s likely to take some time before it becomes readily available.
Some key processes can be automated immediately, however. Expenses, for instance, can be almost fully automated by software-based machine learning, which can recognise employees’ location and past trips and use this information to automatically populate expense sheets.
Professional service-based organisations such as engineers can also use machine learning software to automatically understand the potential profitability of opportunities, or plan projects based on previous assignments that had a similar profile.
‘It is early days, but machine learning has the capability to transform certain sectors like professional services,’ says Thomas Staven, director of corporate product management at Unit4. ‘This is because machine learning will automate many tasks, freeing up time for highly skilled employees to devote their time elsewhere.
‘The goal should be that employees use 100% of their time on their profession, not supporting administrative tasks. This will help businesses compete in ways that were not open to them previously with, for instance, premium service levels or lower fees.’
The early application of machine learning will be in relation to data management and integration. Most corporate data is fraught with errors and omissions, which is problematic when trying to derive business value using traditional methods.
Machine learning provides extremely effective systems to order, cleanse and integrate unruly source data. This will be a focal point for most enterprises that have struggled with outdated data warehouse tools.
But only when quality data integration is achieved can machine learning then be deployed to investigate and even create new hypotheses of business relationships and help users anticipate potential issues and opportunities.
‘For the next few years, the current application of machine learning, such as yield management and pricing, will continue to be applied, but increasingly we will see it spread into inventory and risk management,’ says Keesup Choe, CEO at PI.
However, the most useful, robust machine learning methods require a fundamental redesign of systems architecture, including massively parallel and scalable hardware and software architectures that are a challenge for most organisations to implement or retrofit because of legacy issues and cost.
The science and mathematics of machine learning is also highly advanced and specialised, so the pool of talent with deep understanding of the concepts and the ability to translate these into useful, working computer code is very small.
Further to this, the commercial cloud-based machine learning services are too generic, Choe adds. ‘They take a lot of time and effort to specialise in order to solve the specific problems for a given organisation. The performance of these services to handle data at scale is not quite there.’
One challenge that organisations shouldn’t worry about, however, is any perceived threat to the role of humans in the workplace. From a business standpoint, machine learning will provide greater productivity, reduce costs and augment results. But this doesn’t mean people will lose their jobs.
Since the Industrial Revolution, technologies have automated human labour, and people have always worried that the need for their particular skills would be eradicated.
Everything that computers have ever done has automated the application of computation, delivering vast new benefits to society.
‘Yes, some manual skills will be automated in the coming years – this is the same as the last 20,’ says Crosby. ‘However, there is a worry that this time around computers hold the upper hand in some way. I believe this to be untrue.’
Even the smartest machines are going to need teachers, and those teachers will always be human experts. Moran adds: ‘People simply won’t be comfortable making huge purchases or decisions with a robot or machine – they will always want a human on the other end of the phone.’