​Machine learning: Making it work in the real world

Machine learning and AI are providing new options for business; here's what some leading organisations make of it.

executive guide

What is machine learning? Everything you need to know

Here's how it's related to artificial intelligence, how it works and why it matters.

Read More

Machine learning promises to help businesses automate processes and make smarter decisions based on the huge amounts of data they collect. The result could be sharper insight and better performance. But machine learning and artificial intelligence are still emerging technologies for most companies and few have got further than pilot projects. So how can machine learning best be used to create business benefit? Four industry experts give their best practice tips on making the most of machine learning.

1. Use experiments to augment human knowledge

Dentsu Aegis CIO Gideon Kay says his work around machine learning is focused on automation. The aim is to create algorithms that help the advertising specialist's customers work out what content resonates and to target information to the right audiences around the globe. One area of application is in what Kay refers to as 'media-mix modelling'.

"If you want to advertise a product on television, then understanding the factors that influence demand in certain audiences at certain times can help you know which slots you want to book in a schedule," he says. "We're using machine learning to build-out algorithms and capabilities that will support our analysis in that space."

See: Special report: How to implement AI and machine learning (free PDF)

Kay says Dentsu Aegis is using a range of machine-learning technologies in this process. Some of these services come from the firm's cloud partners, including Microsoft, Amazon and Google. Dentsu Aegis has also set up a global innovation centre in Singapore, co-funded with the Singaporean government, to help build machine-leaning capabilities.

"Machine learning is at an embryonic stage. A lot of people are saying big things about the application of the technology -- but when you look under the bonnet, they're not actually doing it. We're using the technology that's there -- we're experimenting and our work is being used but not at scale. It's somewhere between exploratory and mainstream," he says.

"What we know is machine learning is about thinking how you build a model, which you train and allow to work through significant data volumes and sets to give your organisation some business outcomes. A lot of these technologies and capabilities are about augmentation, rather than replacement of human capability."

2. Investigate the art of the possible now

Richard Corbridge, chief digital and information officer at Leeds Teaching Hospitals Trust, says there is pressure on healthcare leaders to focus on basic IT rather than the "shiny things". In this context, he wonders whether the advent of machine learning and the ability to deploy AI is a must-do.

"Machine learning and the evolution to AI in healthcare will give us the ability to make digital the healthcare actions that are needed to deliver care," he said.

Corbridge says these potential benefits mean the NHS must investigate now what the art of the possible means in terms of machine learning. It is something he is already pioneering at Leeds. Hospitals are scanning large volumes of patients' paper clinical records and storing them in electronic-health records.

However, searching through these records remains time consuming as a patient with a long-term condition might have hundreds of scanned files in their records that clinicians then need to scroll through to reach the piece of information that they actually need. Corbridge says the Trust's first foray into machine learning will be a robotic process-automation development which aims to make it easier to find the right medical records among hundreds of scanned documents, saving time for clinicians.

"It will give us the ability to ask a piece of machine learning to retrieve the right scanned paper notes for each patient, in the context of the clinical speciality, the date and of course the patient, with all this information sent to a mobile device. Once this is deployed, we will then be able to gauge the excitement for machine learning in all that we do," he says.

"Machine learning and the advent of AI into healthcare will bring the delivery of assisted healthcare and inspire a move to augmented healthcare provision that will drive the path to safe and autonomous care that can then become automated."

3. Find an answer to a problem statement

Rob McLaughlin, head of digital decisioning and analytics at Sky, says the television and telecommunications specialist is using AI and machine learning as part of a leading-edge approach to data that uses advanced technology to augment human expertise. The aim is to make smarter decisions about the products and services Sky offers to its UK customers.

"There must be a problem statement - and the most obvious statement for us is we have 11 million customers and their requirements are all unique. We must think about how we embrace that uniqueness, rather than trying to fit them into a box. That means we need to be able to make 11 million decisions at any point in time. Machine learning is a great way of trying to make those decisions as effectively as possible," McLaughlin says.

See: Big data in action: AI, machine learning, cloud, IoT, and more

Companies choosing between machine learning and AI need to think about whether they're using data to make big, strategic decisions or micro, customer-level decisions. "In my department, we care more about micro-level decisions -- that's my team focus; we make millions of decisions for the customer," he says.

"But if I was tasked with helping the business decide which product to deliver next, or which content we should buy from a provider and how it will affect our customer relationships, then I would use machine learning. The aim there would be to use machine learning to deliver insight, rather than action."

4. Build machine-learning capability in-house

Michele Goetz, principal analyst at Forrester, says most firms are already looking at how to make the most of advanced technologies. The researcher says 51 percent of companies have implemented AI capabilities, with another 20 percent planning to in the next 12 months. Goetz says the top perceived benefits are improved customer experience, online security and operational experience.

"These benefits are typical of traditional technology objectives," she says. "However, companies that are obtaining competitive advantage with AI and machine learning are thinking more strategically about their business model and product innovation. The CIOs at these organizations recognize that AI and machine learning are not off-the-shelf solutions.

Rather than thinking sourced commercial offerings will provide a competitive advantage, Goetz says smart digital leaders are investigating how to build in-house capability. She says the successful sourcing of budget will look more akin to a start-up investment strategy rather than the total cost-of-ownership and return-on-investment model of traditional enterprise IT.

"These CIOs recognize that their technology platforms need replacement, their development teams need to include data science in agile development processes, that they will build systems rather than buy systems, and that their strategy to acquire needed technology, data and analytics and talent will come through alliances over purchasing and partnership models of today," says Goetz.

FURTHER READING ON MACHINE LEARNING