ANALYSIS of more than 400 specific use cases across 19 industries and nine business functions shows that application of deep learning techniques based on neural networks can improve performance in a broad range of business challenges beyond what is achievable using more conventional analytics techniques.
Three neural network techniques alone could enable the creation of between US$3.5 trillion and US$5.8 trillion in value annually, or 40 per cent of the value of all analytics. These advanced techniques, while still young, are widely applicable across the economy. Business functions that show the greatest potential for value creation include marketing and sales, and operations, including supply chain, logistics, and manufacturing.
Capturing the value potential of AI requires organisations to solve both technical and organisational challenges.
A new discussion paper by the McKinsey Global Institute (MGI) and McKinsey’s Analytics Practice finds that the most advanced forms of artificial intelligence — deep learning based on neural networks — can play a transformational role in business, with a wide range of practical applications and the potential to capture trillions of dollars in value. The paper, Notes from the AI frontier: Insights from hundreds of use cases, draws on a library of more than 400 existing and potential uses of analytics and AI across sectors and business functions that was compiled by MGI and McKinsey Analytics. It included deep learning techniques that use artificial neural networks—techniques that are in the vanguard of AI today—and maps the specific techniques, problem types and data required to solve these problems, as well as estimates the economic potential of these techniques, across industries and business functions. In all, the cases cover 19 industries, from aerospace and defence to travel and the public sector, and nine business functions ranging from marketing and sales and supply chain management to product development and human resources.
Among the key insights are that these techniques, while still young, can provide an incremental lift in performance above and beyond that from traditional analytics techniques. Indeed, the paper estimates that these AI techniques could generate as much as 40 per cent of the total potential value that all analytics techniques could provide. While they are applicable throughout the economy, certain areas have the greatest potential for creating value -marketing and sales, and operations, including supply chain, logistics, and manufacturing.
“We have explored the techniques at the practical frontier of AI, and find that while adoption will take time, they are applicable to today’s problems,” said Michael Chui, a partner at the McKinsey Global Institute who led the research. “Our use cases demonstrate just how powerful deep learning can be. It has the potential to be applied across all sectors of the economy and with considerable value.”
In all, the deep learning techniques that are the focus of the paper — feed forward neural networks, recurrent neural networks, and convolutional neural networks — together have the potential to create between $3.5 trillion and $5.8 trillion in value annually, MGI estimates. Within industries, that is the equivalent of between 1 and 9 per cent of 2016 revenue.
Neural networks are a subset of machine learning techniques. Essentially, they are AI systems based on simulating connected “neural units,” loosely modelling the way that neurons interact in the brain. Computational models inspired by neural connections have been studied since the 1940s and have returned to prominence as computer processing power has increased and large training data sets have been used to successfully analyse input data such as images, video, and speech.
AI practitioners refer to these techniques as “deep learning,” since neural networks have many (“deep”) layers of simulated interconnected neurons. Prior to deep learning, neural networks often only had three-five layers and dozens of neurons; deep learning networks can have seven to ten or more layers, with simulated neurons numbering into the millions.
Potential practical applications are numerous, the study finds. One application of AI that is applicable across multiple sectors is predictive maintenance, since deep learning is able to analyze very large streams of high dimensional data, from vibrations and other sensor data, to audio and image, that can provide insight into the operation of various types of machines. This means the systems can be used to detect anomalies and predict failures, thereby reducing downtime and operating costs.
Other uses of AI techniques include optimising logistics, for example by optimising routes of delivery traffic and even coaching drivers in real time to reduce fuel consumption. Customer service management and personalisation of sales and marketing “next product to buy” offers are a third practical application that is already significantly lifting sales conversion rates at some companies.
These deep learning techniques require vast quantities and a large variety of data to “train” algorithms. Moreover, this training data needs to be refreshed frequently; the research finds that one in three of the AI systems requires model refreshes at least monthly and sometimes daily. But effectively capturing value from AI requires that practitioners overcome not only technical challenges, but organisational ones as well.
“We know that adopting these advanced AI techniques is a significant organisational challenge for companies, but we can also see from this research, and from our work with clients, that the value they can generate, when organisations bridge the ‘last mile’ problem of connecting superior insights to changes in the way an organisation operates at scale, makes the effort more than worthwhile,” said Nicolaus Henke, a McKinsey senior partner and global leader of McKinsey Analytics. “Ultimately the value of AI is not to be found in the models themselves, but in companies’ abilities to harness them.”
Other limitations highlighted in the discussion paper are difficulties in explaining results generated by neural network techniques, and potential bias in data and algorithms. Societal concerns and regulation can also constrain their use. While there is economic potential in the use of AI techniques, the use of data must always take into account data security, privacy, and potential issues of bias.
James Manyika, a McKinsey senior partner in San Francisco and chairman of MGI, said the study was instructive in highlighting the transformational nature of advanced AI techniques in myriad ways already today. While some previous MGI research has shown that many companies are reluctant to embrace AI because of worries about technical complexity and return on investment, the economic value and practical applications highlighted by the study are compelling. “The scale of the potential economic and societal impact creates an imperative for all the participants—AI innovators, AI-using companies and policy-makers—to ensure a vibrant AI environment that can safely capture the economic and societal benefits,” Manyika said.