There’s a new wave of work automation now starting to impact businesses. Yes, partly this is enabled by Machine Learning technologies and techniques: but there’s more to it. The new wave of automation brings advances in interaction, insights and integration. This report lays out the landscape.
Written by MWD Advisors’ lead analysts Angela Ashenden, Neil Ward-Dutton and Craig Wentworth, this report is the first of a two-part series.
The new wave of automation in business is about more than machine learning. The ‘mainstreaming’ of learning systems techniques is enabling lots of experimentation and many new tools and applications; however there are three distinct layers in the new wave of automation that’s now washing over businesses, and learning systems are only the major factor in one of these layers.
You need an automation strategy. The new wave of automation is bringing many new low-cost, self-service tools into the picture for your business, as well as more complex tools and platforms that may have the potential to radically affect your business processes and operating models; without a clear strategy for how you’re going to engage with new automation tools and approaches, you’ll come unstuck very quickly.
When creating your strategy, your unit of analysis and change has to be tasks (not jobs). It’s easy and tempting to assume that automated systems will substitute for human jobs but the much more accurate way to think about the impacts of the new wave of automation is to think about how different kinds of automation tool and technology have the potential to impact particular tasks in your business.
Prioritise for the biggest impact. Go for either the high-volume tasks or high-value tasks. The other areas will take care of themselves as innovation ‘trickles down’.
Think beyond the technology. Even learning systems need to be trained, and in many cases you’ll need to develop and apply some specialist skills to really get the most out of the technology. You have to thoroughly understand the business domain you’re working in, the nature of individual tasks, and have a very intentional approach to how you insert the technology into work. Domain knowledge in particular is crucial because – for the time being, at least – even the most sophisticated AI needs boundaries within which it can do specialised work.
A new wave of automation
In the past two years there’s been a major uptick in announcements of new ‘intelligent’ services and systems available to both businesses and individual consumers. If you’ve not been exposed to discussion of the capabilities of Apple’s Siri or IBM’s Watson in the media, for example, you’re in a very small minority. These are just two examples at the forefront of a wave of technology and product advances.
It might seem like automation in the workplace is a relatively new phenomenon, but in reality it’s been going on for over 200 years. In 1785, the first automated process, in flour milling, was implemented; 1892 saw the first automated telephone switchboard being used.
Until relatively recently, the waves of automation that have washed over businesses since the late 1700s have shared an important common feature of compromise: we’ve had to deliver and operate automated systems within very tightly-configured and controlled environments. This is obviously true in the context of physical process environments like assembly lines, product packaging lines, automated warehouse picking systems and so on; but it’s also true in the context of software-driven processes that are heavily automated – like trade settlement in banking, service activation in telecoms and bill production in utilities, for example.
So in a sense, the history of automation design is as much about the design of the environments in which automation occur as it is about the design of automations themselves. Think about automation in manufacturing processes and production lines; think about early business systems with crude ‘operator interfaces’. These systems required us to shape our work around the needs of the automation.
In 1959, though, a new course began to be set with the introduction of the first learning program (a chess-playing program invented at IBM). Now, helped by advances in learning systems techniques, a huge shift in the economics of computing and digital storage and changing expectations around the ease-of-use of software systems and services, we’re finally seeing the transformation of this relationship: technology systems are starting to be able to mould themselves around us and our work, and the interfaces they present to us are becoming more and more flexible, natural and intuitive.
What’s being created here is a new wave of automation that will have profound impacts on how people accomplish tasks in the workplace, and on how businesses arrange, distribute and manage work.
The new ‘smart components’ that are beginning to impact business today – chatbots, ‘intelligent’ assistants, and so on – are not really about automation of work or processes in the traditional sense that we’re used to. Yes, the new technology capabilities now entering workplaces are delivered by software that ‘automates’ certain activities; but the endgame and experience is not one of a completely self-contained automated system that works in isolation on an end-to-end business process; but rather of software agents of various kinds working alongside people to augment particular work tasks and processes.
What’s behind it? Machine Learning and more
Advances in Machine Learning (ML) software techniques are the big headline-grabbers that are contributing to the creation of this new wave of automation, but that’s not all that is happening in the technology space – for example, innovation and interest in ML is being accelerated by, and in turn is further feeding, rapid development of big data management and predictive analytics technologies, and the use of these to drive activities and business processes.
What’s also important to recognise is that many of the underlying technologies powering the new wave of automation are not new. Pioneers like Nuance (speech recognition and transcription, speech generation), SAS Institute and SPSS (predictive analytics) have been developing and delivering key component technologies for 20 years.
When we take a broader view, we see that there are three main forces that are together driving today’s new wave of work automation in business (as shown in the figure below):
- Rapidly maturing technology. Firstly, the affordability of high-speed parallel processing that underpins many ML techniques has been transformed by the mainstream availability and packaging of specialised Graphical Processing Unit chips (GPUs), which first appeared in high-end PCs and gaming consoles. And also, there’s now a burgeoning group of open, free SDKs and programming toolkits available (from the likes of Google, Facebook and Microsoft, as well as from smaller specialist providers) that make it much easier for teams to experiment with data and ML applications running against these systems. Equally important, there’s rapidly increasing availability of massive online digital data sets for training these learning algorithms, as well as configuring other kinds of predictive analytics driven by statistical models (think image libraries, digitized speech, online videos, and so on). At the same time, widespread implementation of open APIs for mainstream online tools and platforms (think Twitter, Facebook, LinkedIn and so on) means that access to massive data sets from many sources is much easier and cheaper than it used to be.
- Business pressures. Firstly, there’s the widespread desire to improve customer experiences which is leading businesses to seek ways to ‘act small, at scale’ – using customer knowledge to create much more personalised, relevant services. Secondly, there’s the need to manage knowledge better in dynamic workforces and dispersed operations – decoupling knowledge and insight much more effectively from individuals. Thirdly, there’s the need to drive operations quality and efficiency – finding ways to deliver better, more consistent services to customers, partners and suppliers.
- Familiarity. Many people, particularly in highly developed countries, already have familiarity with and expectations surrounding work automation and augmentation because of their engagement in a range of services that have been developed by earlier adopters of machine learning, predictive analytics and rules technologies. For example, people working in software development teams are already somewhat likely to have come across automation tools and platforms like RightScale, Chef, Puppet and so on. In the wider workplace, team communication and collaboration environments now feature bots and intelligent assistants. In our experiences as consumers, most of us are familiar with automated recommendation features in e-commerce platforms like Amazon’s, as well as the capabilities of assistants like Apple’s Siri, Amazon’s Alexa and Google Now.
Source: MWD Advisors
It’s important to realise that although learning-system technology advances are a big part of the picture, they don’t contribute equally to all areas of this new wave of automation we’re seeing. In fact, there are three distinct layers of change that play in the new wave of automation – interaction, insight and integration – and only the top layer (interaction) is really having an impact principally because of learning-system technology advances in and of themselves. The other two layers (insight and integration) are more part of the new wave of automation picture because of changing business pressures and changing expectations.
The figure below shows these three layers of change and how they relate. We look at each layer in turn below.
Source: MWD Advisors
Interaction layer improvements: sensing and responding in more human ways
Interaction between people and computers in the 1960-1980 period, as computers first entered the business mainstream, was dominated by ‘operator interfaces’ – very low-fidelity, often only barely interactive, tools for issuing instructions and receiving calculation results. Later in this period we moved towards real-time interaction as the norm, but still operators were required to be trained on specialised terminal applications. As Microsoft Windows began its march across business in the mid-late 1980s, we moved towards graphical user interfaces (GUIs) that required less training to use – but still, every application required training for users to be effective.
Now, though, we’re seeing increasing diversity in human interfaces for software systems and services – not only web-based graphical application interfaces that are based on traditional business software design concepts like grid-based data forms, action buttons and so on; but also much more free-form graphical interfaces, mobile apps, notification streams embedded into collaboration and messaging apps, speech-based interfaces (like Amazon Echo and Alexa, Apple Siri, Google Assistant and Microsoft Cortana) and attempts to create synthetic human avatars in software (like IPSoft’s Amelia).
It’s interest in new ‘conversational interfaces’ that are really creating most of the buzz in domain of human-computer interaction today. The main technologies underpinning new conversational interfaces are today concerned with interpreting and generating text and speech (and less so, with interpreting images and video). But innovations aren’t only enabling new interaction media; they’re also enabling less-structured interactions between people and software systems. Crucially, the new interfaces now starting to proliferate are increasingly designed to serve the needs of people who might not necessarily know anything at all about the systems they’re interacting with. Systems fronted by synthetic human avatars don’t need user manuals.
As the figure above indicates, learning-systems concepts like ML and Deep Learning are really the driving force behind a great deal of the interaction layer improvements that we’re now seeing in new software products and services.
Insights layer improvements: interactive, predictive, advisory analytics
Despite the current buzz and hype being focused on the tangible front-ends of new interaction approaches and interface types, there’s a parallel shift in the application of ‘insights technologies’ that are also important for every technology strategist or business analyst looking at the new wave of automation to understand. New applications of existing insights technologies are often (though not always) key components that power the behaviour of the intelligent assistants, agents and systems we’re increasingly familiar with.
In the insights layer, we see changes to the applications of technology happening in three specific ways, all related:
- A move away from the use of analytics tools by highly-trained professionals who create and configure systems and interpret results in isolation; towards the dynamic integration of analytic models into other business applications, and the real-time provision of insights that inform operational tasks and decision-making.
- A move beyond the use of tools to analyse data retrospectively, to understand historical patterns in data; to drive predictions through the use of things like classification models (Is this person likely to be a part of the set of people who upgrade?) and regression models (How much is this customer likely to spend over the coming year?).
- A move away from the production of static results (presented in graphs, tables) to drive ‘offline’ managerial decision-making (for example, informing planning decisions around a new quarter’s promotional campaigns) to the production of large numbers of individual, operational recommendations that are directly presented to operational staff in the context of their digital workflow and workplace environments.
These shifts in the insight layer are partially powered by the influence of learning-systems approaches (for example, through the use of varieties of neural networks to create trainable predictive models). However learning-systems techniques are only one part of a much broader story that is more appropriate to see as being about the operationalisation of advanced analytics.
Integration layer improvements: addressing the long tail
Whereas the shifts we see in interaction (enabling systems to sense and respond in more human ways) and insights (operationalising advanced analytics) are at least partly driven by learning-systems approaches and technologies, the third layer we see as part of the new wave of automation – the integration layer – is today almost completely uninfluenced by the learning systems movement. This doesn’t mean that what’s going on here can be ignored or should be considered separately from advances in interaction or insight, though.
A significant shift in the focus and packaging of integration technologies is having a dramatic effect on the art of the possible in this layer, just as learning systems are having a dramatic effect on the interaction layer. Here, the shift in focus and packaging is away from IT-driven, programmed systems – architected and sold so that they support large-scale, high-volume, centrally-controlled integration scenarios – and towards more business-driven platforms, offering highly visual design tools, that enable teams and departments to rapidly deliver integration value using the more modest resources they have at their disposal.
This shift in approach brings a way for organisations to address the ‘long tail’ of integration needs – conducting projects that would never get a high enough priority from a central IT function – in a way that delivers business value.
In part 2 of this report, The new wave of automation: Unpicking the business impact, we provide a simple framework for thinking about the potential for new applications of automation technologies to impact organisations like yours (or your customers’), and then we work through six different example technology applications, analysing their nature and potential impact of each.