The deluge of stories about artificial intelligence and robots has sparked a renewed interest in the capacity of machines to work better, smarter and longer than humans.
Fuelled by the well-publicised examples of smart systems winning gameshows and trouncing a world-champion in the notoriously complex game of Go, many businesses are considering the potential of automation.
But away from the speculation about the capabilities of near-future AI and robots, what are the practical considerations for any firm thinking of going down the automation route?
The first rather obvious question for a business to ask is whether it is technically feasible to automate a particular activity, or will be in the near future, according to the consultancy McKinsey.
This question shouldn't be drawn too broadly, and should focus on individual aspects of a person's role rather than their job in its entirety.
"The right level is that of individual activities, because there are very few entire roles where all of one's activities are likely to be automated at the same time," said Michael Chui, partner at McKinsey Global Institute.
On a broad level, tasks that are suited to automation will be repetitive and routine, following a pattern of activity that can be determined by a set of rules. According to Chui, these type of activities can typically be found in physical work taking place in a stable and predictable environment, such as work on an assembly line, back-office tasks such as collecting and processing data, as well as some aspects of data analysis. Salary and skill level shouldn't necessarily be a guide when identifying automatable activities, as these type of tasks are carried out by both low-paid, low-skilled workers and by high earners who've undergone years of training, he said.
Once firms have established whether the technology exists to automate a role, McKinsey recommends firms go through the usual steps of weighing up the technology and business change costs against the project's benefits.
These benefits can extend to far more than a reduction in labor costs, also encompassing greater output, and higher quality work with fewer errors. The costs can also include regulatory barriers and the social acceptability of having an automated system carrying out a particular role, for example patients might reject a robot nurse.
"Create a heat map of where there might be potential for automation to have benefits," said Chui.
In order for businesses to identify which roles are automatable, they will likely need to first document their processes in detail.
"If you analyse all of the activities that everyone is doing in the organisation you can get a sense for which of those might be more automatable than others," said Chui.
Neil Kinson, chief of staff at enterprise process automation specialist Redwood Software, recommends creating what he calls a 'robotization centre of excellence' that collates the processes taking place across the business.
"It really is teaching people how to both document their process and turn that documentation into what effectively becomes a robot design," he said, using the term 'robot' to refer to the software carrying out the automation.
Once these processes have been captured, it also becomes easier to identify manual ways of working that aren't efficient in an automated workflow, and that can be changed or even eliminated.
"People tend to replicate the way that they work today," says Kinson, giving the example of a process that might involve placing data in a spreadsheet so it can be manipulated.
While that approach might make sense in a manual workflow, in order to avoid a person having to go into an ERP system to fetch data 30 times, in an automated system he says "the robot doesn't care. Why don't you just extract that directly from the source system?".
"It's very much the business driving the process and it's educating the business as to how to both understand what they do and see what the returns are," he said.
"You really need to look at the entire business process," agreed McKinsey's Chui.
"Sometimes there are steps that just go away if you think about the process differently. Then you understand, which of these pieces machines can do better and which of these pieces people can do better."
When reshaping business processes to fit a mix of manual and automated workers it will often be necessary to designate people to oversee the work carried out by automated systems and intervene on the rare occasion they stumble.
Given the importance of categorizing business processes to automation, firms that have already examined their activities to identify what could be offloaded to a Business Process Outsourcing (BPO) provider may have a headstart. A number of BPO providers, such as Capgemini and Wipro, are also moving into the emerging field of Robotic Process Automation (RPA).
The final step for firms is identifying which vendors they can partner with to realise the automation project and planning their roadmap for deployment, setting out pilot projects and the broader roll-out.
While it falls to individual firms to identify the particular tasks that are best suited to automation within their business, there are general guides as to the types of work that should be in the frame.
A common set of business activities with potential to be heavily automated and that are carried out across nearly every industry are back-office tasks -- the grunt work of wrangling and manipulating data, be it shunting financial information between systems or swapping business information with other people.
A good deal of the data entry and data processing work across all industries could be automated -- some 60 percent, according to a recent report by McKinsey.
"There are a huge number of people in back office functions in most corporates doing things that really are monotonous, repetitive, highly rules-based and not really using any of their thought, judgment or cognition," said Redwood's Kinson.
"We have this notion that our job [as automation specialists] is actually not to take the humans out, it's actually to take the robots out of the humans, so that they can actually turn their back on the slow manual, error-prone stuff that gets in the way of their productivity."
In its recent State of the State report, consultancy firm Deloitte predicted there is a 77 percent probability of 1.3m "repetitive and predictable" administrative and operative roles in the public sector being automated. In UK local government, such a shift would mean the number of admin roles dropping from 87,000 in 2015 to 4,000 by 2030.
Automating admin processes has the potential not only to reduce labour costs, but also to free up highly skilled workers to devote their time to more high-value tasks. For example, McKinsey says that half of the time spent by the workforce in finance and insurance is devoted to collecting and processing data.
Kinson gives the example of accounting, where "a lot of the activities that go on are simply validating balances in different ledgers in different companies in one common ERP system".
"What you want are these, in many cases highly qualified, finance people to focus on the balances that are not where they should be, and where there is a material difference do the investigation -- rather than spending 80 percent of their effort getting the data they need to find out where those imbalances are."
Cathy Tornbohm, research VP at analysts Gartner, says firms can secure good return on their investment by automating back-office activities that would be too laborious to be practical to carry out manually.
"The way I think about it is, 'If I had an army of people, what could I do better?'," she said, giving the example of a firm being able to check all of its contracts with suppliers to discover whether it's overpaying.
A wide range of software and hardware exists for automating back-office data processing roles. There are systems that will programmatically interface with modern back-office ERP, finance and CRM systems or databases -- the likes of SAP, Oracle and Salesforce -- and extract, transform and load data to achieve the same end result as manual processes they replace.
When automating older systems that don't have the necessary software interfaces for automation tools to hook into, it may be necessary to use software that automates tasks by mimicking user interaction with a graphical or text-based interface. However, this approach limits how much more efficient they can be than the manual process and also risks breaking when the UI changes, according to Redwood's Kinson. Given this increased difficulty of automating older systems, ageing infrastructure may be barrier for these types of projects.
Modern automation platforms use image recognition and automatic handling of form layout changes to be able to handle work with a wide range of third party applications, although Kinson says these claims are often overblown.
But Siddhartha Singh, head of BPO for IT services firm NIIT Technologies, predicts that improvements to back office automation platforms will see them incorporate machine learning, allowing them to automate a wider range of activities and systems.
Common technology platforms used for automating the back office today include UiPath, Blue Prism, Automation Anywhere, OpenSpan (now Pega), NICE and WorkFusion.
For automation of back-office tasks to be successful, the underlying technology platform used by manual workers generally needs to be relatively slow changing, without frequent updates to its user and programmatic interfaces.
For suitable back-office tasks, typically between 50 to 75 percent of the activities undertaken by full-time workers can be automated, according to Kinson, either freeing up those staff to concentrate on higher-value tasks or reducing labour costs.
Beyond labour costs, automation can reduce risk, says Kinson, citing a firm that during crunch times had staff working up to 18 hours a day on preparing their monthly financial data, increasing the risk of these key individuals going off ill. After automating away repetitive tasks, these individuals were able to reduce hours closer to that of a standard working day and run a revenue recognition process on a more regular basis, allowing them to identify more revenue for the firm, he said.
"The majority of those benefits are not from decreasing labour costs. They're things like increasing throughput, decreasing errors, increasing the quality of outputs," said McKinsey's Chui.
Technologies to automate back-office processes have been around for years, so why is there renewed interest in the practice?
According to Gartner's Cathy Tornbohm, this groundswell of interest is being driven partly by improvements in OCR technology for digitising paper documents, but more by the higher profile afforded to the automation by current interest in AI and robotics.
"Somebody put the word 'robot' in front of process automation. Now business are saying 'I see this as an interesting thing to explore'," she said.
As back-office automation progresses, McKinsey says systems are moving beyond processing payrolls, generating invoices and tracking bar-coded materials into new areas such as entering paper and PDF invoices into computer systems or processing loan applications.
In the field of analytics, smart systems such IBM Watson are designed to analyse large amounts of data -- so vast it would take a human days or even months to plough through -- and then answer simple natural-language questions about that huge corpus of information.
Watson is being used in many sectors with specialised information needs, including medicine, veterinary science, environmental and geotechnical engineering, education, government, food and beverage, legal, and music and entertainment.
However, most Watson deployments are still at an early stage and are expected to take months or even years of training before they are able to do useful work on a daily basis.
Outside of data processing and analytics, McKinsey identifies the greatest opportunity for automation in repetitive physical roles, including welding, soldering, food preparation and packaging objects. These type of roles are prominent in manufacturing, food service, accommodation and retail businesses, according to McKinsey. Obviously robots already perform similar roles in certain industries -- welding in car manufacturing or placing components on circuit boards for electronics firms, for example. But McKinsey believes that the use of robots in these areas will spread.
Pioneering companies like Amazon have already demonstrated the role that robots can play in the warehouse, with the online giant using knee-high bots to ferry shelves around its warehouses and supporting efforts to develop bots to pick items from shelves. Google also recently demonstrated robot arms that can share what they've learned with each other, in order for the arms to rapidly pick up physical skills such as opening doors, which would otherwise require hours of training for each arm.
The potential for greater use of robotics stems in part from recent advances in narrow fields of artificial intelligence. Advances in machine learning -- where specialised computer systems are taught to perform tasks they haven't been programmed to carry out -- are expected to broaden the use of robots to less tightly-controlled and predictable environments than they inhabit today.
"Obviously we've been automating for well over a century -- some would say two centuries -- so in some sense, conceptually this is not a new thing," said McKinsey's Chui.
"But I think what we've discovered is, with Moore's Law, with lots of people working on different aspects of these technologies -- whether they be the physical technologies around robotics, automated vehicles, 3D printing or the more cognitive tasks such as analysing data -- we're starting to see accelerating developments there."
"People have been surprised, quite frankly, at what we've seen these technologies go through," said Chui, referencing the rapid progress towards developing self-driving cars and the ability for natural language processing (NLP) systems to understand human speech.
In particular, progress in areas such as NLP promises to realise a shift towards partially automating custom relations using chatbots.
Bots are software agents that can have conversations with customers or other bots and automate handling tasks such as hotel booking or order delivery and answering simple queries. Microsoft is betting big on chatbots, with CEO Satya Nadella predicting that these bots will greatly simplify dealings between businesses and customers.
Banks and other businesses are already using the Microsoft Bot Framework to build customer service bots for Skype, Facebook, Slack and "anywhere where people are communicating", said Nadella. Meanwhile, Facebook, IBM and Google are making their own moves towards building out a chatbot ecosystem.
Early experimenters with chatbots include the Royal Bank of Scotland, which is using the IBM Watson Conversation service. The bank will begin using the IBM's chatbots with around 10 percent of Royal Bank customers in Scotland to answer queries such as 'How do I authorise my card to be used overseas?' to 'How do I update my home address with the bank'?
But Redwood Software's Kinson argues that most businesses have far simpler back-office processes they could be automating before considering more futuristic and potentially complex scenarios, such as drone deliveries or automated courier fleets.
"The returns that you get for doing this are available today. They don't require a change in the law -- if you're going to a drone delivery, for example, then you need quite significant reform of civil aviation law across most jurisdictions in the world," he said.
"This is not science fiction; this is stuff that can be done today and we have customers that, in some cases, are eliminating up to 100 percent of the effort in those processes."