Robots have long been in our factories. Now they are on our roads, in our skies, homes and mobile devices in the form of self-driving vehicles, drones and virtual assistants. The robots are already here and the revolution underway. How do those in banking and payments respond?

For millennia, man has been fascinated with machines that take human form or behave in a human-like way. Automata or self-acting machines were known in classical Greece, the Islamic world and medieval Europe. In the public imagination, robots are metal humanoids with small heads, broad shoulders and two legs – even though this is not the easiest or most appropriate form in which to build a machine.

In industry, robots are mainly used for the 4D jobs: those that are dull, dumb, dirty or dangerous. However, robots are increasingly being deployed beyond the factory gates. They care for the elderly, work with children with autism and serve as shop assistants and entertainers. They may not even take a physical form as they exist as software or systems used by millions of people every day. They power spam filters, internet search engines and payment fraud detection.

As voice assistants and chatbots, modern-day robots are expected to behave in a human-like way. As artificial intelligence (AI) and machine learning capabilities, they are expected to locate knowledge, identify patterns, eliminate bias and provide endless capacity better than any human. How do firms implement these new robots? And prepare for a future of potentially radical change?

“To be proper artificial intelligence, there needs to be a learning component.” – Euan Cameron, artificial intelligence leader, PwC.


The new robot revolution is driven by various trends. Firstly, the exponential rise in data. Connectivity is ubiquitous. Sensors are everywhere. Around 2.5 quintillion bytes of data are created a day, such that 90 percent of the world’s data was created in the last two years, says IBM. Secondly, computational power has increased. It is scalable and available instantly and cheaply. The costs of data storage are also falling. This has enabled a third trend: huge advances in analytics, AI and machine learning algorithms.

The race to build bigger, better, more intelligent bots is on. Around $14.9 billion in AI equity funding has been invested across 2,250 deals since 2012, according to research firm CB Insights. There were five rounds of $100 million or more in 2016 alone in sectors ranging from autonomous vehicles and cybersecurity to healthcare and life sciences. Google, Amazon, Apple and other large corporations are investing heavily in AI start-ups to boost their capabilities.

One of the AI growth areas is within language processing, which encompasses virtual assistants, chatbots, text analytics and content creation. Voice and conversation are now emerging as a new platform, operating system or entry point into e-commerce and the online service market. Satya Nadella, CEO, Microsoft is on record as saying “chatbots are the new apps”. Microsoft are betting that voice rather than screen will be become the new way to manage apps. After all, it is easier to talk than type, text or tap.

The future for virtual assistants looks bright. Amazon’s voice assistant Alexa, installed in its Echo smart speaker, could be worth $10 billion in revenue by 2020, according to RBC Capital Markets. The investment bank predicts that 60 million Alexa devices will be sold in 2020, bringing the total installation base
to 128 million.

Voice assistants and smart speakers are the Trojan horses in our homes, cars and phones.They can potentially unlock new revenue for their manufacturers from device sales. Then there is the ecosystem of apps controlled by voice. Developers will also need the supporting infrastructure to create apps, capture data and perform analysis. This only reinforces the frequency and intensity of the relationship between consumers or developers and their providers, and the revenues of Big Tech.


The rise of voice assistants and chatbots pushes on an open door. “People want to self-serve. They want to do it quickly and whenever it suits them. That’s the customer push,” explains Tony Porter, head of global marketing, Eckoh, a provider of multi-channel, integration and secure payment solutions for contact centres. “From the other side, contact centres are trying to automate more to reduce cost as well as meet customer expectations,” he says.

Juniper forecasts that chatbots will be responsible for cost savings of more than $8 billion per annum by 2022, up from $20 million this year. However, the chatbot revolution has stuttered. Facebook launched developer tools to build bots for its Messenger service in April 2016. It has since admitted that its bots have a failure rate of 70 percent. Therefore only 30 percent of requests can be completed without some human intervention. Why?

It is partly because not all chatbots are created equal, according to Frank Carlin, creative services manager at Inbenta, a firm that builds enterprise chatbots. “You can go from building your own chatbot in a couple of hours, which will answer a couple of questions and not much else, to an enterprise-grade chatbot that can help manage your customer queue,” he explains.

Chatbots are getting smarter. They can understand the context of requests through natural language processing, sentiment analysis and image recognition. Carlin gives the example of ticket re-seller client, whose bot receives the question ‘Can I bring my daughter to a heavy metal concert?’

If the chatbot uses keyword matching, it would return results based on what concertgoers can bring into venues – usually not food and drink. A bot using natural language processing would analyse the whole sentence and extract intent. This is whether the enquirer can bring their daughter to a concert. Can they bring a child? What are the age restrictions for a concert? This would inform the response.

While there has been a lot of technological progress over the last three-to-five years, firms often underestimate the work involved in producing a fully functional bot. “Bots are another customer service channel. You wouldn’t set up a call centre in a day and just leave it. They do need to be managed,”
reasons Carlin.

Porter agrees and advocates thinking like a customer when introducing a chatbot. “You need to focus on the questions that customers are asking. Or what information you can gather from a customer to enable the agent to speed up the answer process,” he says. “If you’re going to start replacing some of the mundane and repetitive tasks, start small with the bottom ten percent. Build something that works, test it and then go up to the next level of conversation.”

Voice as the main use of a smartphone is coming full circle. But rather than speak to other humans, users will increasingly speak to machines.

The move from apps to bots is as much a disruption in customer service as it is in technology. Porter recommends treating and training bots as agents, but also re-investing in staff. “If you have super-agents, ensure you give them the skills that customers will come to expect if they speak to a live agent.”

There has definitely been a move towards multiple channels. Customers want to serve themselves by chat, text, phone, e-mail and social media. One of the challenges for contact centres is how they join all the channels up and provide a seamless service. This is the common multi- versus omni-channel debate, which applies to store-based retailers and banks with branch networks alike. How do they provide a seamless customer experience across various touchpoints?

There is no single right answer. Each firm must find the best fit with their own strategy and objectives. They will also need to re-visit their supporting internal processes, particularly around how humans and bots work together. Humans may be able to hand off more tasks to bots. And bots hand off questions to humans with a shortlist of suggested answers. Enterprise bots can also be used to help contact centre staff navigate their internal knowledge resources better.


When almost every second vendor seems to be attaching the words ‘AI’ and ‘machine learning’ to their products like a sticker, it is a case of buyer beware. Many claims around AI are definitely artificial. AI is used to mean automation or ‘doing things with data’. Sometimes even the automation is artificial. A human or manual process is masquerading behind the computer façade.

“To be proper AI, there needs to be a learning component. The system needs to be able to update itself in response to things it has done and outcomes it has observed,” explains Euan Cameron, UK artificial intelligence leader, PwC. “Our definition is machines that sense, think, act and learn in a feedback loop,” he says.

There is no unified theory of AI, so an element of trial and error is necessary in working out what is best for a particular use case. “If AI and machine learning helps you to do a particular part of the value chain faster, cheaper, better, more accurately or with higher quality, absolutely you should investigate whether there is a use case,” says Cameron.

There is a huge opportunity for AI and AI-like techniques in financial services. As an industry with a lot of data, a small improvement in how data is handled and processed can yield large improvements in insight and profitability. “We’re seeing a lot of use cases in the risk prediction area of AI – being able to hyper-segment your customer base so that you have a more granular view of the risk profile,” says Cameron.

There are also a lot of use cases in ‘transparent AI’. These are situations where the black box is forming a conclusion, but this needs to be interpreted to make it accessible to end users. This will become particularly important in a post-GDPR world. Using classifier models to explain AI decisions, make them transparent and communicate them in natural language is a real growth area, according to Cameron.

How do firms implement AI? The response is similar to any other type of technology. It begins with the fundamentals: they must be able to understand and articulate a strategy and business case. Then, how a problem unbundles into different business components. And where technology can be used to solve parts of that question.

“There are any number of vendors out there offering machine learning engines. But what clients need are cars that take them from A to B,” thinks Cameron. If an engine lands on the doorstep, it is not the whole solution. Firms need a chassis, suspension, tyres, controls, a driver and a destination. Being able to assemble the package to answer a business need is critical.

Banks and those in the payment industry need to understand the specific use cases that apply to their own value chain. “There are some common ones in terms of customer interaction, risk prediction, asset optimisation, demand prediction and pricing. But the way that those manifest themselves in specific industries can be very different,” says Cameron.

Firms also must consider the type of data they have and need to power any AI solution. Have they got the right type of data? Is it clean enough and in the right format? Then there is the domain time to implement and train the system. They must also factor in the cultural change necessary within their organisation.

Importantly for a learning system, firms must consider how they operate and monitor the system. Ensuring compliance and that the system is not drifting over time is important, as is improving it. “A learning system doesn’t always behave tomorrow the way it behaves today. That is a barrier to adoption in some cases,” cautions Cameron.


Robots are already here. They have been crunching numbers in the back-office for some time for business intelligence applications, predictive maintenance, fraud detection and forecasting. With the advances in language processing, robots are becoming increasingly customer-facing in the form of applications such as virtual assistants, chatbots and robo advisors. The robot revolution is already underway. There is no going back.

It is no longer a case of man versus machine, but man with machine. This relationship is changing and will continue to change over time. The development of rules-based systems is an example. Rules-based systems have been part of machine learning for years. They are usually programmed by humans. However, with more advanced machine learning systems, the machines devise rules themselves based on the patterns they observe. This has two main advantages.

Firstly, humans are not as good at identifying patterns as machines, particularly in large data sets. Secondly, human patterns tend to be backward-looking and may miss future trends and attack patterns. Previously labour-intensive, upstream, human processes behind machine learning can now be automated, including data hygiene, feature extraction and model updates. Man and machines are still working together to create models and outcomes that are greater than the sum of their parts. The specific tasks are merely distributed differently.

Humans have compassion and morals. They can abstract, dream and generalise, and have common sense. They can see through spurious correlations in data quicker than machines. Machines and technologies are designed, implemented and used by humans. They bear the imprints of their creators, their purpose and their time. Sometimes humans build machines to be more like themselves – more human. Sometimes they build machines to be better or different – super-human. This draws on characteristics that are exclusively
human – for now.

The robot future could be more radical than we can imagine today. Businesses may sell to customers’ AI as opposed to customers themselves. Machine-to-machine transactions could happen without human involvement. Data and decisions that flow from these transactions could create and transform
whole industries.

Every time that there is an exchange of data, there would be an exchange of value. This may in turn involve an exchange of monetary value (i.e. a payment) each time. Firms would need to re-engineer business models, supporting processes and customer journeys accordingly.

The wisdom of American scientist and futurist, Roy Amara, has never been more apt. We tend to overestimate the effect of a technology in the short-run and underestimate the effect in the long-run, he said. However, technologies do not triumph simply because they are the best. They seldom appear in isolation and need the support of other technologies. Socio-economic and cultural factors also influence take-up and the ripple effect of further innovation.

Similarly, no technology – virtual assistants, chatbots, AI – is the silver bullet. Firms must have sound fundaments in place around their sales, marketing and customer services processes, among others, to get the most out of any technology. Technology is not an excuse for broken business processes
and models.

When it comes to the robot future, every company is a technology company and faces the same choice: invest now to become a leader, or pay later to be a follower.

About Author

Leave A Reply