Only what the customer really wants
Interview with Jan Morgenthal
Jan Morgenthal is the Chief Product Owner for the eLIZA project. As the leading product supervisor, he coordinates the project with employees and stakeholders. Whether features are useful, relevant or superfluous - he assesses what the customer wants and tests prototypes to determine market-readiness.
What are the differences between artificial intelligence, machine learning, deep learning and other such terms?
Artificial intelligence is an umbrella-term used to describe the very complicated circumstances that occur at the human-machine interface. It encapsulates a range of technologies, such as image recognition, natural language comprehension and text recognition, as well as issues like machine learning and machine intelligence.
Given that these are themselves governed by a variety of scientific and academic disciplines – linguistics, art, psychology, among others – it becomes clear that the term AI can only ever be considered a top-level collective term. Ultimately, AI is about trying to replicate human intelligence within a machine. It takes cognitive competences, such as vision, speech, smell, hearing, and then uses artificial intelligence to interpret or even recreate them. It requires an understanding about how the brain works to then attempt to translate these purely chemical processes into mathematical algorithms.
The other terms describe sub-categories within artificial intelligence. Machine learning means that a machine can identify and remember patterns within a large dataset. In the context of Deutsche Telekom, this would be identifying identical customer enquiries out of a set of hundreds of thousands. Such capabilities make it faster and easier to categorise and catalogue enquiries, thereby assisting agents in quickly determining the customer’s issue – and the machine in logging it faster. In principle, machine learning might be considered a precursor to artificial intelligence.
Conversely, deep learning takes a very different approach. It works on the principle of neuron networks. With machine learning, the computer’s state is ones and zeroes. In the case of deep learning, there are several levels because networks can be built on top of each other. This means makes it possible to display not just the states of zero or one, but to harness a network to create a hierarchy of multiple statuses.
What general milestones have been achieved within artificial intelligence?
What can we realistically expect in the future?
Essentially, there have been four major phases within artificial intelligence. The first major “Sturm and Drang” phase began in the 1950s and lasted until the early to mid-1960s. This period brought us break-thoughs such as the eLIZA software programme, which was developed by Joseph Weizenbaum in the USA and generated first significant excitement. It was predicted that computers would be able to completely understand human language within only four or five years. We now know that this didn’t turn out to be the case. The speed of development within Artificial intelligence has defied the optimistic predictions of all early researchers within the field.
The 1960s were something of an ice age for AI. Many companies that were involved during its infancy turned their backs on the subject – as did many scientists. It was not until the mid-1980s – and up to the mid-2000s – that things began warming up again. Once more, there were new developments within the field; however, everyone was still very cautious and there was no real hype. It is also fair to say that there were no major leaps forward during this period either. Instead, it was the case that existing projects were brought back to life and slowly developed further.
The first decade of the new century saw companies helping to bring prominence to the issue of artificial intelligence. These included IBM, but also facilities such as the “Deutsches Forschungsinstitut für künstliche Intelligenz” (German Research Institute for Artificial Intelligence) in Saarbrucken. The second “Sturm and Drang” phase occurred between 2008 and 2011; prompting universities to create an incredible number of professorships within the field.
The automotive industry has made major advances since then, with parking pilot and parking assistance technologies – both of which are artificial intelligence. Finally, Siri and Alexa came up with fresh approaches to working with natural language. Microsoft and Facebook both announced their bot frameworks last year, and since then, I think the hype has increased by another 200 percent.
What Artificial Intelligence milestones have been achieved at Deutsche Telekom?
What can we still expect?
Deutsche Telekom started working on this issue in mid-2015, during a relatively small workshop together with T-Mobile Austria. We began without really realising that we would end up in the field of artificial intelligence. Over the course of the first 13 months, we tentatively worked through a wide range of workshops, also to determine which alternative methods were available and what the innovation target should actually be. Our focus from the very outset was how the interface between Deutsche Telekom and our end-customers could be simplified to such an extent that the existing digital contact points could really become easier.
We hit upon the subject of artificial intelligence in autumn 2015, and it was then that we consciously decided to approach the issue of digital assistance. It is very much a current topic, but also one that is beneficial to Deutsche Telekom. We launched last year with our Austrian subsidiary, and we have already seen successes. Deutsche Telekom in Germany officially became part of our programme at the beginning of this year. Our milestones for 2017 have already been defined: our aim is to apply artificial intelligence within customer service and to trial it within private customer sales.
How quickly has AI grown within Deutsche Telekom, and how quickly do you think it will expand?
It is truly remarkable. The dynamism and growth that we are witnessing are enormous. As I mentioned previously, we started out in 2015 and grew relatively quickly to ten employees. Throughout the course of 2015, this figure roughly doubled to almost 20. In the meantime, our current forecast is that we will hire around 50 to 60 employees in 2017. This represents a rather steep curve, and it is a curve that will advance yet further because we will also add employees for other countries. As a result, the job will become significantly more international.
What are the benefits for the customer?
Although they say that digitalisation is supposed to make everything easier, many digital interfaces – such as websites or apps – are far too complex for many customers. They can’t find their way around. Artificial intelligence can help, such as by providing a natural language interface. This allows customers to access these contact points using a text message or their own voice. The result is they can conduct their business much faster and more easily, and present their requirements – whether they are concluding a contract or simply changing a surname or address.
Transaction costs are significantly reduced both for the customer and for us: a significant benefit for both parties. The two essential points from a customer perspective are user experience and time-savings. Naturally, none of this costs them a penny more because it is not a premium service we are offering; rather it is a regular service.
What visible products are you expecting for customers?
Beginning this summer, we are planning to provide a chatbot on the website, and a few weeks later also within the Deutsche Telekom service app. The bot will use text input to offer various services relating to service disruptions or invoices. It will be able to help and advise customers and will also be able to initiate actions entirely autonomously. Later, we will implement a so-called voice claim; allowing customers to not only input their issues in text form, but also to dictate them using natural speech.
The next thing we have planned is an interactive voice portal. This will mean that when the customer calls our hotline, for example to report a disruption, they no longer have to wait until they are connected to an agent. Even though we have the largest customer service organisation in central Europe, things can still currently take a bit longer if customers call during busy periods. In future, customers will be able to talk directly to a digital assistant and have their specific needs processed without stress or cost to their time.
We will also use AI to enhance our internal customer service. This means we will help agents to work better using artificial intelligence. Many of our customers are not keen to talk to a chatbot or a callbot, and would prefer to talk to a real person. In such cases, we will support our own employees using AI. The agent will be able to input a customer’s complete question or just a keyword into a database, and then receive suggestions for resolving it. They are also connected automatically to the customer’s data; helping them to provide the best possible personal and individual advice over the telephone.
Our future plan is to have AI monitor calls and give an answer or propose a solution. The programme for 2018 is still quite broad. We are in the process of consolidation because it is just incredible how many possibilities there are for AI to support us. We already have a numerous ideas and many prototypes.
How do you find out exactly what the customer wants?
We have an entire team dedicated to customer research. There are some surprising moments, when we are convinced that a certain idea is perfect and a customer survey comes to a completely different conclusion. The customer simply says, “No way, I don’t want anything to do with it.” There are also some features where our mindset was simply too conservative and customers had totally different, much more visionary ideas. An idea is only worth something if it really actually works and the customer likes it too.
If I want to work in AI, why should I take a job with Deutsche Telekom?
We offer a really cool package of things. Very few companies in Germany truly work so closely with the customer in matters relating to general innovation – and to artificial intelligence in particular. Furthermore, we are an incredibly young and dynamic team. We have succeeded in creating a start-up within Deutsche Telekom. Of course we have some interfaces with the rest of the company, but we are relatively free in developing our own algorithms, products and new cases, and our work is dynamic and agile – just like in a start-up. This is a genuine USP.