Artificial intelligence moved from being the stuff of movies to become part of our daily lives long ago. Many of us use applications like Siri, Alexa or Google Translate. AI is improving voice control and giving technology a more human face. We highlight five examples of how Swisscom uses artificial intelligence.
Ladina Camenisch & Sascha Bianchi & Bruno Böhlen, 12 October 2017
The data deluge is mounting incessantly: more than 500,000 GB of data alone is transmitted on Swisscom’s mobile network. Computers, smartphones and sensors generate huge volumes of data. It is already virtually impossible to analyse and use this immense mountain of data but artificial intelligence or AI can make it easier.
Mostafa Ajallooeian, Swisscom Innovations explains in the video how artificial intelligence can give old blurred videos a new sparkle. Video total length 1:00 min.
The term artificial intelligence is often used in the same breath as Big Data. But what is the difference between the two? Big Data and artificial intelligence both use huge volumes of data to produce results. The difference lies in the anticipation of intelligent behaviour: Big Data analyses data and presents the desired result at the end. Artificial intelligence also analyses data. At the same time, however, it learns and identifies correlations that we didn’t even realise existed before. And even more importantly, AI proposes a solution at the end. We spotlight five examples of where and how Swisscom relies on artificial intelligence.
Customer Service employees have a hard job. Every incoming request increases the volume of information – and it is also stored in all different places. Consequently, employees are stressed and customers unhappy if they have to wait too long to have a matter resolved. Solutions are already available.
“I considerably reduced my average handling time by using the new AI solution.”
Manuel Tschanz, Customer Advisor
Cosmos is used for e-mail or contact form requests from residential and SME customers. Cosmos recognises customers’ requests. In the case of simple routine requests the system advises customers that they can settle the matter online in the Customer Center – and sends them the relevant links and help pages. So customers are no longer restricted to hotline or Shop business hours. Cosmos forwards more complex requests to the relevant department. Thanks to artificial intelligence Cosmos constantly enhances its knowledge. If a question is incorrectly routed, it simply needs to be sent to the right department to train up the system.
Marmo is deployed in corporate business and uses artificial intelligence to automatically seek solutions across all sources in the existing infrastructure. The call agent enters a couple of keywords and the system automatically checks for similar requests in the past and offers a possible solution in a matter of seconds. The call agent then decides whether the proposed solution is useful. The system learns and gets wiser as it goes along. Customer issues can thus be dealt with better, faster and more individually.
The chatbot, which Swisscom aims to introduce at the end of 2018 as an alternative to traditional service channels like the Shop and hotline, will work along similar lines. The Service Bot offers help at the press of a button. It deals with administrative customer requests on PIN and PUK codes and technical issues. It can provide simple product advice. As part of an app the bot is designed for people who prefer to do things online – at any time of the day or night. Like Cosmos and Marmo, the Service Bot also builds up knowledge with each request thanks to artificial intelligence. If the bot can't help, it forwards the request to a call agent, so far from being competition, it is more like an electronic colleague which supports the team. The Service Bot will be trialled in-house in the upcoming weeks and with selected customers at the beginning of 2018.
Swisscom TV 2.0 subscribers can use the search function to find their favourite series. They used to have to enter a search term on the remote control – but it’s been a lot easier since 2016. You just need to say the name of the programme into the microphone and the new UHD TV-Box searches for films, programmes or football matches. And what’s more, the remote control also understands Swiss German. Swisscom plans to use more voice detection in the future. It’s an option for call centres: the time-consuming “and now press button two” scenario might soon be a thing of the past. Instead, customers will just need to ask a question in their own dialect to be immediately understood and connected to the right person in the company.
“A major challenge for Swisscom in voice search development was to train algorithms in our national languages and dialects.”
Dr. Felix von Reischach
Head of Product Management Artificial Intelligence, Swisscom
But how does a machine learn Swiss German? Experts feed the computer with standard and Swiss German to show when the two are talking about the same things. Hours of film material and countless books were scanned in for the purpose, using sources available in both standard and Swiss German. The more data accumulated in different dialects, the better artificial intelligence works. Dialect words are thus added to the language pool for new AI solutions.
First level customer advisors receive hundreds of e-mails a day. They answer some directly but many have to be sent to experts. The latter are sent all over the company. E-mails are frequently sent on an odyssey through the company before they finally reach the right person. As part of a pilot project Swisscom developed an algorithm that eliminates manual e-mail classification for a leading Swiss bank. The system reads the information in the e-mail and automatically forwards it to the right recipient. However, the solution is capable of far more: it anonymises confidential data, recognises the language and identifies specific key words or customers.
Before the artificial intelligence training, the customer’s old data was first evaluated and the algorithm trained with basic functions. The bank's employees then monitored results for four weeks and rated them with a click. Was the e-mail forwarded to the right department? Employee feedback was directly incorporated into further development of the algorithm so the system kept learning all the time. The algorithm now has a higher hit quota than a customer advisor.
At Swisscom various teams are working on the development and integration of artificial intelligences solutions. The current focus is on increasing the efficiency of existing processes, developing network security and voice detection in Swiss German. With these aims, Swisscom both develops its own algorithms and uses existing publicly accessible algorithms and combines them with its own data sets. It is supported by EPFL in Lausanne and other universities.