AI in Customer Service
There are two types of agent: the Martini-drinking womaniser in Her Majesty’s Secret Service and the sleuth who tracks down information at the call centre. The multiplying data sources of the digital era are confounding this mission. Swisscom Customer Care’s secret weapon is artificial intelligence (AI).
Text: Bernhard Imboden, images: Jana Wicky,
«Have a relaxing weekend, goodbye!»
Manuel Tschanz ends the call he was making in the softly lit Customer Care Center in Ittigen (Bern). The 22-year call centre agent has been assigned to the day shift. Of course, talking about an «agent» makes James Bond come to mind.
«A customer server has frozen» Manuel explains and takes a sip from a bottle of water. After all, he’s not the kind of agent who drinks Martinis. «That’s really typical for this type of server. A soft reset should get the system back up and running. Problem solved. But finding out this sort of information hasn't always been this easy.»
Problems that aren’t everyday occurrences are inherently complex. And while Manuel goes searching for a solution in all the different systems and databases, the customer is edging closer to a nervous breakdown. As that other agent, 007, is only too aware: fast and fail-safe facts are decisive.
In particular during the graveyard shift. When there’s a skeleton workforce, the transfer of know-how between colleagues comes to a halt. And there's no Miss Moneypenny on duty to help.
Manuel and the other call centre agents started using a secret weapon a few weeks ago. It’s called AI for Customer Support. With little further ado, a new tab, «Easy Solution», was added to the familiar call centre solution. This tab features a search function that makes full use of artificial intelligence. It is linked to all relevant sources of data. The customer only has to start describing an issue and the system will start searching for the right solution using its specially developed algorithms.
All the agent has to do is enter keywords into the search mask. Manuel likes the fact that the tool gives him certainty when he’s giving advice – along with self-assurance. «I don't need to make as many inputs as I used to, and it gives me faster returns.»
Assigned to serve the customer: Manuel Tschanz, call centre agent for Swisscom
David Rüfenacht, AI Product Owner for Customer Service, is preparing our meeting with Manuel Tschanz. He is telling us about the challenges being faced in a cell centre.
«The product portfolio is growing larger and larger, the communication channels are multiplying, the IT environment is complex and there is continued cost pressure. This all makes the work our agents do more difficult. The ones who ultimately suffer are the customers, who spend more time in call queues.
Management became aware of the solutions using artificial intelligence available on the market. Implementing a solution like this to improve efficiency seemed the obvious thing to do. As a leading ICT provider, Swisscom places value on utilising its own expertise and develop a solution of its own.
AI for Customer Support searches and indexes open and closed tickets, existing infrastructure, orders, document and
This gives the agent a 360° view of the customer's history.
Improved efficiency: David Rüfenacht, AI Product Owner for Customer Service
Developing a solution like this can only succeed in an interdisciplinary and agile team. This solution needed iterative development and involved intense teamwork between agents, team leaders and supervisors in order to make ongoing optimisations.
Everyone knew exactly what they wanted to achieve, but no-one knew exactly what the solution would end up looking like. It was only after an intense analysis of all the available data that the building blocks for AI were defined. Software developers weren't the only ones involved – experienced data scientists were called in to develop the logic and to programme the algorithms.
New features were released for testing over two-week periods, and were improved on straight away. «This ensured the solution continued growing. It only took ten weeks for us to implement all the functionalities» as Rüfenacht says with pride.
A system like this only brings benefits, however, when it can master the demands made of it. This means it needs to be capable of learning.
Each time a problem is solved, the agent rates how constructive and relevant the proposed solution is. Systematic evaluations like this allow the system to learn. The best, and therefore most plausible, solution always appears at the top of the list of results, saving time.
And all the intelligence available to our agents suggests that customers will start profiting from artificial intelligence even more directly in future. Because by then they won't have to call the call centre at all any more.
Back to Manuel Tschanz. In order to find information about the problem described to him beforehand, he simply enters the name of the customer. He soon finds the server in the 360° view, but without additional details such as the type or its serial number. It only takes him a few clicks to locate the solution that appears on his monitor.
When asked whether he might one day lose his job to artificial intelligence, Manuel waves any concerns aside. «They’ll still need someone to tell the machine what to search for.»
James Bond would have put it a little differently: «I’ve got something the solution doesn’t have. I’m licensed to give competent advice.»
More on the topic