When I tested the chatlines of Swiss financial services providers two years ago, I came to a pretty sad conclusion. The chat support was generally basic, often tedious, and in some situations even a complete waste of time.
But there have been some developments in the field of text-based assistants since then. Artificial intelligence is being talked about all over the place, chatbots have become a source of entertainment, and companies like OpenAI are promising us the moon with regards to the capabilities of GPT language models.
In light of this hype, I wanted to know how Swiss banks’ humble chatlines might have changed over the past two years. Has GPT been integrated into existing models? And is this really the revolution we have been promised?
Frankly and Yapeal have pulled the plug
Interestingly, two of the chatlines I tested in 2022 were no longer available when I repeated the test in mid-November this year. Frankly and Yapeal no longer offer this service. ZKB’s Frankly responded to a moneyland.ch inquiry by saying that it made the decision to terminate the chatbot when relaunching its website. They would not share information about how heavily the chatline had been used. Yapeal did not respond to the moneyland.ch inquiry at all.
The disappearance of these chatlines did not likely come as a major loss for customers. Although Frankly received a “good” rating in my last test (Yapeal received a “satisfactory” rating), both of these bots only let users choose from a limited selection of predefined questions. That can severely limit the usefulness of the service for individual use cases.
Only one newcomer
As if to compensate for the lost bots, the Migros Bank has launched a new chatbot-based customer service solution. Migros Bank told moneyland.ch that the chatbot is based on GTP 4 Turbo. That means the orange giant has hopped onto the GPT bandwagon.
I am surprised that there are not more service providers offering this kind of chat service. One would think that that AI hype would drive wider adoption of chatbots for customer service. But that does not seem to be a priority for financial services providers at this time. Instead, many banks, including the likes of UBS and the Basler Kantonalbank, let their customers arrange a chat or call with a customer support employee. Some chatbots give you the option of communicating with a real person when, for example, the complexity of a topic is above what the bot can handle.
How do chatbots perform in specific situations?
I performed a spot check to see how good a chatbot is at answering very specific questions. For example, I asked chatbots to tell me the maximum amount of interest I could earn, or how much money I could still contribute to the pillar 3a. Additionally, I also did a separate test in which I typed my questions in Swiss German in order to see how chatbots coped.
The tests were done as a non-customer. The resulting ratings are based on how useful the interaction was from the perspective of a potential new customer – and on whether the information I got from chatbots is actually correct.
Different kinds of chat-based customer support
Chat-based customer services can be divided into three different categories:
- Chatbot: Users enter their questions into a text field and receive answers from a bot.
- Multiple-choice: Users can select predefined questions from a list to view the corresponding answer.
- Human: Users type their questions into a text field and receive answers from an actual person.
The results of my financial service chatline test
Here I have written concise evaluations of the chatlines I tested, starting with the best:
Category: Human
The chat service from Viac combines the best and worst aspects of communicating with a human. The good: The best thing is that I am actually understood. Unlike AI, a human can pick up the nuances in my questions and can work to answer even the most obscure questions. Viac told moneyland.ch that its chat service is its most widely used point of contact for customers. All the more reason to keep this service run by actual humans.
This example shows how a customer service employee can actually understand my situation: When I wrote that I would like to open an account, I was not only told how to open an account, but was also informed about the limitations that would apply when I withdrew the money. Apparently, my human chat partner had sensed that my “interested customer persona” was not too financially savvy because I had also asked whether there is any benefit to using the pillar 3a, and they wanted to make sure that I did not make an uninformed decision. That is customer service.
The bad: To err is human. The first answer also contained the first typo. I wanted to know how much interest I could earn. The answer was that the current interest rate for the pillar 3a was 0.%. But before I could express my disappointment with the poor interest rate, the correct interest rate with the missing decimals popped up on my screen. A bit embarrassing maybe, but not really a problem.
What was not ideal was that in the course of the chat I was shown outdated information in screenshots and links. The exact case was that I was shown the 2022 limit for pillar 3a contributions. Although they did write in parentheses that the limit shown applied to 2022, one cannot really expect the customer to question what that means. If someone tells me that a figure is from 2022, I assume that there have not been any changes since then. When asked about this, Viac explained that the reason for the mistake was that the linked article is a blog post that is not regularly updated and that Information provided on other parts of their website is up to date.
Category: Chatbot
Before revealing the results of my test, it is important to say that Swissquote’s chatline does give you the option of asking for human customer support. I tested that option separately two years ago and came to the conclusion that it is one of the best chatlines there is. This year, I tried to get customer support from the chatbot only.
From the beginning, the experience just did not feel right. The chatbot seemed cluttered. When I opened the chat, I got a list of commonly asked questions plus prompts to other language versions of the bot, and a newsflash that it is now possible to trade pre-market. That is a lot of information – and I could have definitely done without the completely unrelated newsflash, at least.
The interaction did not feel like a chat. The bot just spat out links to information on the topics that I mentioned. My hopes of actually getting concrete answers from the chatline were not fulfilled. Swissquote responded to an inquiry by saying that they plan to migrate to a GPT bot in the near future. It is possible that the bot will soon be able to engage with the customer more effectively.
The bot also does not understand Swiss German. When I said I wanted to open an account, it replied with information about account fees. But because the bot only provides general information about topics you inquire about, it was able to more or less function for questions in Swiss German as well. When I included the term 3a in one of my questions, the bot answered with a link to information about pillar 3a accounts – regardless of the specifics of said question.
- Postfinance: Satisfactory
Category: Multiple-choice
In my last test of Postfinance’s chatline, I was blown away by the efficiency with which the multiple-choice options guided me to my end goal. This year, however, I was somewhat disappointed. Seemingly, the new increased focus on concrete questions is simply more than the bot can handle. The first problem I encountered was that one of my questions was too long – the maximum length allowed is 130 characters. That is hardly enough space for a complicated question.
In general, concrete questions resulted in a curious exchange. It took several clicks before the bot finally revealed the limit for pillar 3a contributions. Some of my questions were not understood at all. The bot’s Swiss German seems to have gotten worse, with it interpreting “Konto uftue” (open an account) as a request for information about how to terminate an account.
This test revealed the limitations of spot checks. On the technical side, nothing has changed since my last test. In response to an inquiry, Postfinance said that the current bot, which is based on its own language recognition model, has been in use since 2017. That means I chatted with the same bot in my last test, but I had a much better experience with it that time.
The bank says that it is working on a new version. Additionally, complicated questions that cannot be answered by the bot are handled by employees.
Category: Multiple-choice
Two years ago, Yuh told me that they would launch a new chatbot with additional features in 2023. But in 2024, the chatline still feels almost exactly the same. You have to click through menus that (usually) lead to general information on the topic you are interested in. I am not sure which new features were meant to have been added. Yuh responded to an inquiry by saying that there is one chatbot in the app and another on their website. Possibly the upgrade only applied to the first.
Another thing that is not new, but is worth a positive mention, is that the Yuh chatbot can help you in certain emergencies, such as if you suddenly lose access to your account. Having a collection of information for situations in which you need help fast is useful, from a customer perspective.
- Migros Bank: Unsatisfactory
Category: Chatbot
Of all the chatlines I tested, this one felt the most like ChatGPT. It replied to questions with complete confidence in well-formulated sentences. But in spite of the polished façade, one could not shake off the suspicion that the answers given may well be utter nonsense.
And sure enough, the bot responded to my question about how to earn the most interest by telling me about an investment account called MB+. It even gave me links to documents that were meant to provide more information. The only problem: The MB+ account does not currently exist, and there was no mention of it on any of the linked documents. According to the Migros bank, the bot’s database was outdated. After moneyland.ch had made it aware of the error, the mistake was corrected. But obviously it is not the job of the customer to check whether the information they are fed by a bot is actually correct.
What is more, certain questions clearly confused the chatbot. It seemed to be unaware that Migros offered solutions for contributing to the pillar 3a. Swiss German was also a challenge for the bot, but I was able to eventually get an answer after being asked multiple times to explain what my question meant. The answers, when correct, were very helpful and detailed. But that is a weak tradeoff for not being able to rely on receiving correct information. For that reason, this chatline does not qualify for a better rating.
Category: Human
I'm at the end of my tether with Valiant. The bank was already extremely slow to respond in the past. This time too, I had to wait more than five minutes before I received the first answer. When I did, all that it said was “it depends.”
After that, the waiting times got even longer. Eventually I just broke off my test. If I were a potential new customer and I had to wait more than a couple of minutes for an answer in what is supposed to be a live chatline, I would probably just stop waiting. In that case, the quality of the answers themselves would be irrelevant.
When asked about the long waiting times, Valiant said that it is possible that the customer service line was dealing with a large number of support requests at the time of the spot check. That is certainly possible, but it is unlikely to make a difference for potential customers. Additionally, this is already the second spot check in which the chatline performed poorly. Valiant Bank also said that the chatline may not appear on its website at all during times when there are numerous support requests to deal with. They could consider implementing that more rigorously.
My verdict: Have chatlines gotten better?
My latest test did not leave me with the impression that chatlines had gotten better. In the best case, the interactions were the same as when I did my last test, and in the worst case they were less satisfactory. It is also possible that my expectations have gone up now that chatbots and artificial intelligence have acquired a certain omnipresence partially fraught with exaggerations.
Whether or not you want to chat with a bot at all is a matter of personal preference. Opinions about the usefulness of applications like ChatGPT are massively divided. Some people are impressed by how convincing AI is, and how smooth and fast the interactions occur. Other people are annoyed by the unreliability of the answers provided, and the questionable data harvesting behind these applications. The test does show that customer support chatlines are not completely useless. But it also reveals that even today, technology has a long way to go. Human intelligence is more likely to get you the answers you need. But, as the test of Valiant’s chatline shows, you have to be able to reach a human assistant before they can help you. Bots are generally more readily available, including at night, for example.
On the whole, my opinion is that bots will need substantial improvements before they earn their right to be used for customer service – especially as an alternative to human specialists. And because it seems that such a development is a ways away, the decision which some service providers have taken to terminate their chatbot services may just be a step in the right direction.
More on this topic:
Swiss financial services provider chatline test 2022
Compare Swiss savings accounts now
Swiss pillar 3a robo advisors compared
Deal of the Day 
