Chatbots are gaining some momentum in the financial services industry in Finland. OP Financial Group has been at the forefront of launching a relatively large amount of experiments. Other Finnish banks and financial institutions have been rather passive but at the moment OP Financial Group is setting the bar quite high.
Kotipizza’s collaboration with OP Financial Group is an example of a transactional chatbot for ordering food (simply order through Facebook Messenger and pay with Pivo app), last year OP’s Pivo Penni (a chatbot for students) turned out to be a short-lived test, and in addition OP launched an invite-only financial management chatbot Pivo Alfred (spending analysis) last December. Unfortunately, there has been no further news about S-Bank’s collaborative chatbot project since last year. In the US, many banks and other financial institutions have already deployed chatbots. As many commentators have already pointed out previously, there are various forms of chatbots ranging from customer service, personal financial assistance to content.
On the other hand, an Estonian bank, LHV Bank, launched a stateless customer service chatbot based on Facebook Messager just a few months ago. Although LHV’s customer service chatbot is a passively informational, stateless and straightforward optimizer constructed on a very mere tree-like logic, general everyday information can be found out (of course, the question arises if this information is easier to retrieve just by visiting their website). Although customer’s jobs cannot be done solely relying on the chatbot (i.e. you are instructed to call customer service, follow links to LHV’s website, and the relatively generic information is provided), LHV’s bot probably serves its purpose for now. LHV’s bot is a type of more interactive FAQ, nothing more, nothing less.
Accenture’s global study of nearly 33 000 financial services clients across 18 markets shows that even today 71% would use entirely computer-generated support for banking, 74% would use entirely computer-generated support for purchasing insurance, and surprisingly nearly 80% would use entirely computer-generated support for investments.
It’s evident that there are multiple interesting use cases for chatbots in everyday retail banking, and both financial service providers and their clients are curious about new technologies. There promising new startups like Avaamo, Kasisto, Kore.ai, and Teller transforming the “conversational banking” into reality.  Conversational banking, as Keith Armstrong argues at Financial Brand, is “redefining [the bank-client] relationship again, bringing two-way interactions back to digital banking.” Capgemini’s Damyana Stoyanova is right pointing out that for “highly regulated industries such as financial services the technology is not at the necessary level to deliver the desired results.” In February 2017 TransferWise introduced a bot for Facebook Messenger to carry out international money transfers.
Chatbots can’t fix the root cause
As Matt Hooper from IMImobile explains, “the ecosystem for chatbots is already quite complex, with many options to choose from.” For a financial services company, there are notable differences between the different customer journeys. Although Hooper himself argues that banks need to rely on “solutions that are enterprise grade.” I don’t think that the decision is so simple after all as different chatbots can serve various purposes, e.g. take a look at the example of LHV Bank in Estonia. Not every banking bot has to scale, change quickly, provide detailed reporting, etc. but instead these requirements should be clearly based on customer and organizational needs rather than assuming that every banking bot has to be this or that. Hooper, on the other hand, is quite right when pointing out that, “While we don’t expect large volumes of banking customer interactions to move over to chatbots within the next year or so, it’s important to start dipping your toe in the water.”
In practice, this means that financial service companies should try to figure out appropriate use cases as Hooper puts it. This might be one way to approach the issue, but I don’t think that it actually works in the long run. As I explained in my earlier post, chatbots should not be treated as demand management technique, but rather they should be appropriately deployed to handle dull routines, simple inquiries, requests for information, etc. If the goal is just to manage demand, the potential of chatbots is undervalued as they are treated just as a dam between the clients and the company. Instead, the best consumer experience relies on the principle of minimum interaction, i.e. start from eliminating the demand rather than increasing supply of customer service as Bill Price and David Jaffe explains in their magnificent book. My simple point is that chatbots are not the answer to the root cause driving the demand for customer service, i.e. things that are beneath the surface and deeply anchored in the service delivery.
As long as the service delivery chain is well-established and bots are deployed to drive efficiency and scale, there is no problem with this. If bots are employed just to boost the supply of service without any thought put into the service system itself, more questions will follow. Bots should be treated as a way to make processes fast and straightforward, measuring things that really matter, and matching the service to the customer. Today it’s easy to start building (chat)bots as Mats Kyyrö points out in a recent article but the decision of using bots has to be based on the fundamental rethinking of the whole service system as bots on their own can’t develop a truly exceptional service culture.
Many customer service departments – even those who are already drowning in customer contacts – are built around very old metrics, i.e. the number of contacts each customer service agent must handle. This is self-defeating. Companies must primarily focus on reducing or meeting their current demand for customer service, but not willingly increase it or make their service less efficient (as people are quite impatient). So before coming up with the actual real life use cases of (chat)bots, it makes sense to actually find out the root cause your customers are contacting you today, and establish a closed-loop to challenge demand as Price and Jaffe argue. This is the phase when specific solutions like chatbots come in as one should know by now how to challenge the demand, and decide on next steps tackling with demand (simplify, automate, exploit or eliminate). Bots, as pointed out, are one effective way to challenge the demand for (human) service.
What about wealth management chatbots?
Efi Pylarinou wrote a pretty good article on applying chatbots to wealth management about a year ago. She discusses a couple of interesting case examples (trade execution is the most interesting one) and lays down a case for chatbots in wealth management. Chatbots could provide information about “positions, cash, risk units, industry exposures, correlations to existing holdings etc.” and also, chatbots might be used “in discovery of investment ideas and curating financial information.” If chatbots are fully integrated with trade execution, buy and sell orders can be executed with the help of a chatbot. This is a brilliant idea but, of course, only execution of the idea really matters.
Fundamentally, we view wealth management as a software problem.
– Davyde Wachell, CEO, Responsive AI (The Globe and Mail, 1.5.2017)
I see great possibilities in integrating more advanced loyalty chatbots with social trading (“Facebook/Twitter for investments”) and robo-advisory (automatizing portfolio management). Wouldn’t it be cool if a robo-advisor would deploy a chatbot that proactively would inform you about the things you want to and bring up particular investment ideas based on social sentiment? I think that it is. Of course, there are compliance and regulatory issues to be resolved for every new service, but these are still things that can probably be adequately addressed. Robo-advice and chatbots face a particular set of similar challenge that can be dealt with the help of hybrid-bot approach so that people are not left alone to transact purely with bots in an infinite loop. Currently, chatbots are still highly domain-specific so accomplishing a set of pre-determined goals is (pretty) easy but learning by doing is something that chatbots aren’t really able to do for now.
There are multiple challenges with chatbot adoption in wealth management. These are related to data quality, decision-making and the importance of human factor in the traditional wealth management service itself. Also, chatbots are still relatively simple as they rely on basic input-output communication, and there is no way to have a natural interaction with a chatbot for the time being.
There are numerous relatively well-defined tree-like discussions that a wealth management client might have with a semi-stateless chatbot. In the future, much more difficult scenarios can take place. For example, one example of this kind of more nuanced interaction could be something like this (although I don’t really think that text-only conversational bots are the way to go as there are various problems with this kind of interactions).
Mike the Chatbot: Hi Thomas? How are you doing?
Me: Hi Mike! Could you show my latest transactions?
Mike the Chatbot: For how many days or weeks? Do you want to include all the instruments or just some of them?
Me: Show all transactions from this year.
Mike the Chatbot: Sure. Do you want them in XLS or PDF format?
Me: Give me ’em in both.
Mike the Chatbot: Will do. Give me a moment, and I’ll prepare transactions listing in a moment. Anything else I can help you with?
Me: Update my risk preference.
Mike the Chatbot: Actually, we could have a virtual meeting with John from our Copenhagen office on Friday 3:45 pm. We could check out your current setup as it was done more than 6 months ago, ok?
Me: Yeah, sure. Send me a reminder and ask John to prepare a couple of portfolio strategies based on factor ETFs. Bye!
Mike the Chatbot: I will send an SMS to you one hour before. I will inform John that you’d like him to “prepare a couple of portfolio strategies based on factor ETFs.”
Mike the Chatbot: By the way, the transactions document you asked for has been sent to your personal secure inbox.
Mike the Chatbot: See you soon!
This conversation mentioned above, by the way, is way too complicated for currently available chatbots to handle but I just wanted to demonstrate one possible way forward. At the moment, best (chat)bots I’ve seen are like interactive creatures living on an already existing platform, i.e. they are preferably available 24/7/365 and they don’t need separate setup.
Conversational chatbots, those which rely heavily on text and therefore interacting with them can be somewhat frustrating (see my example above), probably won’t win the game if the backend technology (natural language processing, machine learning etc.) is not fully integrated with some kind of already existing factory-like settings which allow simpler interaction, e.g. preset functionalities which can be chosen from a list.
So will we see wealth management or trade execution chatbots soon? Actually, there are simple trade execution and trading idea-generating chatbots available already. Although I personally find wealth management, financial planning, and stock brokerage as potential areas for adopting chatbots, the adoption will be slower than we expect in the short term and faster than we expect in the long term.
There will be multiple obstacles on the way but I think that it makes sense to actually believe it from the dualist point of view, i.e. what are the most irritating things bothering both the clients and the company at the moment, and is there a possibility to leverage customer experience with the help of chatbots in order for customers to buy more and sell back time. In the end, chatbots are just one piece of the puzzle in realizing service-dominant logic.
It remains to be seen which service provider will put forward the first chatbot for traders and investors in the Nordics.
Errata: 4.7.2017 – Fixed a pronoun typo (thanks Efi!).