Conversational Banking Landscape
Conversational interfaces are gaining momentum both in banking and pretty much every other industry. However the landscape of technology to support can be confusing and complex. The framework pictured here is not a technical architecture, but a functional view of the key components of a “conversational banking framework”.
Conversational interfaces are gaining momentum both in banking and pretty much every other industry. However the landscape of technology to support can be confusing and complex. The framework pictured here is not a technical architecture, but a functional view of the key components of a “conversational banking framework”.
There are many conversational platforms and 3rd party development tools that support some or all of these components. This framework can be used to understand the capabilities of a 3rd party framework or to develop your own.
Conversational Platforms
There are three mainstream types of conversational platforms, text, voice and chat (four if you include visual sign language automation from companies like www.signall.com). Over time we may see these converge i.e. voice providers supporting chat, chat providers supporting voice etc. These interfaces are delivered through screens/apps and increasingly through physical devices like home speakers, in-car systems and even robots! In the future we may also see a rise of AVATARS for example www.soulmachines.com have extremely life like characters.
It’s easy to think that with the rise of smartphones that SMS text will soon super-ceded by chat and voice. However 2019 could see the rebirth of text by Rich Communications Services. Initially this will only be supported on Android by Google and Samsung, but it is rumored that Apple will follow.
The big challenge for developing conversational interfaces is that these platforms are very much like native mobile development, they come with their own proprietary development kits. However 3rd party solutions increasingly offer multi-platform development capability, thus a single development platform for multiple conversational platforms.
Speech Processing
Speech processing comprises of two parts, recognition and output. The key differentiator across platforms is accuracy, and when reviewing providers things to look at include the ability to handle different international languages, ability to filter noise and accents. When it comes to speech output, in the past gaps between words were very distinguishable but nowadays many providers have smooth language output simulating very natural speech output. Differentiators on output would be the ability to choose or personalise the output, there are a number of text to speech providers that allow you to do this. In the future organisations will create “personalities” to match their brand values, or even simulate the voice of their sponsored brand ambassadors.
Natural Language Processing (NLP)
This is the heart of a conversational platform and has two parts: Understanding and Response.
Good response processing allows for some personalisation so that responses can be varied in content, style or tone/accent. This can be quite powerful depending on the platform, for example in some solutions local user dictionaries of language are created for each user, their language is then reflected back in responses. This is a common technique people use to build rapport and relationships.
Natural Language Understanding is a key capability as it is what translates what the user is requesting into a specific INTENT. So for example a user could say “I want to cancel my gas bill” or “I want to cancel a regular payment to my gas company”. NLP will translate that to “Cancel standing order” as an intent, and identify that “Gas Bill” was the beneficiary required field for that intent. Clearly the effectiveness of this governs the real power of the conversational platform. Many 3rd party solutions will embed use of platforms like IBM Watson, Microsoft Luis or Google Natural Language services. Some platforms like Kore, Finn and Clinc have pre-built dictionaries for banking.
Some platforms also have the ability to pass requests to a human agent to process requests rather than simply output a “Sorry I do not understand the request” response, or if the request is particularly complex.
In screen based solutions like chat, responses can include content or interactive screens and thus provide very rich interaction so much so that some talk about chat replacing apps!
Chat History
Some NLP’s are pretty basic platforms that recognise speech and pattern match synonyms, others offer more advanced machine learning capabilities utilising neural networks. Real machine learning solutions help the recognition improve over time, and adapt to wider language requests without further development.
Micro-flows exist to handle a request that requires several parameters, where the NLP has identified the user has not provided them all. So for example if a user says “I want to pay Fred”, the micro flows would handle getting the amount, date and Fred’s bank details if he wasn’t a known beneficiary.
Conversations should be stored, especially in banking to assist with audit and non-repudiation issues. They will also help in reporting and understanding effectiveness of the services provided on the platform.
Digital Engagement
Digital engagement interjects further proactive dialog into a conversation. Driven by data and events/sensors, engagement could be used for a variety of reasons e.g. using personal finance management (PFM) analysis to provide advice, using CRM to make a sales approach, using fraud data to notify of suspicious activity etc… Engagement is driven by access to bank and non-bank data.
Whilst much of the focus for banks will be on self-service banking automation, already banks like DBS have made huge strides in automating more general queries. However non-bank data can add much greater dimensions to conversational banking to truly contextualise conversations not just down to the individual user, but the user, their emotion, their location, etc. The list shown is merely a set of examples, as always with data, the possibilities are almost limitless.
Conclusion
Whilst the landscape for conversational may seem broad and complex, fast in-roads can be made with a reduced scope and focus. I hope this overview is helpful in helping you to plan for the bigger picture and a future that could be truly transformational in terms of human – computer interaction.
I hope to cover more on the topic of conversational banking in the coming weeks, and welcome your feedback and suggestions for future posts.