Speaking Fintech: Integrating AI, NLP, Conversational and Voice
We had an incredible night of 4 talks and 3 panelists at Bots and AI tackling Fintech and much advice that is very relevant to anyone creating a bot or voice experience.
Joti Balani — Building Enterprise grade conversational AI services
Joti started her consulting firm about a year ago because she loves product development, loves understanding psychology. Finance and retail are aggressively going after conversational AI — especially for their customer experience in both B2C as well as B2B.
Not many people have the opportunity to involve psychologists in the process of web or mobile app development and with chatbots and AI for the first time technology is trying to understand the psychology of the users.
Bots are nothing but a new digital channel for selling goods and services and authentically delight the customers in that experience using a natural language interface to do that. The expectations for humans to use bots are very different from mobile and web applications.
Bots can be programmed for IQ, but EQ is a different ball game. Interjecting emotional intelligence into bots is an important aspect of a good user experience.
Hence designing and building them requires a multidisciplinary team. Joti proposed a 7-step framework (Ideate, Plan, Design, Build, Test, Launch, Measure) with a multidisciplinary team (Business Stakeholders, Product Manager, Data Analyst, Technology Architect, Software Developer, QA Engineer).
Joti ended by saying that for many enterprises a good place to start bot deployment is with their employees because if they can take good care of their employees, the employees, in turn, will take good care of their clients.
Chris Butler — Conversational Agents for Finance
Chris’ talk covered the ground for how to think about conversational agents in general and how to make the right decisions about what to build when building one.
In finance, importance is placed in having a single view of the customer — requiring the merging of different available data about the customer into a single interface.
In the process, more often than not Conway’s law kicks in: when you build things your organizational structure comes through in what you build. Which is to say, oftentimes all the organizational dysfunctions end up getting exposed to the people you are trying to serve. Instead, if you focus on having a ”single conversation” with the customer it becomes important to pay attention to things like topic switching, using analogies and language as people use them and providing customers with enough information to help them interpret what your organization does.
Chris framed the building of conversational agents along these key aspects:
- Understanding the common conversation patterns — regular chat (B2C or B2E), whisper agent, group chat, escalation and learning
regular chat — one on one chat with a chatbot
group chat — multiple humans in conversation with each other as well as a chatbot (slack bots are a good example)
whisper agent — when a chatbot assists human agents with standard operating procedures or to answer customer queries without directly interacting with the customer
escalation — diverting the chat to a human agent when the chatbot is unable to assist
learning — chatbot acts as or escalates to a subject matter expert to respond to unfamiliar queries, with a potential feedback loop where those queries could then be handled by the bot in future
2. Understanding how and where the conversational agent fits in the myriad service options offered by the company — typically it would be safe to split responsibilities as below:
⁃ Simple habitual (repeated) tasks >> apps and websites
⁃ Complex SOPs ( done rarely, complex process steps, but standard mostly undeviating process) >> conversational agents
⁃ Complex anomalies >> humans
⁃ Chaotic special cases >> highly specialized humans
3. Interpretability and explainability — making your org chart easy to interpret to customers, and systems that need interpretation need conversational interfaces
4. Gaining and establishing the context in the conversation — using profile and behavior information available about the customer, starting to predict what the customer wants based on this information, picking up from a previous conversation and finally just asking them outright. (Sentiment ranked very low for Chris in the use cases he deals with because conversation interfaces based on standard operating procedures don’t care about emotion.)
Chris walked through a use case selection framework for Amelia (IPsoft) as a means of judging the success of conversational agents. The rubric checks for whether a use case is feasible (in Amelia), usable (provides some value) or has a real ROI (is of substantial value). So for example, answering FAQs using Amelia is feasible, but being able to submit PTO in workday using Amelia (end to end resolution with APIs) provides significant value to the HR functional group.
When thinking about what to build (when building conversational agents) ask the question what if the customer could opt out and you are penalized for it?
And when it comes to automating, how do we keep the meaningful parts of the job available to people and automate just the tedious parts.
Chris concluded by reminding everybody that even when we talk about automation we are still talking about technologies that simply intermediate to humans.
Eric Seay — Audio Branding for Conversational Banking (how small sounds have a huge impact on Voice)
Sonic branding — distilling grand values into audio assets, much like visual brsnding is doing the same with things like graphics and your visual logo.
Eg: the jingles for Nationwide or Liberty Mutual. These audio assets can be small or large, but more specifically in Alexa skills, they can find their way into what we call earcons — a pun on the more familiar term icon.
Looking at design principles of something like this, sounds happen at certain key moments to give users confirmation that certain things are happening. Without them, Alexa skills would be the equivalent of plain HTML text without the visuals.
Brands like Mastercard and Visa have seen a significant increase in brand awareness by including sonic branding in their purchase confirmation. Eric shared that a lot of financial institutions have been approaching Audio UX for building their sonic branding because it gives customers a sense of comfort and assurance of safety when used effectively.
Speaker and Panel Recording on YouTube
