Building a Micro Chatbot Platform
When was the last time you went to a website and clicked through layers of navigation to find a piece of information? How was the experience? Time consuming? Frustrating? These days, we all seem to have a lower tolerance for hunting and pecking our way to an answer. Instead, we'd rather just ask the question to Alexa, Siri or Google. At the very least, we want to open a messaging app or chat window and have a conversation with a bot/person. We're starting to have those same expectations at work, too.
Here at Northwestern Mutual, we're starting to see departments researching how to simplify cumbersome user interfaces, databases and services using chat functionality, and our innovation team aims to unlock that potential. We're building a proof-of-concept Micro Chatbot Platform that enables engineers from across the company to build, test and deploy their own micro bots. For those who have an Amazon Alexa, our micro bot approach is very similar to Alexa installing additional skills that teach her new functionality. We think that by decentralizing the bot creation and functionality through micro bots, engineers will be able to build responses unique to their business areas. It’s called a micro chatbot because the bots are consolidated under one parent bot, so any question can be answered from one chat. Over time, the platform will get smarter as engineering teams contribute to their individual bots.
Unlocking chat also has another potential benefit: New or updated information can be deployed fast. For example, to create a web application or update website content, you'd have to set up the servers/cloud infrastructure, develop the functionality with a good user experience, test the site in different environments and operating systems and make sure you are on point with the branding guidelines—all of which might take months. But when you're delivering content within a chat window, there's no fancy user interface to care about; you're just responding to a message and getting data into the hands of the user. A bot can be built, tested, secured and deployed in a matter of weeks (or less), not months. It’s still important to build a good user experience, but in the case of a chatbot, the most important factors are conversation flow, mapping out the variations in utterances (questions/responses to the bot) and entities (nouns in the utterances).
Our experiment is initially focused on our home office use-cases and micro bots. However, the learnings and infrastructure will be able to be redeployed for other parent bots, including one directed for our financial advisors in the field and another for our clients. By starting internally, it gives us a chance to vet the technology integration, build up engineering chatbot development skills and test the adoption of the platform. Also, some of the same data sources can be reused, such as, "What’s the status on policy xy123?" and can be modified slightly to respond to the financial advisor or client. As you can imagine, the level of compliance, testing and security increases as you go from home office, to field, to client.
We're also tackling one of the most common complaints users have with some existing chatbots—that they quickly deflect unknown questions out to the Internet: “OK, I found this out on the web for ...”
To combat this, we built two features into the platform:
- A monitor that tracks all questions asked and flags those that go unanswered, so we can use that information to build bots to fulfill new use cases.
- A dictionary of bots that surface the types of bots that exist, their functionality and the variations of utterances you can ask. This dictionary provides insight into what has been built, so users aren’t blindly asking questions with no response.
Our findings from the proof of concept will inform our rollout strategy for the Micro Chatbot Platform. Initial internal hackathon results have concluded that engineers can deliver a micro bot in a matter of days, without having a background in chatbot development. We are gearing up to provide Getting Started documentation and examples for teams across the organization, enabling them to build robust micro bots.