Putting Alexa to Work: Moving Conversational UI from Hype to Reality
The rise of conversational UI signals exciting progress for the BI world but there are pitfalls to be avoided. This blog presents 3 considerations for guiding your conversational UI implementation to ensure success and maximize the value of your data analytics.
By Mary Grace Glascott of Narrative Science.
If you are lost, you ask Siri for directions. Hungry? No need to pick up the phone, there’s a Domino’s Pizza bot that will calm your grumbling stomach. Wondering if you need an umbrella before you leave the house? Ask Alexa about the weather (and if you don’t have an umbrella, just ask Alexa to order you one).
We are used to interacting with conversational agents in our personal lives, depending on them to execute the mundane tasks that we were required to do ourselves in the past. Though a small portion of us may be nostalgic for the days of flipping through phone books to find a number or fumbling with maps to locate a destination, those days are gone. We’ve entered the age of conversational systems, and there’s no turning back.
Powered by artificial intelligence (AI), Siri, Alexa, Google Assistant, Slack bots and their friends are getting smarter and smarter: more accurately translating our speech into text, better analyzing the nuances of our intended sentiment, and getting it right more times than wrong when responding to our questions.
Now, AI-powered conversational systems are making their way to the enterprise, with chatbots and other types of virtual assistants powering automated dialogues in call centers, business intelligence applications, and internal employee portals. With voice-enabled applications like Alexa acting as a front-end to data and analytics tools, this dialogue is increasingly happening in our everyday work environments.
Take the mundane tasks associated with analyzing data, interpreting dashboards, and writing reports. Instead of spending time performing these processes, what if you could simply ask your mobile dashboard, “How are sales this quarter?” In turn, you would receive an automated response, “Sales are just over $1 million this quarter, driven primarily by a big win with Acme Corporation at $800,000. Pipeline looks good for the remainder of Q1, and sales are projected to exceed their bookings by $200,000.” That reality is not far away, with Gartner projecting that by 2019, 40% of enterprises will be actively using chatbots to facilitate business processes using natural-language interactions .
The future looks bright but currently, the conversational UI experience is one of limited success in the enterprise. User frustration levels with virtual agents are heightened due to chatbot experiences in our personal lives. Expecting a similar experience to consumer applications in terms of relevance and immediacy but being disappointed by results, many enterprises view most conversational integrations as “experiments.” Coupled with the fact that “getting it right more times than wrong” is not a sufficient bar of success in the workplace, it’s necessary to separate the hype from reality.
Whether working with an internal team or an external vendor, here are three considerations for guiding your conversational UI implementation so it’s a need-to-have, and not just a nice-to-demo:
- Make it truly conversational.
Think about a memorable conversation you recently experienced. More likely than not, it was a dynamic two-way dialogue, moving beyond a simplistic Q&A paradigm, where you dug deep into various areas and ended up at a different place from where you began. So too should be our experience with intelligent enterprise systems. Instead of simply asking a question and the machine responding with an answer triggered by a pre-built library of responses, it needs to understand your intent: why are you asking this question?
Back to the sales example response: it not only contained the answer to your question (“sales are just over $1 million”), but an explanation of why (“driven by a big win”), with relevant follow-on information that you care about (“pipeline looks good”). In order to engage in a dialogue, these systems cannot be purely driven by search (read here for the different types of chatbots, there is a time and place for search-driven systems.) The system needs to have an understanding of your intent and include the analytic capabilities to identify the insights to meet that intent. For example, “driven by a big win” requires the system to perform a driver analysis to identify the key contributors of performance. Finally, those insights need to be transformed into conversational language. Advanced Natural Language Generation (Advanced NLG) systems are able to perform these capabilities.
A BI bot from Sisense, conversing in natural language powered by Narrative Science’s Advanced NLG platform, Quill.
- Teach it about your business.
Facebook recently “benched” its enterprise chatbots due to their 70% failure rate, acknowledging the bots did not have the required domain expertise to be valuable . So how do you teach bots relevant domain expertise? It’s all about the knowledge base, or in other words, the intelligent systems’ ability to convey concepts about your business. It must have the contextual understanding that salespeople have quotas, benchmarks, and pipelines. Without this knowledge base, you won’t have a very insightful or articulate dialogue.
- Avoid the black box.
If you are unleashing conversational agents to interact with your clients or explain business performance, you need to make sure it is able to explain the reasoning behind its decisions. It’s imperative that all automated communication from virtual agents contain a transparent audit trail which traces every query, analysis, and word choice back to the system of record. Without an identifiable logic trace, you have a black box, and you don’t want a black box communicating on your behalf.
Conversational intelligence represents a huge opportunity to broaden access to relevant information across the enterprise. While the potential is exciting, it is important to remember that a truly impactful virtual assistant will be one that understands our intent, speaks with the expertise of our business, and explains the decisions it’s made. If you remember those requirements as you embark on your pursuit of conversational capabilities, then talk won’t seem so cheap anymore.
 Gartner, 4 Use Cases for Chatbots in the Enterprise Now , Feb 2017.
 Facebook Inc’s Chatbots Hit a 70% Failure Rate, Motley Fool, Feb 2017.
Bio: Mary Grace Glascott is a Director of Product Marketing at Narrative Science.
- Building a Bot to Answer FAQs: Predicting Text Similarity
- Artificial Intelligence and Speech Recognition for Chatbots: A Primer
- A Concise Overview of Recent Advances in Chatbot Technologies