With over 1500 attendees, and more than 100 speakers, the Chatbot Summit is the place to be to learn about and engage in conversation about the latest developments and future of chatbot technology. The conference focused on bringing the leading chatbot community closer together helped in developing and bringing the world’s best digital experiences based on natural language technologies to market. Here are the top 4 lessons I learned from this summit.
Lesson 1: A homogeneous technology
There is a huge number of bot technology providers, but the vast majority share the same vision of how to implement a chatbot, which can be summarized in the following three principles:
- use of machine-learning based NLU to map user input with application’s Intents. (The Intent is the intention of the user, or users interacting with the chatbot.) The Intents must be manually defined for each application. The client must provide a lot of examples to give the bot a correct understanding of the user input;
- the dialog is designed using static decision trees, often called “stories” or “scenarios”. Providers usually propose a dedicated UI (user interface). Some base their representation on standard tools, such as BPML (a language for visually modeling business processes), while some enhancements propose slot-filling mechanisms at points of the dialog tree;
- the answers provided by the bot are static: either canned answers or sometimes templates filled with variables collected during the dialog;
The main pain-points in setting up a chatbot are thus the time for collecting examples and the (manual) tracking of Intents that are often very similar to each other and thus can lead to misunderstanding. A tool that is working to address these pain points is ChatBase, sponsored by Google, which uses the transcripts from support centers to automatically propose Intents from conversations. This feature is still in Beta version and requires massive input data. A step in the right direction, but still an area where we look to see further transformative innovation to deliver a truly great customer experience.
Lesson 2: Mix bots and humans
Some bot vendors try to escape the standard paradigm: user chats with the bot, escalates, if necessary, to a human in the support center. A Swiss insurance company implemented Experflow, a solution that mixes the bot with human interaction, using the following mechanism: if the bot can answer, then it does, else the human takes the control. This ability to change between the bot and human operator continues for the whole dialog.
This has technically the side-effect of a constant human monitoring of the bot and the possibility of adapting the decision trees by tracking the paths where the bot fails.
This is a very innovative way of keeping the “human in the loop” in support centers., The main difficulty of this approach is the acceptance by humans of this automated assistance that can take the lead at any time during the conversation.
Lesson 3: New conversation media
As the era of “smartspeakers” is rising (Amazon Echo, Google Home, Apple Home Pod), the attempt to counter Facebook Messenger’s dominance is becoming clear. As privacy concerns begin to tarnish Facebook’s image, several players are capitalizing on this opportunity for other conversational platforms:
- Apple’s iMessage Business Chat allows the integrator to customize an SMS/iMessage-like UI with specific branding, and expands the UX to integrated applications,
- WhatsApp proposes a similar approach with WhatsApp for Business,
- A product called Google My Business is also in preparation, providing enhanced messaging capabilities on Android devices.
Lesson 4: From a bot that works to a bot that wows
Several visionary talks agreed on the following point: a bot will really change the user experience if it can deliver a truly personalized experience and show some proactivity in its dialog. This is where a bot will really make the difference to the customer experience and stand out from the crowd.
If we summarize it in more technical terms, current bots are – more or less – good at understanding (provided you feed them with enough examples) but restricted to static dialog patterns (the dialog trees) and really lack NLG technology to generate personalized answers instead of canned texts.
This is where Yseop technology can really make the difference: with an advanced dialog engine and best-of-breed generation and personalization capabilities, Yseop Smart Personal Advisor can really be a game-changer in the chatbot field.