Battle of Bots
Deconstructing the Bot-scape
The bots are officially here .Over 30,000 bots were launched in 2016. Ever since Facebook opened up its bot framework to developers around the world, several other biggies like Google, IBM, Amazon, Microsoft have forayed into the arena and have offered comprehensive bot frameworks. The space has seen been high profile consolidations in 2016; api.ai was acquired by Google , wit.ai was acquired by Facebook and Amazon recently acquired angel.ai .Thousands of established companies and startups alike have sprung up, boasting of their AI/NLP platforms and bot building capabilities. Select players focus on niche capabilities like bot analytics, bot discovery, bot builder etc. , making the landscape more complex. Add to that, the limitations of language and local ecosystems like commerce and payments.
Here is a quick guide to navigate the space :
1. Messaging Channels
There is a potential to reach out to 1 bn+ active monthly users on face-book messenger and 10 m + users on other messaging platforms like Skype ,Slack, Kik, Telegram, SMS etc. Based on consumer journeys and bot goals, it is important to ascertain the key channels in your bot roadmap and evaluate if the all these messaging platforms are supported . If the brand has an existing web presence or app with a substantial monthly engaged user base, it merits launching the bot on these channels.
2. Bot Platform Capabilities
There are primarily three approaches to cull out a bot solution .Each have their own upsides and downsides
One; there are bot frameworks like MS Luis, IBM Watson, Facebook’s Wit.ai, Google’s api.ai etc., offering services and tools that can be used to put together a bot solution.
Two; there are players like reply.ai, msg.ai etc. that have built out custom AI platforms (using micro-services for NLP,NLU, text/ video/image understanding ,sentiment analysis and other AI capabilities) .These provide additional capabilities like bot builders and conversational scripters and advanced capabilities like buffered learning.These mostly work on a licensing model .
Three; there are players like Gigster, Assist that have NLP experts who help customize default models based on domain specific needs and essentially assemble the ecosystem.These guys are premium and work on either a Platform as a service model or user journey based pricing .
Deconstruct the platform capabilities in terms of intent learning, text classification, entity extraction, context modelling, action, dialogue scripting, language understanding, machine learning and sentiment analysis. Design for GUI based intent, entity and dialogue visualizer/manager .Ensure you have capability to personalize experiences based on history and hence plan for maintaining state . Understand how context needs to be modeled and embedded in conversations .
If your product road-map has commerce capabilities, understand how search component and knowledge graph is integrated .
Design for handling long conversations, returning users, non- domain questions, semantics, synonyms, profanity, sentiments etc.
3. Analytics Framework
Analytics lies at the heart of a smart bot .Bot analytics, as an area is still evolving. It is crucial to design the core analytics framework and key impact metrics early on . Evaluate the framework offered by key players in this area-Botalytics, Dash-bot and Bot-metrics. Ensure that the core framework is extensible and scale-able . Make sure that the user journeys and bot interactions are saved and can be monitored & assessed as a funnel.
4. Handling transition to human assistance
Bots is an emerging area; on-boarding consumers and navigating them through the conversion funnel could be a challenging task. Users have low tolerance to bot breaks and retention could a key challenge. Bot automation with on demand human assistance has proven to be a recipe for retention .Ensure you design for handover to humans when needed and with required context for personalization. Custom portals for call center agents to supervise bot conversations, halt conversations and jump in where required would prove to be a valuable feature . Alert management capabilities to trigger mails /SMS to activate smooth and timely transitions would be good additions to your product road-map
5. Localization : Ensure that vernacular is considered when laying down the bot roadmap. For e.g. if the bot is being launched in India for a target group that essentially converses in regional languages or Hinglish (Hindi + English), understand how those languages will be supported. Though NLP/NLU in non-English languages is a research area, there are vendors who are increasingly building out intent entity libraries in regional languages and would be instrumental in supporting these languages. Other localization considerations could be payments ,wallets and commerce integrations .
6. Visual Bot Builder & Conversational Scripter
Mapping out conversation is a key first step to designing a bot .A visual bot builder would enable your designers to create and manage intent and conversational mapping at run-time in a plug and play manner. There are few niche players, which have entered this arena like Flow X, Sequel etc. It is a visual editor which would allow intent to be added and conversations to be tailored on the fly . This feature would provide speed to market and enable writing and piloting bots without worrying about server side framework. It is definitely good to have if the bot is a long running discovery or commerce bot where continually new intents would be unearthed, updated .
7. Copywriter & Design Capabilities
In the long run ,mapping out the humanized conversation trees and designing compelling user experience would be key to retention in the bot-scape .So ensure that there are interaction designers /copywriters dedicated to the engagement . Bot personality is super critical .Having a psychology major on the team to lay down the personality of the bot would go a long way .A/B testing conversational copies and carousel elements would help design for friction-less user experience .
8. Curated Content Portal
If you are building out a discovery or commerce led bot, curation and presentation of personalized content as part of conversation on the bot is critical. The UI elements (Images/cards/videos/content tags etc.)presented to the user should have relevant and compelling content to ensure a higher click through rate to the brand website and to reduce conversational drop offs .Make sure domain experts are on-boarded to curate content for conversational elements .
For conversational bots ,plan to build out a curated knowledge portal to dynamically enrich content tags and curate data would go a long way in building a compelling discovery bot.
9. Supervised learning Visualization
Once the bot has been piloted and there have been size-able user interactions with the bot, the bot should be able to improve itself by active supervised learning. Most good bots learn in a supervised fashion so it is important to have a NLP GUI which can be used to iteratively add new intents , entities and channel active learning .The visual interface would guide the learning in form of identifying and re-enforcing responses and building additional hooks for sentiment analysis.
