In part one we discussed about the importance of chatbots in the enterprise and what it takes to make chatbots enterprise ready.
Continuing the discussion, we will cover more areas with respect to making chatbots enterprise ready.
Human – Machine interaction has never been easy. We all adhere to Language being a key differentiator that makes us human .Humans would like to interact with machines in the same way they interact with other humans. This humanization of machines has led to the growth and the need for AI. Machine learning and deep learning.
Chatbots are an early and novel way to begin this man – machine interaction journey.
In chatbot terminology, interactions between humans and chatbots can be codified as dialogs.
Almost every enterprise customer would like to customize the dialogs to make them more tuned, relevant and crisp for their customers, prospects and stakeholders. They may also choose different conversation flows to improve the user experience among different business needs, functions, countries and customer segments .
Some enterprises may want to white label the chatbots meaning they want to personalise the chatbots according to their specific requirements. Visual element changes may also be requested by certain enterprises to make the chatbots more user friendly.
Additionally, integration with internal systems will have to be customized if enterprises do not provide standardized integration interfaces.The chatbot platform would have to provide the above features to build customized Bots. Building every bot from scratch would be time consuming. So, providing bot templates and bot platforms like Engati www.engati.com for various use cases or industry segments will accelerate the bot building process.
Natural language understanding (NLU)
Do we need NLU for chatbots? Isn’t NLP sufficient for making bots understand human language. This is debatable and depends on the uses cases and how much sophistication and accuracy is required from the bots.
We will all agree that we would like chatbots to interpret tone, sentiment, emotion, analogy, simile, metaphor, perception, abstraction and experience to understand conversations with humans. Most chatbots of today employ a retrieval based model – the responses are retrieved from a database based on the input provided.
Generative based model is the future and this model will enable generation of responses in real time i.e. responses need not be pre-defined. Generative models based on deep learning techniques are most promising.
Since enterprises will be using chatbots to augment decision making and in many cases the chatbots will be performing critical functions, it is but expected that they understand conversations and respond with a high degree of accuracy.
Most of the use cases handled by chatbots are related to understanding the queries of the business. Thus, intricate knowledge of the business will be a prerequisite for employing chatbots in business. For example, if we talk of a customer support chatbot for e-commerce, the chatbot should understand business terminology and processes to go about answering questions related to a customer’s order.
A significant part of this domain knowledge will have to be imparted through training. In many cases machine knowledge may have to be augmented by human knowledge to provide better service.
Most enterprises will employ chatbots to solve business problems or improve efficiency of existing business processes. As such domain specific knowledge will be needed rather than generic knowledge. SMEs on the domains have to design and train chatbots for specific domains. In future, we will have B2B chatbots that will conduct business transactions. High degree of domain knowledge (at par with humans) will be needed for such use cases.
Every enterprise will have regulatory requirements for auditing business processes and transactions. Since chatbots fall in the path of business flows they will have to log all events and interactions between users and the enterprise. This will create a path to keep track on traceability, reconciliation and for resolving conflicts, if any.
As chatbots take on more and more business-critical functions, providing a foolproof audit mechanism will become a mandatory requirement by enterprises. The information captured by the chatbots can be used to detect fraud, find out irregularities in order to enforce ethical business practices. Just as customer support calls are recorded for analysis and training purposes, the recording of events and interactions of chatbots can be used to train the chatbots for better performance. Chatbots providing superior auditing capabilities will have a distinct advantage over the others that don’t, when enterprises are evaluating them.
Chatbots are a global phenomenon, thus most global enterprises will require chatbots to cover internationalization (i18N) requirements. At a minimum chatbots should provide i18N support similar to those provided by business applications such as translation into multiple languages, multi byte character set support, character encoding – preferably UTF-16, multiple time zone etc.
If the chatbots support voice recognition feature, then they should support this feature in multiple languages.
The biggest challenge will be providing NLP in multiple foreign languages. Translating the user input into English and then using an English trained NLP engine will not work out.This could be due to the quality of the translation or the training provided in English, since in different languages an expression may be interpreted in different ways.
The best way would be to train the NLP engine in the foreign language using native expressions. However, this will create a problem of scale, since as many trained NLP engines will be needed as the number of supported languages.
As we have seen from the discussion above, it is fairly challenging to meet the requirements of an ‘enterprise chatbot’. Those chatbot platforms that can provide these features will have a significant advantage when enterprises are evaluating chatbots for their needs.
By: Anwesh Roy
Also published on Medium.