Impact Analytics Blog
Before it is too late: How to start with chatbots today to reap the benefits tomorrow
MARCH 13, 2018
Team Impact Analytics
What are Chatbots?
Leading retailers and other organizations around the world are investing heavily in chatbots to provide a delightful customer experience. Chatbots are being used for meeting many other business objectives such as employee onboarding, service request handling etc. as well. But what exactly are chatbots? Chatbots are artificial intelligence programs based on Natural Language Processing (NLP), Deep Learning (DL) and other similar advanced machine learning technologies, that are designed to understand a user’s query and simulate human-like responses. They are indeed ‘super-agents’ - not only can they process limitless amounts of information under the hood, but also can engage customers more efficiently by providing a personalized experience. Since chatbots save precious shopping time for customers and reduce human-related errors, retailers need to dive head-on into this technology right away to get an edge over their competitors in future.
Benefits of a Chatbot
Types of Chatbots
Chatbots are versatile applications – quick in response and highly productive yet cost-effective. They relieve the burden of manual and error-prone labor and the costs associated with it. Several prototype chatbots are being used in different business contexts, these are:
- Retail Agent Bot: An AI-based chatbot that helps retail customers with product inquiries, ordering, finding store locations, etc. IA has developed such a bot for one of its clients which can provide an engaging experience to its customers by helping them to browse items and generating personalized pairing recommendations for them. One of the most unique features of this bot is that users can upload a picture of a product they are wearing, and it will recognize the product using an Image Recognition engine (developed using a Hybrid Convolution Neural Net) and suggest products which pair well with that.
- Virtual Business Analyst Bot: It is designed to provide Business Intelligence to the enterprise internally. This chatbot helps users (like product managers, store leads/managers, sales heads, etc.) to find out how specific locations or products of the enterprise are doing in terms of different key metrics, by churning data from company’s internal databases. IA has built several such bots which helps IA’s clients understand their sales patterns and promotion performance.
- FAQ Handling Bot: This is an AI-powered bot that handles the FAQs of external or internal users. The core of this bot is built on Artificial Neural Networks and helps provide all the information (regarding aspects like company information, vacancies, team info, policies etc.) One such bot is on our website and helps visitors with information about the company.
The working and user-base of these chatbots are quite different but they perform the basic function of compiling the service-request data, accomplishing the assigned task and engaging the users instantly. Their cost-effective and ‘learning over time’ nature through artificial intelligence gives them the biggest advantage over error-prone manual labor.
Chatbots mainly use Natural Language Processing (NLP) to understand the users’ inputs. NLP is a kind of Artificial Intelligence technique which can convert natural language data into the machine-understandable format. The architecture of a chatbot typically comprises the following modules:
- Chat Client/UI: the user enters the query into the chat client/UI.
- Chatbot Engine: the acquired data from the user goes into this engine. The chatbot engine is responsible for identifying the context which is used for further processing of the text.
- Natural Language Processing Engine: this extracts the user’s intent and entities from the given utterance and sends it back to the chatbot engine. The whole process of the machine ‘understanding’ what to do is undertaken by the NLP engine.
- Data Services / Database: the intent and entities identified are then passed to this database to eventually provide the relevant response. The generated response is then returned to the chatbot engine which finally packages it into a proper format to be displayed to the user through the client.
Impact Analytics has developed a ‘specification building framework’ that helps create a model of the conversation flow that a chatbot is supposed to handle. It serves as the equivalent of a design specs document in the context of traditional software development. The detailed architecture of the chatbot in terms of specific modules and functions is defined based on this document.
NLP helps machines ‘read’ text using advanced computational linguistics algorithms. This process simulates the human ability to understand the language. Have you noticed that your email automatically tries to correct text when you are trying to send a message? This is one of the many ways in which NLP helps us perform day-to-day tasks. The most widely used Natural Language Processing tasks in building the chatbots are:
- Working Principle
Entities are first marked using the existing training data and the machine learns from that data about the syntactic information and predicts how to process the new entities.
This helps identify entities such as names of people, organizations, locations, etc., and extract relevant information from these entities.
- Working Principle
The machine is trained to identify the categories using the training data. Both semantic and syntactic information of the data is used to identify the structure and positioning of the words.
This NLP application aids in identifying sentiments and opinions of the customers and understand their views about a client’s products and services.
Generates word clouds and in-depth analysis of a range of metrics (positive, negative aspects of services) and provides valuable feedback to Business Intelligence to take the right call.
- Working Principle
Different training sentences are tagged with ‘intents’ that enables the machine algorithm to identify them. This tunes the algorithm to give the best output.
It helps extract the user’s intent using a varied set of categories by applying deep learning.
- Working Principle
This process identifies the query and sends this to the SQL database. The query is processed, and data is returned in the identified format.
Pre-defined answers are already present in a database and using cosine similarity, a score is generated that defines which answer will be displayed to the user’s query.
Queries have graph-based, summary-based, table-based and single digit answers available for the natural language queries of the user.
The Future of Chatbots
All the above applications of NLP empower retailers to understand their users better and provide a personalized experience to them for an enhanced retail experience. Retailers now understand that capitalizing on chatbots today will help them reap enormous benefits tomorrow. In fact, Gartner predicts that more than 50% of enterprises will spend more per annum on bots and chatbot creation than traditional mobile app development. 
Building fully functional chatbots is time-consuming. Most chatbot application platforms have generic data available to them but for a particular retailer, they will need specific data-set(s). The early trial chatbots which have been released by various retailers right now may not be very impressive, but they are building a rich dataset of user-interaction-pattern information, years before their competitors have even started. It is very important for other retailers to start investing in pilot chatbots now so that they can start deploying fully functional chatbots in 2019. By providing personalized recommendations and an engaging customer-experience, these tireless and super-efficient chatbots will start ruling the retail world very soon.
How retailers can get a context intelligent chatbot for Christmas 2018