50 Chatbot Platforms: 2019 Edition

46 min readNov 29, 2019

In 2017, we published our article “25 Chatbot Platforms: A Comparative Table”. Recently we updated the information on chatbot platforms and added 25 more chatbot building tools. We created a new version of the comparative table where we paid our attention to artificial intelligence features, especially to Natural Language Processing, to each platform’s use cases, and to industries in which the tool can be used.

A fragment of the comparative table:

Summary

All the fifty tools are different and have their unique features. Some of them are platforms that do not require any programming skills and include visual flow builders. Others are frameworks containing advanced tools for developers like APIs, SDKs, IDEs, and others. During last few year many local chatbot building tools were developed: Recast.AI in France, Xenioo in Italy, Rasa in Germany, DeepPavlov.ai and Electra.AI in Russia, AgentBot in Argentina, Botsify in Pakistan, Engati and Morph.ai in India, and others. The newest startups that appeared in 2017–2018 are mostly platforms with code-free visual conversation builders that provide beautiful dashboards with analytics. The analytics shows metrics of the chatbot’s and agent’s performance like total users, user engagement and activity, number of conversations, average conversation length, hot topics and keywords, most frequently used intents and exit percentages, answered and unanswered questions by the bot, and so on. Usually, such platforms don’t include their own NLP engines, but integrate third-party NLP tools (conversational AI). The most popular and functional NLP tools are IBM Watson, Amazon Lex, Microsoft LUIS, Google Dialogflow, Wit.ai, Rasa, DeepPavlov.ai, Electra.AI, DigitalGenius, and Semantic Machines. Some of the chatbot building tools are provided with step-by-step instructions of the bot building process in text or video format. Many of them grow communities where it is possible to find answers on questions about the platforms. Several startups were acquired by bigger companies: Api.ai was acquired by Google, Semantic Machines was acquired by Microsoft, Motion.ai was acquired by Hubspot, KITT.AI was acquired by Baidu, ChattyPeople was acquired by MobileMonkey. There are even platforms (Imperson, in particular) that offer to create a chatbot that will speak using the right voice and a special unique personality for your brand.

Below is an overview of the most popular bot platforms.

AgentBot

AgentBot, an automatic customer service solution, was developed in Latin America and supports English, Spanish, and Portuguese languages [1]. The tool doesn’t require any linguistic or technical skills and can work with any text or voice channel. The platform provides its own algorithms, dictionaries and meanings database (the largest base of different ways of expressing the same meaning). AgentBot uses Aivo’s natural language understanding engine. The platform has enough memory to maintain coherence during long conversations, gathers customer information to deliver customized solutions, and continually evolves. The NLU engine understands users’ intents even if it contains multiple ways of asking, local terms, regionalisms, jargon, grammatical errors, and other language deviations. AgentBot is also able to recognize voice, emoji, and stickers to detect keywords, to segment customers, to predict the next question. AgentBot learns constantly, from every interaction. AgentBot answers may contain formatted text, videos, related FAQs, buttons, web view extensions, carrousels, forms, pdf files, images, co-browsing, emoji, integrations or external forms, or maps complement. AgentBot integrates with any CRM, internal system, human chat, or third-party application through the REST API.

Amazon Lex

An easy to use Amazon Lex service allows you to build, test, and deploy conversational interfaces for different applications using voice and text [2] with embedded deep learning technologies for speech recognition, speech to text conversion, and natural language understanding. With Amazon Lex different types of bots can be created including call center bots (to request a balance on an account, or schedule an appointment), informational bots (to access the latest news updates or weather), application bots (to book tickets), enterprise productivity bots (to check sales data from Salesforce), bots for the healthcare industry, and for Internet of Things. An Amazon Lex bot can be developed via Console or REST APIs [3]. The key features of Amazon Lex are high quality speech recognition and natural language understanding, multi-turn conversations, slot filling, intent chaining (transferring control dynamically from one intent to another based on end user input), confirmation and error-handling utility prompts, one-click deployment to multiple platforms, rich formatting capabilities, and powerful lifecycle management capabilities [4]. AWS Lambda, a serverless computing service that runs your code in response to events, allows you to execute your business logic, create your own back-end services, and retrieve data and updates. Java, JavaScript, Python, CLI, .NET, Ruby on Rails, PHP, Go, and CPP SDKs are available.

AmplifyReach

AmplifyReach is a conversational automation platform for sales, marketing, insurance, banking, travel, adTech, and support teams. AmplifyReach is already integrated with top CRMs and can be integrated with any backend system using REST APIs [5]. The platform does not require any machine learning or coding knowledge. AmplifyReach provides a conversation builder, some default workflow to explore, canned messages, rich UI controls, team management, smart routing of chats to available agents (a round-robin functionality), screen sharing and co-browsing, chat transcript, and proactive chat. The Core NLP engine built on deep learning techniques allows performing entity extraction, intent detection, keyword extraction, taxonomy construction, semantic relations, relationship extraction, title extraction, language identification, and text extraction [6]. The embedded Natural Language Classifier executes on IAB standard taxonomy-based classification and is applied for web pages and text categorization. Without training the chatbot replies fast to FAQs from your knowledge base and other sources, distinguishes between FAQ and active task. Analytics tools and a Dashboard allow you to track current chats, chatbot performance and failure points, agent performance, visitor initiated and proactive chat details with respect to channels and more [7].

Azure Bot Service

Azure Bot Service is an integrated environment that allows you to create, test, deploy, and manage from one place chatbots for information, commerce, enterprise, and other scenarios [8]. A web server, Bot Builder SDK, Microsoft Bot Framework, Azure Cognitive Services, and Azure Storage are the key elements of the Azure Bot Service. Using Cognitive Services it is possible to process images and natural language, recognize speech, perform searches, map complex information and data. Out-of-the-box templates for scenarios are available. LUIS allows adding conversational intelligence to understand natural language, including user’s intent, speech, or even spelling errors [9]. QnA Maker lets you build question and answer bots. Text Analytics API makes it possible to evaluate sentiment and topics to understand the user’s intent. To check and correct spelling, Bing Spell Check can be used. Bing Speech API converts speech to text and back. With Bing Web Search API you can obtain improved search details from billions of web documents. Analytics generates conversation-level reports on user, message, and channel data. Users graph shows the information on how many users accessed the bot using each channel during the specified period. The activities graph is used to understand how many activities were sent and received using which channel during the specified period. Command line tools, .NET and JavaScript SDKs are used. You pay only for what you use.

The Bot Platform

The Bot Platform founded in 2011 allows you to build bots on Workplace and Messenger [10]. World’s largest brands including BBC, Dyson, Nivea, Pandora, Samsung, and others trust and use The Bot Platform to automate their internal knowledge base, instantly answer commonly asked question and respond to customer service enquiries, engage their audience, increase productivity, automate their marketing activity, drive loyalty and monetize their audience, onboard new employees, improve company culture, and so on. The Bot Platform is easy-to-use, and it does not require any coding skills. The tool provides pre-built bots and lets you install templates quickly and easily. The platform’s cloud infrastructure and next-generation technology allow handling a large number of users, and a 99.9% service-level agreement (SLA). Messenger’s built-in NLP extracts entities like greetings, thanks, bye, date, time, amount of money, phone number, email address, and others. With analytics, it is possible to track ignored messages in which the bot did not match any keywords, audience analytics, or journey analytics (funnel analysis), and to view the most popular messages of your bot. You can integrate third-party services or add custom features using APIs.

BotEngine.ai

In BotEngine.ai, developed by LiveChat the main building block is a story, a conversation scenario between customers and the bot [11]. The story’s elements are interactions, webhooks, entities, and attributes. Interactions are responsible for what your bot does and how it behaves. The elements of interactions are: “User Says,” Bot Responses, welcome interactions that greet your clients, and fallback interactions that pop up when the bot cannot recognize the user’s input. Webhooks (HTTP push API) help you send information to your web services and retrieve the information from web services. Entities are used to extract the necessary information, for example, names, phone numbers, e-mails, locations, and so on. Attributes are usually used to store users’ information. To understand the user input and match it with the right interaction, two approaches are applied, machine learning and keywords. Machine learning uses Natural Language Processing and algorithmic probability and outputs the matching score that is compared to a confidence score letting you decide if the bot can answer the question or the help or a live agent is needed. In the second approach, the system searches for the keyword; if it finds the keyword, the matching score is equal to 1 (100%). The training tool is another useful feature that lets you add to your interactions all of the unmatched queries that had been responded by a live agent. To faster build your chatbot templates are available. BotEngine.ai is focused on English, but using AI, it works well with almost every language.

Botkit

Developers assert that their Node.js framework, Botkit, is the best developer tool for building chatbots, applications, and custom integrations for major messaging platforms [12]. Botkit includes a convenient visual conversation builder and fully featured SDK with support for all major platforms including powerful open source libraries [13]. Many plugins and middlewares are supported. It is possible to create custom dashboard metrics and statistics to stay informed about particular events or messages that have been sent by your users or your bot. For NLP tasks, third-party services like Microsoft LUIS, IBM Watson, Amazon Lex, Google Dialogflow Recast.ai, Wit.ai, Rasa are used [14]. If you have any question about the platform, you can find the answer in the Slack community or on Botkit’s GitHub.

Botpress

Botpress is an open-source based on a modular architecture flexible framework created for developers to build, deploy and manage production-grade bots in record time [15]. It doesn’t require any dependencies and can run on a computer on your premises or on any cloud provider. Botpress offers visual flow editor, authoring UI, an administration dashboard, dialogue manager, chat emulator and debugger, and a built-in NLU Engine. The NLU Engine can perform many tasks including intent recognition, entity extraction, language identification, and sentiment analysis [16]. When the bot receives a message from any channel, it processes the message using an NLU Engine and then responds back to the user with help of the Dialog Engine that uses a more complicated logic than just “if/else” logic. To add your code it is possible to draw on actions (called by the Dialog Engine to extract data, or call external services, or implement custom reply logic) and hooks (that are executed in the context they are located in). To create Botpress Modules, Botpress SDK can be used. Besides Botpress’s functionality off the shelf, many modules were created by the community and can be used by anyone [17]. Botpress was developed for enterprise giants, but also can be used by smaller businesses and start-ups.

Botsify

Botsify is an easily customized platform for building chatbots for websites and Facebook [18]. Usama Noman, the co-founder of Botsify, asserts that the platform was created for people who don’t know how to code and know nothing about chatbots [19]. In spite of simplicity for users, Botsify provides many useful features. Smart AI lets the system extract phrases and entities from the user’s input and gives the best answer. For query matching, Botsify uses intelligent algorithms and various matching criteria. If the chatbot is not able to answer the user’s question, you can temporarily turn it off using the “human fallback and takeover” feature. Furthermore, all unanswered questions are recorded and used to teach your chatbot. Botsify supports several integrations including Alexa Integration, WordPress Integration, Shopify Integration, and Zapier Integration. To quickly get started Bot Store provides templates. Fully-fleshed documentation around each feature is available.

BotStar

Botstar is an online platform that lets you visually design and develop chatbots as it has a powerful drag-and-drop visual editor [20]. You can apply ready-to-use professional templates provided by Botstar and other developers. A chatbot can be built for sport, entertainment, jobs, HR, banking, news and media, healthcare, travel, transportation, e-commerce, and other industries [21]. To understand the user’s input and extract the necessary information Wit.ai or any Natural Language Processing engine can be used. The NLP engine will allow you to perform smart training (an adaptive version of keyword training) or using NLP entities within connector conditions in the Main Flow Editor [22]. Training is a very important step to improve the ability to understand a user’s input especially for text-heavy sources including FAQs, diagnoses, or small talks. A human has an option to take over the conversation if it’s necessary. Analytics features allow you to be aware of different metrics on conversations and audiences. Illustrative graphs will show you how many messages your bot has sent and received or the user growth over a certain period of time. To add your third-party applications or to make the chatbot more functional, BotStar API (Curl, Ruby, NodeJs, Python) can be used.

Chatfuel

Tens of thousands of chatbots were created with Chatfuel to communicate with millions of users worldwide [23]. The platform does not require any programming knowledge. The main building element in Chatfuel is a block (similar to a webpage) consisting of cards that may include messages and other content (audio, video, images, text, complex). Several blocks can be linked together. The most popular functions that bots can perform were gathered into templated sets of blocks called ‘skills’ [24]. Different bot templates are available at Template Marketplace. The bot understands the user’s input and replies to user’s questions and keywords, sometimes with random responses. You can set up a few AI rules for better performance. You can also change existing rules by analyzing user inputs the bot’s AI was unable to recognize [25]. DialogFlow can be used. Analytics tools allow you to track your bot performance to evaluate its effectiveness using simple metrics like total users, conversion rates, user engagement and activity, user activity graph, or metrics applicable to your specific case (e.g., direct sales numbers, sources of traffic). Third-party analytics tools (Dashbot.io, Botanalytics, and Yandex Metrica) also can be integrated. Advanced functionality is presented by custom fields, math operations, setting up a random default answer, geofencing, routing users through your bot, responding to Facebook stickers, user-defined searches, Account Linking, and others. Many plugins are available including Google search, Bing search, RSS Import, Subscribe plugin, Digest, IFTTT, Zapier, user input, and LiveChat. The JSON API plugin enables you to integrate the backend into your chatbots. Chatfuel supports about 50 languages and offers free and PRO plans.

ChatScript

ChatScript, the next generation chatbot engine, was developed by Bruce Wilcox in 2011 and won the Loebner Prize four times. ChatScript is the basis for natural language processing tools for a variety of tech startups [26]. ChatScript is an open source rule-based engine. Rules are written in program scripts through a process called dialog flow scripting using a scripting metalanguage (a script) as their source code. UTF8 support allows scripts written in any language. The ChatScript engine’s key features are powerful pattern matching aimed at detecting meaning, simple rule layout combined with C-style general scripting, built-in WordNet dictionary, extensive extensible ontology, local machine control via popen/tcpopen/jsonopen, structured JSON data reading from websites, and others. ChatScript has a POS-tagger and parser. ChatScript runs on Windows, Linux, Mac, iOS, or Android. Integrated tools to support, maintain and test large systems are available. ChatScript is available under the MIT License.

ChatterBot

A ChatterBot is a language-independent machine-learning based Python library and a conversational dialog engine that allows generating responses based on collections of known conversations [27]. A ChatterBot is able to have multiple concurrent conversations. Training is required since a new chatbot doesn’t have any knowledge. When a user types a statement, the ChatterBot saves it with the response. The more inputs and responses are added the more accuracy increases. To respond to the user’s statement, a trained bot searches for the closest matching known statement and replies with the most likely response based on how frequently each response was used before. The additional ChatterBot’s components include a special data utility module for training chatbots, built-in adapter classes for connection to various databases, preprocessors that modify the input statement before it gets processed by the logic adapter which determines the logic for selecting a response, input and storage adapters, filters, and utility module containing useful functions. ChatterBot uses several machine-learning techniques to generate its responses including search algorithms and the naive Bayesian classification algorithm.

Converse.AI

With Converse.AI it is possible to easily create intelligent chatbots or augment human interaction without coding [28]. In particular, Converse.AI can be applied to solve many use cases like pre-sales inquiries, funnel progression, conversational commerce, customer service, post-sales engagement, re-engagement, and, of course, delivering complete end to end services over messaging networks. The features of the Converse.AI are a complete UI, integration into multiple messaging channels and third party or backend tools, directly integrated e-commerce capabilities, query and analytics engine, plain text and rich media conversations, and its own built-in NLP Parsing Engine. One of the most important blocks in Converse.AI is a conversation template (providing an answer or engaging a user in a conversation) which determines the actions after input is matched against a template [29]. Templates can be easily built using a drag and drop interface. To understand user’s input it is possible to use pattern matching or natural language parsing. In NLP Parsing Engine, phrases are the core element of every template and are used to teach a bot how to react, trigger and extract information within a particular conversation [30]. If the user’s input is not understood human escalation or fallback are used. Contexts for information extraction and as validators within the flow of a conversation are very important and widely used. To integrate third-party applications a webhook system and the fully featured API can be used. You can extend the platform by Chatflow Modules or API Event Endpoints.

DeepPavlov.ai

DeepPavlov.ai is an open-source conversational AI framework built on TensorFlow and Keras [31]. DeepPavlov.ai was developed for complex conversational systems and chatbots building, NLP, and dialog systems research. The library consists of different building blocks. Component (a function) is joined to skill (a large NLP task). An agent is a multi-purpose dialogue system that includes several skills. The framework contains pre-trained NLP models, predefined dialog system components, and pipeline templates. Besides that, developers can train and test their own dialog models [32]. The tool also includes a benchmarking environment for conversational models. The key DeepPavlov components are slot filling and NER, intent/sentence classification, automatic spelling correction, sentence similarity/ranking/ tf-idf ranking, morphological tagging, ELMo training and fine-tuning, pattern matching, and question answering [33]. The framework’s main skills are Goal (Task)-oriented Bot (an RNN combined with domain-specific knowledge and system action templates), Seq2seq Goal-Oriented bot (a bot based on a RNN that encodes user utterance and generates response in a sequence-to-sequence manner), Open Domain Questions Answering (a task to find an exact answer to any question in Wikipedia articles), eCommerce Bot (helps you to identify the most relevant product items according to your search query.), Frequently Asked Questions Answering (a FAQ skill that helps to classify incoming questions). ELMo embeddings and FastText embeddings for the Russian language are available. Users are able to perform Hyperparameters optimization and view the Parameters evolution for DeepPavlov models. Data processors including preprocessors, tokenizers, embedders, and vectorizers can be also used.

Dialogflow

Dialogflow (formerly Api.ai, Speaktoit) is a platform that allows you to develop the voice and text-based conversational interfaces powered by AI [34]. Google bought Api.ai in September 2016. Dialogflow is built on Google infrastructure, optimized for the Google Assistant, and powered by Google’s machine learning. Dialogflow is integrated with popular messaging platforms, Actions on Google, Amazon Alexa, Microsoft Cortana, and others. Dialogflow developed very functional and convenient natural language understanding tools to achieve a conversational user experience. The main element in Dialogflow’s NLU is an agent. It understands the human language and converts it into the format that computers understand, into actionable data. Usually, an agent includes several intents. An intent consists of training phrases (example phrases of what users can say), actions and parameters (entities), and responses (a text, speech, or visual response to the user). Intent can also be invoked by an event. Contexts let the agent carry information from one intent to another. When a customer says an utterance, the agent matches it to an appropriate intent, extracts parameters, and sends a response (prompts customers for more information or ends the conversation). Rule-based grammar match or machine learning (classification) match are applied. Automatic spell correction is used. Knowledge connectors allow parsing knowledge bases, FAQs, or articles to find responses to user requests. Response handler allows returning simple, static, containing minimal logic responses. It is possible to apply logic called fulfillment (using webhook) to return more dynamic, intelligent, and useful responses. Teams can collaborate on an agent. Dialogflow has prebuilt agents (cover specific use cases) and small talk agents (provide responses to a casual conversation). Agents can be exported and imported. Agents are multilingual. Analytics lets you know agent performance based on metrics like a number of sessions and queries per session, most frequently used intents and exit percentages. Available SDKs are Android, iOS, Cordova, HTML, Java, JavaScript, Node.js, .NET, Unity, Xamarin, C++, Python, Ruby, PHP, and Botkit.

DigitalGenius

DigitalGenius is a customer service automation platform and a deep learning agent efficiency tool supporting any language [35]. DigitalGenius understands conversations and customers’ objectives, drives automated resolutions through APIs, and executes repetitive processes. The platform incorporates AutoPilot, Copilot, and Control Center. AutoPilot allows end-to-end case resolution of repetitive journeys without human intervention. With APIs DigitalGenius AI connects the conversations to back-end processes that makes it possible to perform repetitive customer queries such as refund requests, order status inquiries, validate accounts, cancellations, or others without agent involvement even if it requires additional information or actions from third-party systems. In particular, the Flow Builder allows you to construct, visualize, improve customer journeys, and test different actions and their outcomes. Copilot offers the best answer to the agents liberating them from searching through templates or knowledge articles. The answer is automated if the confidence level is high enough. Copilot extracts data from customer service console and CRM and trains a neural network using historical customer service transcripts. After training, it is possible to predict case relevant meta-data. Classifying tickets and customer requests Copilot is able to route them to the right agent or team. The Control Center includes an analytics dashboard and the ability to fully control your AI models and automation. DigitalGenius integrates to any agent console or CRM such as Genesys, Oracle, SAP, ServiceNow, industry-specific systems like Sabre and Amadeus, Shopify and Magento, and others.

Electra.AI

Electra.AI was designed in 2017 for customer service, banks, retail, telecoms, and other large companies to help them optimize their work by detecting and automating repetitive actions [36]. Electra.AI makes it possible to process multiple requests at many accelerations in the service departments. It reduces service department response time by 2 to 11 times. The solution includes machine-learning algorithms, integration with RPA, and convenient AI administration tools. The tool contains dictionaries for banks, telecom, and retail, language models trained on over 3,000,000 conversations, advanced speech and language analysis technology, NLP tools for clustering, intent classification, named entity recognition, spell checking, DSL for script writing, and so on. Another important feature is real-time auto-learning. Electra.AI lets you create simple bot scripts without any coding skills or quickly write and update complex scripts using Python programming language. With Electra.AI you can also forecast the economic effect of robotization.

Engati

Engati was developed by Coviam Technologies in 2016 for fast chatbot building without programming [37]. The multilingual platform is used in e-commerce, travel, health care, banking, news and media, and automobile industries to automate a large set of tasks. The examples of tasks are order tracking and fulfillment, finding hotels and flights, medication reminding and tracking, nearest ATMs search based on the user’s location or zip code, tours and events informing, and so on [38]. Engati has an NLP engine that lets you use machine learning, NLP, and NLU to easily train your bots, manage the conversation flow and path with tags, copy your bot, apply private labeling to create custom bots instantly for different channels and customers, and upload your FAQs from csv or xls files [39]. Voice-to-text support, human takeover feature, and livechat are also available. Analytics evaluates different metrics including most frequently asked questions, actions, or cohort analysis. For integrations, JSON API and Mobile SDKs are used [40].

Facebook Messenger Platform

With the Facebook Messenger Platform, you can build different types of bots from bots handling shopping lists to e-commerce to customer service bots [41]. Connected Stripe/PayPal accounts and tokenized payments are supported. Platform Design Kit allows you to design your bot using drag and drop gestures. Example Messenger Bots include Wishlist, Shopping List, and Account Linking. Built-in NLP makes it possible to understand and extract meaningful information from a user’s input called entities (greetings, thanks, bye, date/time, amount of money, phone number, email address, distance, quantity, temperature, volume, location, duration, URL, sentiment), identify intent, automate replies, route the conversation to a human, and collect users’ data [42]. It is possible to expand the functionality of the built-in NLP engine by adding Wit.ai. Audio, video, images, files, structured messages and text, assets and attachments, message templates, buttons, quick replies, sender actions, welcome screens, and persistent menu conversation components can be added. Messenger Bot Analytics shows how your bot is being used and what users think about it. Facebook Analytics presents detailed demographic info about users and lets you log your own custom events. Page Insights allow seeing spam and block rates. With Ratings and Reviews, you are able to see a 5-star user’s review and feedback on your bot. It is also possible to integrate existing analytics providers. Other useful features are AR for Messenger Platform (beta), The Graph API Explorer for testing GET, POST, and DELETE Graph API requests, a special API for IDs and Profile Information that gets users’ profile information to personalize your bot messages. Integration components include APIs, web plugins, Messenger Extensions JS SDK, Web Plugins, Chat Extensions, full webview (allows you to offer experiences and features like picking seats to book, or browsing available dates), and webhooks. Chat Entry Points allow initiating users’ engagement.

Flow.ai

Flow.ai is a browser-based platform with a beautiful drag and drop user interface for building AI assistants or voice and chat apps for different channels [43]. The main element is a flow, a conversation topic such as ordering a pizza or greeting a customer. Flows use context and always start with triggers (speech, text, or another type of event like sharing a location). Triggers are created to reply with content or execute actions. Branches let you reply within context. Actions help developers add complex business logic and functionality. The core feature is an intent classification that needs to be trained with examples. The best practice is 10 examples for an intent. If you add five or fewer examples, only exact matching will be used. It is possible to reuse the same intent in the context of another flow. Flow.ai is able to extract data from user input using entities. To work with voice Alexa is supported. For situations when your AI engine fails to understand or handle a message, fallback scenarios are used. To get information on your customer service through HTTP posts, webhooks are applied. You can extend Flow.ai’s capabilities with Cloud code, a JavaScript that runs in a sandboxed environment. The Flow.ai works with any human language, but the fully supported languages are Dutch, English, French, German, Italian, and Spanish.

Flow XO

Flow XO consists of Flow XO for Workflow and Flow XO for Chat platforms. The first platform makes it possible to create workflows (flows) that connect things that happen (triggers) with things you want to do (actions) [44]. When a workflow is triggered, this is called an ‘interaction’. The flow listens for the specific keywords or phrases as triggers and responds with answers, web links, images, or payment requests. [45]. The second platform lets you create online chatbots quickly and easily without coding [46]. The chatbots created in Flow XO can perform several tasks including providing a virtual welcome mat to your business, gathering information, answering simple questions, pre-filtering leads, switching to live chat when a chatbot isn’t able to assist your customers, accepting payments or even providing light-hearted entertainment. Templates for common use cases are available. The analytics tools allow you to track some metrics including total users, active users, new users, triggers, messages sent, messages received, top messages, top flows.

Gupshup.io

Gupshup is an entire bot lifecycle tool that enables you to build, test, deploy, host, monitor, and analyze chatbots [47]. Bots can be created by anyone, coders, or non-developers [48]. For non-coders Flow Bot Builder, a graphical tool to build bots, is available. In IDE Bot Builder developers can create bot conversations using a scripting tool. The Bot Builder contains a simple code editor, a publishing mechanism and a diagnostics program [49]. NLP features include the semantic search, extraction of intents and entities from user queries at runtime, the ability to keep NLP logic along with other code without using separate NLP systems, pre-trained dynamically updating models, and training-free NLP or ‘NLP on the fly’ (is not necessary to train the NLP model before implementing it) [50]. The platform has built-in Wit.ai integration. Examples of bots that have been created with Gupshup include the PropWorth Bot which lets you get the valuation of your property depending upon the location, type, and many other parameters; Finder Bot which acts as a search engine for places around you; Restaurant Finder which lets you search the restaurants based on your location; Movie Trivia lets you search for information about movies, TV shows, actors; Wordster bot challenges you to solve the word puzzle and can be really fun; Recharge Bot lets you recharge your mobile number using a chat platform; Travel bot lets you search for flights between two cities; and Ecom bot lets you search for books on Amazon.com. Template Bot Builder provides pre-defined templates created specifically to business needs.

IBM Watson (Natural Language Understanding)

According to the research study by Mindbowser in association with Chatbots Journal, IBM Watson is the first choice as a bot-building platform for 61% of businesses [51]. IBM Watson Natural Language Understanding is a tool built on a neural network (one billion Wikipedia words) that provides advanced text analytics features for thirteen different languages. IBM Watson understands unstructured data, concepts, sentiment, extracts entities, relationships, keywords, semantic roles, categories, emotion, metadata, and so on [52]. It is possible to use custom domain entities and relations and create a custom model for some APIs. There are several sample apps whose purposes include banking chatbot creation, product review analysis and shopping guide generation, analysis of Twitter and SMS messages for sentiment and content, document analysis from different sources, and others. Besides IBM Watson Natural Language Understanding, other IBM Watson tools can be applied for building a chatbot. These tools are AI OpenScale (extends deployments enterprise-wide), Watson Assistant (determines in which case it is needed to find an answer in a knowledge base, in which case to ask for clarity, and when to route the customer to a human), Watson Discovery (helps to unlock hidden value), Knowledge Studio (teaches Watson the language of your domain), Language Translator, Natural Language Classifier (analyzes text, labels and organizes data into custom categories), Personality Insights (predicts personality characteristics, needs and values), Speech to Text, Text to Speech, Tone Analyzer, Visual Recognition, and other tools. Android, Java, Node.js, Python, Ruby, .NET, OpenWhisk, Salesforce, Swift, and Unity SDKs are available in IBM Watson for developers.

Imperson.ai

Imperson.ai, developed by an Israel-based company, is a platform for building, testing, deploying and analyzing enterprise chatbots for sales and marketing without coding [53]. Built with the Imperson.ai bot will speak natural language and have the right voice and unique personality for your brand if you provide the keywords and phrases that evoke your desired brand messaging [54]. The chatbots will also support throughout the marketing and sales funnel including awareness, engagement, lead generation, and sales in a two-way dialogue that deeply engages customers, addresses customer questions, resolves issues, influences purchase decisions, improves loyalty, and increases transactions [55]. The examples of created bots are Jack-in-the-Box (promotes new menu items and chats about all restaurant topics), Hyundai-KONA-Bot (answers questions about the brand new cars), Nice-inContact (promotes the customer experience platform), Hennessy (teaches about the world of Hennessy), Wonder-Lionsgate (encourages users to #choosekind in their daily life), Genius-Nat-Geo (converses with users about show’s title character’s life, how he feels about the March for Science), The-World-of-Avatar-Disney (the bot is an ecological specialist who sends guests on educational missions in the park), Steve-Aoki-Skype (invites fans to chat), Judy-Hopps-Disney (fans are able to chat and follow a series of clues to help solve four different cases from Zootopia Police Department), Miss-Piggy (makes the user experience entertaining and authentic) [56]. Imperson.ai works with many interfaces including text, voice, video, AR, and VR. NLP abilities are based on deep dialog context, NLP user intents, user sentiment, and relationship memory. The system learns constantly, remembers every user, what he said, the response he received, and how his next request relates to the previous responses. Advanced analytics lets you track metrics like the number of conversations, average conversation length, average turns per conversation, number of recurring users, user sentiments, hot topics and keywords from a central dashboard.

It’s Alive

It’s Alive is a platform that enables you to easily build, test, and deploy chatbots and services for Facebook Messenger [57]. A chatbot will answer frequently asked questions or connect people to a human if it’s necessary, collect information, enable people to subscribe to content and broadcast messages [58]. Recipes (scenarios or conditions) are the key feature in It’s Alive. This feature helps to automatically respond when the users write specific keywords or phrases. Recipes are organized into folders and consist of user triggers (conditions that automate your content) and the bot’s answers. You have to define a set of keywords to trigger a recipe and to create conditions [59]. It is possible to link all the recipes together. Thus, a conversation stream from one end of the content to another can be constructed and the decision trees can be created. It’s Alive recognizes keywords and adds the keywords that your chatbot has missed to the new or existing recipe. The platform enables you to send periodic content (RSS) automatically or manually sending a message to every chatbot subscriber.

Just AI

Just AI is a language and industry-specific, NLU-powered high-performing conversational AI that helps to take over up to 85% of recurrent tasks [60]. The platform consists of three components: Aimylogic, Aimychat (a sales tool), and Aimybox. Aimylogic allows building, training, and deploying voice and text chatbots using a simple drag-and-drop editor of the Universal Chatbot Builder. These chatbots understand speech using NLP core and become industry-specific by adding named entities. The industries where these bots are applied are sales, support, helpdesk or HR tasks. With Aimybox you are able to build talking devices performing simple tasks like weather forecasts, taxi hiring, or currency converting. The NLP features of Just AI include speech recognition and synthesis ASR + TTS, conversational skills for Amazon Alexa, Google Assistant, and Yandex Alice, intent recognition entities extraction, context retention, deep cognitive abilities, hybrid machine learning and rules-based linguistic methods for fastest chatbot training. Dashboards allow tracking various metrics including the volume of issues solved, number of failed cases, or time of issues solving. To expand offered Just AI functionality API was developed.

KITT.AI

KITT.AI allows creating conversational interfaces for home automation, conversational commerce or chatbots [61]. KITT.AI built its own platform, ChatFlow, that has a simple drag-and-drop interface which visually describes a dialogue and at the same time implements the flow that can be executed on the server as the dialogue is designed. KITT.AI also developed a fully customizable engine, Snowboy. The platform’s key features are hotword detection (no internet required), semantic parsing, natural language understanding, conversational engine (multi-turn dialogue support), neural network powered machine learning model, usability and multi-platform, quick reply buttons, and so on [62]. KITT.AI’s solutions are deployed in smartphone apps, speakers, appliances, web chat, cars, homes, conference rooms, offices, hospitals, and even telephone lines [63]. In 2017, KITT.AI was acquired by Baidu, a company that specializes in artificial intelligence products and internet-related services.

Kore.ai

The Kore.ai platform provides tools for chatbot design, development, training, testing, and publishing before rollout [64]. A built with Kore.ai chatbot can perform different tasks including execution of actions from a multi-part user request, offering guided assistance through task completion, delivering information or reports on-demand and per schedule, answering questions by finding answers from FAQs, documents and websites, holding multi-faceted, business-logic driven dialogs [65]. The Kore.ai Bots Platform has many useful features like a simple web-based bot builder tool, a graphical dialog builder, co-development support, pre-built bots, two NLP engines based on fundamental meaning (won the Loebner award, Chatscript) and machine learning, ability to use deterministic and probabilistic approaches to NLP, intent and entity recognition with declarative approach, NLP Analysis Graph for a pictorial representation of the NLP models, Knowledge Graph, and component reusability in different bots [66]. NLP tools are able to recognize proper nouns, recognize the communication of numeric values as words or digits, enable addition of synonyms, process singular and plural nouns, expand contractions, transfer vocabulary from one chatbot to the next, process tensed verbs, include pre-programmed synonyms, and personalize messages. Kore.ai includes a speech recognition engine and a built-in intelligence engine that makes it possible to define multiple forms of context for your bots, use short and long-term memory models, apply supervised and unsupervised learning based on interactions, perform sentiment analysis that can detect six emotion types, handle and execute multiple intents from one utterance, and ‘pause and resume’ intents throughout conversations. To execute the functions of the several bots a Universal Bot can be built. A Bot Analysis tool allows you to view necessary information including successfully handled and unhandled user utterances, task execution failures, script, and service performance. Platform-Specific Bots SDKs (Bots Web SDK with HTML5 and JavaScript libraries, Bots Native iOS SDK, Bots Native Android SDK) and BotKit SDK, and APIs are available.

Microsoft Language Understanding Intelligent Service (LUIS)

LUIS is a cloud-based API service that allows you to add natural language processing functionality to apps and bots [67]. LUIS can be used for conversational interfaces of internet accessible devices; in scenarios like banking, travel, and entertainment; or for a bot that answers questions defined in the FAQ. LUIS is based on machine learning. LUIS understands users’ intents and sentiment, extracts entities from sentences and uses them in high quality, nuanced language models. Entity dictionaries with billions of entries are available. You can use pre-built apps and pre-existing, world-class, pre-built models [68]. Models deployment to an HTTP endpoint is a one-click operation; it returns easy-to-use JavaScript Object Notation (JSON). Speech recognition is supported. LUIS learns and improves continuously. Active learning of endpoint utterances, phrase lists for domain word inclusion, and patterns to reduce the number of utterances needed are LUIS’s features. LUIS offers powerful developer tools. Command Line tools, LUIS Programmatic APIs, Visual Studio, Java, Node, Python, Ruby, PHP, .NET, Xamarin, Android, iOS, Swift, and Windows SDKs are included. LUIS is integrated with the Azure Bot Service that allows you to create more sophisticated bots.

ManyChat

To create a bot with ManyChat, you can use a visual drag and drop builder that does not require any programming skills. This bot will be able to welcome new users, send them news and notifications, schedule posts, set up keyword auto-responses (text, pictures, menus), automatically broadcast your RSS feed, and so on [69]. Dialogflow can be used for NLP problem solving. Flow Builder allows you to manage your flows based on messages, actions and transitions between them. Your messages can contain not only text but also other blocks like images, cards, galleries, lists, audio, video, files, user inputs, delays (making an impression that your message is not sent automatically but by a human), dynamic (get info from an external server and display it). All the flows are stored and managed in a Flows Manager. A Bot Cloning feature allows sharing pre-built solutions (templates). With Custom Rules you are able to execute actions based on different global triggers (date/time-based trigger, the tag applied, tag removed, subscribed to a sequence, unsubscribed from a sequence, custom field value changed, new subscriber). The appropriate action will be executed in case an event happens. Using Bot Fields it is possible to store the bot data and set values depending on the subscriber’s actions. Randomizer sets up an A/B split step inside the flow. The Handover feature allows passing control of the conversation to the LiveChat where agents can answer users’ questions, manage subscribers’ data manually, or send them messages. In a Dashboard, you can analyze statistics on how many subscribers you have right now, the number that represents subscriptions minus unsubscriptions for each day, users who subscribed and unsubscribed from your bot. Zapier, HubSpot CRM, ConvertKit, Google Sheets, and MailChimp are integrated. Dev tools and ManyChat API (currently in beta) give you more flexibility in customizing your bot. ManyChat offers Free and Pro plans.

Meokay

Meokay can be used to create an intelligent messenger bot for your business. Meokay analyzes and takes orders from Facebook Messenger and Facebook comments and at the same time it can push your offer or deal to users on demand, and suggest additional products and services [70]. It is able to recognize variations on the trigger words or phrases. In addition, it is already integrated with Stripe, PayPal, bKash, and other popular payment systems. Customizing discussions makes it possible to make emotional connections with patrons.

Meya.ai

Meya.ai platform allows developers to build cognitive apps using the command line, BFML (Bot Flow Markup Language), and Python [71]. Web IDE, flow and code editor, Bot visualizer, Bot Engine, test chat, live debugger, cloud computing, Bot code repo, Meya API and webhooks are among features of Meya.ai [72]. To store data for bots a fully managed, highly scalable cloud-based database is available. Different types of integrations are possible, including voice integrations (Amazon and Google technology), customer service integrations, messaging integrations (Messenger, Twitter, Slack, Kik, Telegram, Twilio), and CRM integrations. A Dashbot integration helps you track metrics of bot behavior (sentiment analysis and message counts) and user experiences (user, engagement, and retention, and demographics data). The main bot building blocks are intents, triggers (determine which sequence of actions the bot should perform), flows written in YAML (sequences of related actions), components (describe the action your bot will perform, for example, asking the user to enter some information), and transitions (tell your bot what action within the flow to execute next based on the value of a variable, a user’s input, a button click, or so on) [73]. Dialogflow, LUIS.ai, and Wit.ai are used as NLP engines. Meya.ai is able to handle multiple intents in one input. A Logging tool that stores all logs for up to 90 days can suggest the insight into your flows, components, and integrations, and it lets you understand processes under the hood. To manage bot’s content without programming the Bot Content Management System (CMS) can be applied. Meya.ai is mostly used to

support text-based chatbots,
respond to text-based triggers,
support multi-step flows or transactions,
store, retrieve, and search data (flow, user, or global scope),
run custom code components in Python,
make API calls to web-based services,
host your bot’s settings, code, and data in the cloud,
define schedules or delays to execute bot actions on time [74].

MobileMonkey

MobileMonkey acquired ChattyPeople in 2018. MobileMonkey is considered the world’s best Facebook Messenger Marketing Platform [75]. The platform provides code-free interactive Facebook Messenger marketing chatbots creation added tools. You can drag and drop different types of widgets to qualify leads [76]. The chatbot builder’s error handling tool gives you information on any issue. The Chat Blast feature lets you send messages to multiple users. With broad-matched keywords and Q&A triggers, it is possible to quickly reply to users’ questions and serve intent-based content to your customers. The bot can answer with custom text or a specially created page [77]. The chatbot learns from every interaction. You can train your chatbot using the most common unanswered questions that MobileMonkey shows you. A live operator takeover feature is also applied. Custom variables are available. You can use simple, quick, and free conversational forms that engage users and support follow-up offers. MobileMonkey developed a set of marketing chatbot templates including e-commerce chatbot, real estate chatbot, lead generation chatbot, beauty salon chatbot, restaurant chatbot, auto repair shop chatbot, dentist office chatbot, gym chatbot, personal coach chatbot, marketing agency chatbot, survey chatbot, and contest chatbot. You can clone a chatbot with just a click. MobileMonkey supports more than 30 languages.

Morph.ai

Morph.ai, used by more than 1500 businesses, is a tool that lets you build, deploy, and manage (analyze, train, converse) your chatbot [78]. Morph.ai focuses on marketing and sales [79]. Many useful features and properties like conversational UI, Converse, human fallback, Announce, natural language processing, live training, real-time analytics, support for API call, Google sheet, CRM integration, custom code, and others are available. Conversations, messages, FAQs, subscriptions, and polls sections are used to build a chatbot. Morph.ai developed its own AI (Artificial Intelligence) Engine and supports integration with Dialogflow. The AI Engine is able to parse and understand user’s input, extract entities, find the user’s intent using classification algorithms, and reply to the user. AI Understander understands Natural Language statement and identifies intents and entities. A special scoring algorithm to determine the confidence of the understanding is applied. In Morph.ai distributed models are used. Each model is responsible for a single intelligence. This makes it possible to work with about 90 languages. The AI engine is configurable and lets you train any part of it to teach the bot new things. If you need additional functionality, you can use a custom code module with Java or Python programming languages.

Motion AI

Motion AI that was recently acquired by HubSpot is a platform that enables you to create and customize bots without coding skills [80]. By the end of 2018, Motion AI will be closed in favor of HubSpot Conversations. To import your Motion AI bots to HubSpot, rebuilding is necessary because of differences in features [81]. Created with Motion AI bots allow you to qualify leads, book meetings, automate and scale chat conversations, answer common customer questions, and so on. Visual editor simplifies the process of bot building. You can tweak the copy of bot or develop it from scratch using HubSpot’s easy-to-use bot creator. Canned bot templates are available. Motion AI works with other NLP tools by means of APIs [82]. Translations into 104 languages are supported. Every bot’s reply corresponds with a “Module”. The conversational flow is defined by several modules with different behavior and created with Connections. Connections act as “if” statements and manage the flow of the conversation based on the extracted data from users’ responses and custom variables. Modules tasks are the extraction of data for yes/no questions, email addresses, URLs, telephone numbers, user’s name from user input, ISO string date stamp, representations of time and parses them into total seconds, and converting numeric and written numbers into integers. API in cURL, Node, Ruby, JavaScript, Python languages are available.

Msg.ai

Msg.ai is a scalable, reliable, and secure multilingual platform that creates AI solutions for excellent customer service [83]. It helps resolve repeatable customer needs and lets human agents focus on important tasks. BMW, Heinz, Grundfos, Nespresso, Signal, Singtel, Sony Pictures, Target, The Anne Frank House, Tommy Hilfiger, WestJet, and others use Msg.ai. Msg.ai doesn’t require any coding skills. The platform includes its own NLU engine based on deep learning and performs sentiment analysis as well as intent and tone recognition. Multivariate testing allows monitoring of user interaction with the bot, measuring the tone of the bot, and using media [84]. Msg.ai also supports deep interactive smart cards and A/B testing. Msg.ai uses short and long-term memory to remember all conversations with each customer and takes into consideration situational context. Deep reinforcement learning allows choosing the most effective action that will help customers solve their problems, propensity recognition, and even needs anticipation. Msg.ai constantly learns by working with variations and scenarios, more sophisticated conversations, and customer behavior and feedback.

Octane AI

The Octane AI platform allows easily creating and managing a Facebook bot that helps merchants to increase revenue [85]. Custom Flows let you use a targeted series of messages that are sent when triggered by a specific date, time, event, or customers’ activity. The bot is able to send follow up messages and acquire new customers, automatically answer questions sent in by your customers (training is required), recover carts, ship alerts, and provide information on new products, deals, announcements. The bot also collects data. The Questions feature helps to gather audience-generated content, opinions, polls, surveys, and ideas. Octane AI is integrated with the best ecommerce tools like Shopify, Klaviyo, Yotpo, Recharge, Gorgias, and others. The Octane AI’s stats provides metrics on revenue, products viewed, products bought, orders, customers, followers, people, messages, Convo views, Convo reads, gender breakdown, and so on.

Pandorabots

Using Pandorabots, an online web service, you can build and deploy chatbots using Artificial Intelligence Markup Language (AIML) [86]. The good use cases for Pandorabots are a FAQ chatbot connected to a messaging platform, website, or a “Concierge” chatbot designed for making recommendations about a service or product to the customer. The key elements are categories (delimit a base unit of knowledge), symbolic reductions (allow the bot to recursively call categories after transforming the user’s input), and wildcards (capture many inputs using only a single category). To create a bot, it is necessary to script many input/output pairs (“rules”) but ready libraries and integrations are available. There are also additional libraries for math, Boolean, and string operations, and other basic operations that can be used for more complex AIML programming involving SRAI handling of end-user input. The bots work very fast; it is always around 300 milliseconds, even if bots have 300,000 intents defined. The platform is multilingual. With Pandorabots API it is possible to integrate the bot hosting service or NLP engines for your application. Java, Node.js, Python, Ruby, PHP, Go SDKs are available. A large community is represented by 500,000–1,000,000 AIML developers worldwide. Two very famous chatbots developed by Pandorabots are OpenWeather Bot (allows you to access OpenWeatherMap.org current weather information through their Weather API) and Mitsuku (a four-time winner of the Loebner Prize Turing Test, the world’s best, most human like, conversational chatbot that speaks with millions of people monthly).

Pypestream

Using its patented framework of ‘Pypes’ and ‘Streams’ for natural language processing, Pypestream creates AI-driven conversational solutions [87]. The industries where Pypestream can be used are aeronautics, automotive, cable, call centers, business process outsourcing, financial services, health care, insurance, logistics and shipping, retail, telecom, travel and hospitality, and utilities. Pypestream’s Natural language understanding engine is based on machine learning algorithms. To understand the user’s intent and extract entities, machine learning classifiers are trained. Entities help improve the engine to bring in more context and detail. Pypestream’s NLU provides 1000 pre-built entities and allows creating custom entities. The platform can be integrated with modern live agent vendors that use REST APIs. An open and flexible API platform also allows the development of third party connectors, plugins, and extensions. The platform includes the Smart Messaging Framework, Pypeconnect SDK, Pypemanager, the Pypestream mobile app, as well as API plug-ins [88].

Rasa

Rasa is an open source tool for creating contextual AI assistants built for developers, by developers [89]. Rasa includes three components, Core, NLU and Platform. Rasa Core is a chatbot framework with machine learning-based dialogue management [90]. Rasa NLU is a library created for natural language understanding. Rasa NLU allows you to determine intents (including multiple intents) and extract custom and built-in entities from a user’s input with a confidence score [91]. To start working with Rasa, NLU you have to prepare your training data then define and train your machine-learning model. You have the ability to provide training data as markdown, as JSON, as a single file, or as a directory containing multiple files. It is possible to apply tensorflow_embedding pipeline (doesn’t use any pre-trained word vectors, but it fits these for your dataset, and it can be used if you have 1000 or more total training examples) and spacy_sklearn pipeline (uses pre-trained word vectors from either GloVe or fastText; it can be used if you have less than 1000 total training examples). Rasa NLU works with any language using tensorflow_embedding pipeline since it trains custom word embeddings for your domain. Other approaches (using spacy-sklearn, MITIE, Jieba-MITIE) are based on available pre-trained word vectors and don’t support all languages. You can save your models using S3 and GCS. Models from Wit, LUIS, or Dialogflow can be used in Rasa NLU [92]. Rasa NLU has an evaluate mode which allows developers to understand how is the model performing, if there enough data, and if the intents and entities well-designed. It is also allowed to create a custom component for a specific task. Rasa Platform offers APIs, a graphical user interface, and enterprise-grade support. To deploy a bot with Rasa Platform, you need to train and make models available to the Platform’s Rasa Core and Rasa NLU containers. Also, you will need to create a docker container for actions execution [93].

Rebot.me

Rebot.me is a very simple, free tool for creating chatbots for personal purposes and benefits [94]. To start developing a chatbot, first you need to create a free account. Then you can customize the bot using various tools and teach it to respond to certain questions. The more examples you provide the more questions the chatbot will be able to answer. It is recommended to teach the chatbot to respond to questions users will likely ask [95]. The chatbot also can be applied to product or services promotions, answering simple customer’s questions, gathering information on what the customers are thinking about your product. Many bot examples are provided.

Recast.AI

Recast.AI, a French bot platform developed in 2015, is a collaborative tool to build, train, deploy and analyze intelligent bots for developers [96]. 50,000 bots have already been built with Recast.AI. Recast.AI includes three modules: Bot Builder, Bot connector, and Bot Analytics. There are several steps of the Bot Builder process. The first step is getting the user’s input from a messaging channel by Bot Connector. The next step is Natural Language Processing that can be done using NLP API. NLP API returns JSON that is used in the last step, managing the conversation and context. A chatbot developed with Recast.AI contains two elements: skills and a training dataset. A skill that can be business, floating, or fallback is a part of the conversation with a concrete purpose that your bot will perform. A skill includes triggers (the conditions that need to be completed for your skill to be initiated), requirements (determining the information the bot needs to retrieve from the user, and how to retrieve it), and actions (performed by the bot when all requirements are complete) [97]. A training dataset includes sentences organized into intents obtained from users’ interactions. NLP API allows the bot to understand user’s intents (that should contain at least 30 expressions) and automatically detect 31 different entities such as date, time, location, person, and so on. A list of words belonging to an entity is called a gazette. Different formats for messages are embedded including texts, cards, buttons, quick replies, carousels, lists, images. To retrieve business information or connect to an external system, webhooks are used. Analytics is available on conversations, users, messages received, average messages by conversation, most used intents, entities, and skills. Recast.AI developed several open-source libraries and hosted them on GitHub SDKs.

Reply.ai

Reply.ai is a visual bot builder that allows successfully building, launching, and scaling chatbots from prototype to production [98]. The industries the bot can be used in are insurance, travel & hospitality, food & beverage, consumer electronics, agencies & resellers, and SMBs. The examples of bots are Education Bot that answers questions about insurance, Policy Bot that collects information on your customers, and The Morton’s Brine Time Alexa Skill that teaches how to brine a turkey. In Reply.ai machine learning and natural language processing are widely used. It is possible to integrate wit.ai and DialogFlow, or use the Reply NLP engine [99]. Other features of Reply.ai are a built-in CRM and customizable, real-time analytics dashboards, support of multiple languages, markets, and channels, enterprise-grade security, out-of-the-box integrations, and scalable deployments.

Semantic Machines

The Semantic Machines focus on the development of fundamental conversational language-independent AI technology that goes beyond understanding commands to understanding conversations [100]. The Semantic Machines’ use cases are business, productivity, shopping, e-commerce, concierge, travel, calendar, search, and automotive [101]. The key platform features are the Conversation Engine, deep learning, speech recognition, speech synthesis, reinforcement learning, and a large-scale training corpus [102]. The Conversation Engine extracts semantic intent from text or voice multi-turn conversations, maintains contextual understanding, and creates a self-updating learning framework for managing dialog state, salience, and end user goals. Natural language generation (NLG) technology allows formulating communication with the customer taking into account the dialog context. Neural network systems (deep learning) are used for semantic analysis, acoustic and language models, natural language generation, and speech synthesis. Reinforcement learning permits the bot to learn new domains and enhance semantic understanding. The speech recognition platform provides unique capabilities and performance. Speech synthesis will enable conversational computing for the first time. Semantic Machines created the world’s first large-scale training corpus for machine learning spoken and written dialog (automatic annotation and alignment of data at scale). Semantic Machines also provide Developer Tools to add the conversational AI technology to their applications or teach new skills within existing domains. In 2018, Semantic Machines was acquired by Microsoft.

Smooch

Smooch is mostly used by customer-centric software makers and bot platforms because it allows embedding conversational capabilities to applications [103]. Smooch Conversation Cloud lets businesses get in touch with their customers across popular messaging channels. Smooch automates conversations, manages conversation participants, and builds conversational intelligence using third party NLP and AI engines. Rich messages templates can be created and sent to users. Smooch provides API libraries for JavaScript, Ruby, Python, and Java, as well as fully customizable iOS, Android, and Web SDKs.

Streebo

Streebo DX Accelerator lets you build a chatbot using ‘no or low code’ tools [104]. Streebo provides ‘drag and drop’ interface, chatbot branding and UI/UX customizations [105]. In Streebo a user’s input can be understood using NLP engines like IBM Watson, Alexa, or Wit.ai that identifies keywords and determines users’ intent. If the input is not understood, humans can take over. Real-time chat analytics are provided. You can easily integrate CRM, ERP, and messengers like Facebook, WhatsApp, Slack, or other third party applications.

Twyla

Twyla learns from agent or customer live chats, blends machine learning and rule-based algorithms, answers questions and deflects tickets, increases revenue, and lowers support costs [106]. To generate the Conversational Intelligence that powers TwylaBot Twyla’s NLU works with structured data (historical support conversation logs) and unstructured data (knowledge bases and FAQ content). TwylaBots are trained for enterprise systems like ERP, CRM, eCommerce, support, helpdesk, and live chat systems. Analytics tool tracks KPIs and measures ROI leveraging powerful analytics dashboards.

Wit.ai

Wit.ai is an open and extensible natural language processing platform [107]. It is one of the most usable platforms. Forty-five percent of the respondents in a study by Mindbowser said they trust Wit.ai more than any other bot-building platform [108]. Over 180,000 developers already use Wit.ai for bots, mobile apps, home automation, wearable devices, and hardware. Wit.ai understands the user intent from text or speech, extracts different types of entities (returning confidence level), performs sentiment analysis, and works with context and actions. Wit.ai provides built-in entities [109] that were trained across all Wit.ai’s datasets. It is also possible to create custom entities. To configure Wit.ai, it is needed to provide examples showing what customers may say. Wit.ai also learns constantly from every interaction. All learned examples are saved and shared across developers of the community. A useful Wit.ai’s feature is the ability to reuse the data from the community that lets developers fork each other’s code. There are several official clients including Node.js, Python, Ruby, and HTTP API for other platform integration. Developers can use Wit with iOS, Android, Windows Phone, Raspberry Pi, Python, C, and a JavaScript plugin. Wit.ai supports more than 50 languages, and it is free.

Xenioo

Xenioo is a platform that allows you to build a chatbot without coding using a Chat Flow Designer [110]. Behaviors, interactions, and actions are three main concepts of Xenioo chatbots. An example of a behavior is “Support and Feedback.” Examples of interactions are “Ask about a problem,” “Ask for input,” and “Evaluate Input.” Xenioo provides an NLP engine. The main element in Xenioo AI is the intent with an expressions list. Xenioo AI Engine also automatically extracts any defined entity. The variable and tag Condition Switch Operation feature lets you switch the flow accordingly instead of simply making the flow go on. Using automatic redirection feature, it is possible to activate a specific behavior or Interaction whenever a given intent is detected [111]. The Xenioo chatbot can be linked to IBM Watson Assistant or Dialogflow. At any stage of building your bot, you can test it using the preview panel. The Execution Diagram shows you a detailed information on your bot execution steps. Many existing templates, tutorials, and the in-application data can help you create your chatbot.

Summary

In this article, fifty chatbot building tools were described and a comparative table for them was composed. All the tools are different and have their unique features. Some of them are platforms that do not require any programming skills and include visual flow builders. Others are frameworks containing advanced tools for developers like APIs, SDKs, IDEs, and others. During last few year many local chatbot building tools were developed: Recast.AI in France, Xenioo in Italy, Rasa in Germany, DeepPavlov.ai and Electra.AI in Russia, AgentBot in Argentina, Botsify in Pakistan, Engati and Morph.ai in India, and others. The newest startups that appeared in 2017–2018 are mostly platforms with code-free visual conversation builders that provide beautiful dashboards with analytics. The analytics shows metrics of the chatbot’s and agent’s performance like total users, user engagement and activity, number of conversations, average conversation length, hot topics and keywords, most frequently used intents and exit percentages, answered and unanswered questions by the bot, and so on. Usually, such platforms don’t include their own NLP engines, but integrate third-party NLP tools (conversational AI). The most popular and functional NLP tools are IBM Watson, Amazon Lex, Microsoft LUIS, Google Dialogflow, Wit.ai, Rasa, DeepPavlov.ai, Electra.AI, DigitalGenius, and Semantic Machines. Some of the chatbot building tools are provided with step-by-step instructions of the bot building process in text or video format. Many of them grow communities where it is possible to find answers on questions about the platforms. Several startups were acquired by bigger companies: Api.ai was acquired by Google, Semantic Machines was acquired by Microsoft, Motion.ai was acquired by Hubspot, KITT.AI was acquired by Baidu, ChattyPeople was acquired by MobileMonkey. There are even platforms (Imperson, in particular) that offer to create a chatbot that will speak using the right voice and a special unique personality for your brand.

Resources:

50 Chatbot Platforms: 2019 Edition

Written by Data Monsters