Chatbots Overview In Non Technical Terms
Written By: Jerry Witkowicz firstname.lastname@example.org
1.0 What is a Chatbot?
A chatbot is an intelligent robot referred to as bot. There are two types of bot models, rule-based and AI (Artificial Intelligence) chatbots.
A chatbot is basically a robot application that has been trained (programmed) to perform specific tasks by interpreting users requests submitted in voice or text commands and to respond to users with responses that range from simple scripted and pre-defined responses to self-determined by a chatbot in human like responses. The level of capabilities that chatbots can deliver is essentially determined by the level of AI functionality that the chatbot utilizes and the data it contains.
The simplest one to deploy in any business is the rule-based Chabot model and is the most widely used in the industry. The more sophisticated AI chatbot models require considerably higher level of Artificial Intelligence technology, research, data and expertise.
2.0 Chatbot Models
The two most commonly deployed chatbot models are rule-based and AI chatbots. Conceptually they can be deployed to perform similar functions but with different levels of capabilities and human intervention. As the AI technology continues to evolve, vendors are creating a hybrid of chatbtos that fall between these two models.
2.1 Rule-based model – The rule-based chatbots behave exactly what the name implies; they follow a predefined conversation script. This model of chatbots is basically trained to respond to user’s keywords or intents and to respond with pre-defined scripted responses that are designed to accomplish specific tasks. The user asks a question or states a request and based on the data that tells the bot how to respond, the chatbot will respond according to the provided data.
This model is a bit more complicated than that. In this simpler model, the chatbot uses natural language processing (NLP) technology to recognize users requests submitted in voice or text.
For this model to continually improve, ongoing human intervention is required to monitor; how well the chatbot responds to users, where does it fail, where scripted information is missing data, where transfer to human agents could have been avoided and more. This type of monitoring enables the business to continually fine tune the chatbot to continually improve its intelligence and ability to handle more tasks.
This model is also less likely to generate errors as the responses are pre-defined but at the same time this model is also limited to conversations that have been pre-defined in the script.
By comparison to the AI chatbot, this model is significantly simpler to design and deploy and there are number of vendors which provide technology platforms and services to help any business design the conversation scripts and integrate with the business environment.
2.2 AI chatbot model – An AI chatbots as the name implies uses sophisticated Artificial Intelligence technology that may include machine learning and deep learning algorithms to name a few. Unless the business plans to develop its own AI chatbot and its own technology, there are vendors who specialize in developing AI chatbot models.
Developing an AI chatbot model requires significant investment and expertise. For example, Google chatbot Meena which is a human like chatbot utilizes 40B words mined and filtered from public domain social media conversations and uses 2.6 B parameter neural network.
Meena and similar Artificial Intelligence chatbots use deep learning algorithms to improve the quality and relevance of their responses. Instead of scripted responses based on specific text input, they detect users’ intent. This type of chatbot requires significantly more effort to design and deploy.
3.0 Business Applications for Chatbots – In a survey conducted by Drift (chatbot vendor) in 2018, the survey results (chart below) showed where businesses intend to use chatbots. Clearly these results indicate that customer facing business functions are the most popular candidates for chatbots to be deployed in.
4.0 What is involved in deploying chatbot?
In our research to answer some of the basic questions that business leaders are asking such as; how much effort is required and what are the overall costs in deploying a chatbot, we were unable to find a standard and proven model that held such answers. What we did find were common best practices used by businesses when planning and deploying chatbots in their environments.
4.1 Manage Expectations – Set specific expectations what the chatbot should do and in which business function. Define what is the overall goal the chatbot is to achieve, what improvements and where are expected? As obvious as this may sound, companies often fail to do this and are disappointed with the results.
4.2 Manage the scope - Avoid trying to do everything initially. Select an area of your business where your business experiences most frequent and repetitive interactions with customers. Interactions that can be automated by a chatbot. You can always expand as the business learns by monitoring and evolving the chatbot.
4.3 Create training data source – This is one of the most critical steps in deploying a chatbot regardless of which model is chosen. An effective chatbot requires a massive amount of training data in order to quickly solve user inquiries without human intervention. The primary requirement in chatbot development is to obtain factual, task-oriented dialog data to train the chatbot.
Chatbot training data is the critical information that enables the chatbot to understand what users are saying, and how to respond. To build an effective chatbot, the business must first compile this data and install it in the chatbot. This is one of the more significant efforts involved.
For a business which is considering deploying the simpler rule-based chatbot, conversations scripts will need to be defined, scripts which are specific to the customer/user interactions with the business.
There are vendors which sell generic data that may help establish a base script. However, every business has is unique in their business model and in their relationship with their customers. Bulk of this unique data will have to be captured and recorded by the business and from within the business.
Conceptually, all of the required data already exists within the business. It may be in many formats and many places. For example, if the chatbot is automating specific human handled interactions with the customers, the required data may already be recorded. Many companies alert customers “this call is being recorded for training purposes” which suggests that customer conversations are already recorded. E-mails received from customers and responses would create another source for this data. A social media interaction is another source of conversation exchanges between the business and customers.
These data sources will enable the business to design and deploy a chatbot that is unique to the business and its brand. Only the business knows how much, where and what state of readiness such data exists. And only the business can assess how much effort will be required to compile this data.
4.4 Learn and evolve – More specifically in the rule-based chatbot, it is critical for the business to continuously monitor its deployed chatbot and learn from it. In order to understand where chatbots respond well, where they fail and where they could have done more, it is essential to monitor its performance. Chatbots should not be considered a static application. Chatbots even in the rule-based model can improve and take on more tasks with additional training.
In the AI chatbot model, human intervention will also be required. Depending on the deep learning algorithms that are deployed, the chatbot will not be 100% accurate. It will make mistakes and it can learn incorrectly. Human intervention will be required to monitor how well the AI chatbot is responding, how accurately is it responding and where mistakes are made. The level of expertise for monitoring both models will be similar. Both chatbot models require technical knowledge of the chatbot operation and have the ability to modify data as needed.
4.5 Where to deploy it – Selecting where to deploy your chatbot is as important as developing one. Consider where your business wants to or is interacting with your users/customers. For example, is your company website used to make purchases, reservations, inquiries or access support? Are messaging apps like Facebook Messanger, WhatsApp, Slack, Telegram used by your users to interact with your company? Are mobile channels messaging like SMS and USSD used to connect with your company? These are some of the channels business could consider where chatbots could perform a routine and repetitive tasks by rule-based model.
A more capable AI chatbots could be considered in customer facing functions where the business wants to upsell products. For example, an AI chatbot could have the capability to monitor; what visitors click on the company website, which products they view and how many times, at what step they abandoned their navigation, and using this real time learned behaviors data, the chatbot could suggest another product, offer special deals, ask the visitor questions on what they were looking for and more.
4.6 Overall effort required - Majority of rule-based chatbots are being deployed using technology platforms that many vendors provide. This type of chatbot can be developed by the company with little to no coding knowledge. Vendors offer consultations to help businesses expedite deployment of such chatobot. Time required to design and deploy a simple rule-based model chatbot has been suggested to take 3 – 4 weeks. We need to stress that this is a simple rule-based model where script data is readily available within the business deploying the chatbot.
The effort to deploy a more complex chatbots even when using the rule-based model cannot be simply guessed. As indicated in this tutorial, the state of the critical training data will impact the effort needed to deploy even a rule-based chatbot.
Deploying an AI chatbot is a custom design and development. It is impossible to estimate the effort for deploying this model of chatbot at least at this time. Perhaps as this technology evolves and becomes available in a more off the shelf type,it will be possible to develop a typical deployment effort model.
5.0 Examples of Chatbot Deployments
Today and as the chart below suggest, chatbots will be used and are used across many industries and in many different capacities. There are many actual deployments of chatbots today far too many to list here. We share some examples of chatbots in use in some of the key industries e.g. healthcare, food, travel and retail.
U.S. Chatbot Market by Vertical, 2014 – 2025 (USD Million)
5.1 Virtual Doctor – Babylon a well known British online subscription service that uses bots to offer consultation based on personal medical history, and can even connect you with a live video consultation from a doctor.
Babylon’s interactive symptom checker asks the patients questions and analyses their conditions. Acording to the company website, “Babylon's AI system has been created by experienced doctors and scientists using the latest advances in deep-learning. Much more than a searchable database, it assesses known symptoms and risk factors to provide informed, up-to-date medical information.”
5.2 Travel Booking – One of the more successful chatbots used by Expedia.com to helps its customers make hotel reservations, car rentals, cruises, and even vacation packages via their website or Facebook page. To get relevant offers, travelers need to provide the bot with their requirements such as destination, date, type of accommodation, price range, and so on.
The sequence of screenshots, illustrate Expedia’s chatbot. First the Expedia bot presents the visitor with a conversation start offer “Start Exploring”. Once the visitor clicks on the start exploring, the chatbot presents various vacation themes like “Breathtaking Island Escapes” followed the bot presents specific island and when the visitor clicks on one of the island, the chatbot displays hotel packages specific to the selected island. This chatbot keeps the visitor engaged and when the visitor selects a specific hotel, the chatbot redirects the visitor to Expedia website where hotel booking can be made.
5.3 Placing Orders – Chatbots are being used in the food and retail industry to place orders online. Chatbots gather the necessary information from the customer such as their postal code to determine the closest location and recommend products that fulfill the buyer’s requirements in the best possible manner. They also provide real-time support by answering all the queries the user may have during this buying journey.
For instance, Pizza Hut deployed a chatbot that allows users to place orders for pizza and other products from their Twitter and Facebook accounts. Users can reorder saved or older orders and also access information on Pizza Hut’s latest promotions. Likewise, Taco Bell’s TacoBot allows you to place an order through the instant messaging service, Slack.
5.4 KLM Blue Bot (BB) – KLM airline turned to AI based chatbot to help automate its customer service front line. And according to KLM “BB is not just another smart assistant. She's a self-learning system (or: Artificial Intelligence) BB has her own professional, helpful and friendly character, but be warned; she can also be a bit cheeky from time to time. Aside from that, she's always eager to learn more in order to be of better service to you.
According to Martine van der Leee, the AI solution has an accuracy growth rate of about 2.2% per week. “We’ve seen a 55% accuracy improvement since January (2018),” she said. “To us, this is a strong signal that AI is working and that it is learning over time.”
6.0 How and where to start?
If your business is considering using chatbot solution, review section 4.0 of this tutorial and select a business function where an automated robot like chatbot could perform or supplement the existing function.
Armed with your selection and goals, it’s time to search for a vendor among the 100’s of vendors on the market. This is where it becomes challenging. Instead of searching the Internet, visiting and reading vendor websites or talking to their persistent sales people, we came up with a better way. We created a portal (www.180find.com) where you can describe your business needs or interest, publish it and let the vendors come to you with their best proposals. And do that while you stay anonymous and communicate with vendors directly. Our portal and service to businesses is free. Once you select the best solution, you can go off line and contact your selected vendor directly and complete your transaction.
Not all vendor solutions are the same. And not all chatbot technologies are the same. Some offer more advanced tools to build a chatbot and some even offer generic data. Selecting the right vendor can be overwhelming. To help understand the key differences in the solutions that vendors offer, here are some key questions a business considering chatbot should ask vendors.
Rule-based chatbot solutions:
1. Is there a minimum model (volume of calls, number of interactions, complexity etc.) where chatbots should be considered?
2. What specific improvements has your solution delivered to other customers?
3. What industry is your solution most commonly used?
4. What is a typical ROI for a chatbot solution deployment?
5. Are there reusable chatbot functions your solution delivers or is everything custom configured for our business?
6. What off the shelf function does your solution comes with?
7. What level of accuracy (answering requests correctly) does your solution deliver post deployment?
8. Is there a need to convert our training data to a specific format to be used with your solution?
9. If we don’t have the expected training data format, what options are available to us to use our data?
10. What expertise do we require to define and configure conversation scripts in your solution?
11. What assistance and experience does your company have deploying your solutions in similar companies to ours?
Deployment and Post-deployment
12. How long do you anticipate that it will take us to configure your chatbot solution for our business?
13. How will our training data be used to train your chatbot solution?
14. What integration capabilities does your solution offer with our systems (name your systems e.g. website, Twitter, Facebook if you use them)?
15. How do we monitor the performance of the chatbot?
16. What expertise do we need in our company to monitor and fine tune your chatbot solution?
17. How much human intervention is typically required and when after your chatbot is deployed?
18. If we want to expand the initial chatbot solution to include other functions (name them if you know) with the deployed chatbot?
19. What advance capabilities do your solution offer to enable your chatbot solution to learn and to improve itself?