Studies show that over 50% of customers expect a business to be available 24/7. Customer service call centers can be expensive, especially when factoring in the number of employees needed to manage thousands of customers for 24 hours a day. That’s where chatbots come in. 40% of chats are for routine questions such as, "Where's my account number?" and "I want to return my socks." A chatbot that is able to handle those simple requests saves time and money, while still building a connection with that customer through direct interaction.
Primitive chatbots are keyword-centric. Also, most vendors that have a packaged chatbot application are not enterprise-ready and these bots do not integrate seamlessly with legacy enterprise systems as required to retrieve customer information or gracefully fall back to a traditional support line for human-assisted support.
We built a smart chatbot in less than two weeks that processes thousands of inquiries each day. We did so with our rapid prototyping platform Interplay, where we are able to quickly snap together an AI engine, a training data set, ties to the order management and inventory management systems, and the CSR queue.
Smart Chatbots can easily be built via our platform by connecting together the required elements. A smart chatbot needs good Natural Language Processing (NLP) software, as well as a large number of recognized issues and correct responses (sometimes called “intents”). If the chatbot is programmed to respond to a wide array of different scenarios, it will perform much better than if it only recognizes five or six statements. A good bot will know up to dozens if not hundreds of possible intents and be able to adequately respond or to otherwise gracefully pass things over to a human for an answer. The chatbot’s responses need to be conversational and standard grammar. The responses also need to pull from real-time connections to the order management system, the inventory tracking system, and the customer support queue. Think of the earlier “I want to return my socks”. The chatbot needs to instantly look up who “my” is to get the customer name, “socks” from the inventory system to get the SKU from the last order in the order management system. At any given point, the customer might just type “human please” so we need to be able to kick this over to the customer support queue with a complete chat record and any supporting inventory/order/customer profile information available (in order to avoid the dreaded “Hi can you please tell me the issue?” question form the CSR that forces the customer to start over from square one. That’s a lot of connections, and that’s the difference between a truly enterprise-grade chatbot and a simple conversation-bot you wrote in CS300 class at university.
The bonus here is that a smart chatbot that gives correct answers to a customer will earn trust very quickly, which helps overcome the second risk of people hating on robots. There’s a real opportunity here for organizations to differentiate themselves between the many companies with dumb chatbots or no chatbots, and the smart chatbots who answer questions immediately (average response time is usually under half a second).
Try it out: https://www.ulta.com/guestservices/
A good AI engine is the first step for building a smart chatbot. This AI engine needs to have natural language processing, a large trained data set, and an iterative process for improvement. If a chatbot gives an incorrect answer or inadequate reply, it needs to have the potential to learn from it and not make the same mistake again.
After the AI engine is in place, it needs to be trained and tuned to the specific intents and recognize entities that are present in the dialog. A database of previous questions and responses is needed in order to train the chatbot about potential questions it may encounter and the proper way to reply. An expansive database can help by teaching the chatbot as much as possible, giving it a strong foundation. The chatbot then needs to encounter actual data, where people are reaching out to it and making requests. If it makes a mistake during this phase, it can then be trained to make the proper response next time. Through this system, the chatbot is able to learn and refine itself.
Interplay is a low-code middleware platform. Prototyping is rapid, iterative, and flexible. Interplay comes with modules for several popular smart AI engines already in place. To deploy an AI engine for your chatbot, simply connect the AI engine, your training data set (a list of words, phrases, etc), and your desired output method (e.g. an iframe, https, text file, etc). Creating this ‘flow’ is a matter of a few hours, compared to days or weeks in a traditional coding environment. You can then spend your time configuring, refining, and training your chatbot. Launch when ready.