Chatbots and Conversational Commerce are major buzzwords in the e-Commerce industry right now. An all-new approach to web-based commerce is attractive to retailers looking to increase user numbers and sales. As we explained in the previous installment in this three-part series, Conversational Commerce does hold the potential to let your customers connect more deeply with your site and brand, thanks to the elusive personal touch that many feel is missing on a bog standard commerce site.
Now let’s say you’ve decided to pursue a Conversational Commerce approach and want to integrate a chatbot in your site or use one of the messenger-based solutions that are readily available. There are some things you have to keep in mind when you kick off such a project, especially when it comes to chatbots built into a website, which will be our focus here.
First off, there is the timeframe. Many e-Commerce managers get carried away with the possibilities of Conversational Commerce and expect it to be a one-day job to implement. Let us assure you that it isn’t.
What should my chatbot do?
The process must start long before any development or bot training commences. The main question you need to ask yourself at this stage is: What is my chatbot’s purpose?
Simply implementing a chatbot in your website because everyone seems to be doing it is not only a waste of your money, it could even hurt your sales. A poorly thought-out bot that delivers nothing of value is only an added barrier between your customer and a sale. It is like having a motivated sales clerk greet your customer at the door of your physical store only to then direct them to a storage closet. Not only will that make it harder for you to achieve a sale, it also makes it unlikely that this particular customer would ever return to your site.
A chatbot’s purpose can vary, but it usually hovers around the areas of sales support and technical support. The best chatbots do both equally well, but this involves a lot of careful training before launch.
A major part of defining the chatbot’s purpose is setting goals and measuring the progress toward those goals. Very likely, your target would not be simply amassing a large number of users for your chatbot. Ideally, those users would then also purchase a product and return to your site at a later date. Conversion and retention should be two of your main targets. After the launch, you will be able to monitor exactly how your bot is performing and whether it is meeting those goals. We will have more on this post-launch analysis in the third and final part of this series.
The most important part of the chatbot process: the training.
The best approach is to start with actual questions asked by your customers and users that were collected prior to training. Using real questions is much more likely to make your project successful than trying to come up with questions you think people might ask. Chances are that guesswork will not hold up in reality.
This valuable data must then be fed to your chatbot, within the framework of the bot technology you are using. Questions are given answers and your chatbot designers will have to establish variations in wording that could lead to those same answers. Make sure to account for typos, regional spellings and other potential causes for misunderstanding.
The process differs quite dramatically between scripted or rule-based bots and true AI bots. Scripted bots can only answer predetermined questions in certain ways, unable to learn by themselves. They rely solely on the data and rules provided by chatbot trainers, which means they can only simulate a true conversation. They do not analyze and try to understand a user’s question as much as they check whether it fits within the criteria of questions established beforehand by the chatbot trainers. If a user veers from the beaten path, it’s unlikely that you can cover all bases and answer an unorthodox question in a satisfactory way.
A truly conversational experience can only be delivered by bots that utilize machine learning. They are able to learn autonomously from previous chats and conversations, which means they improve their performance with every instance of use.
Retrieval models, which can be likened to scripted bots that also use machine learning, can improve the accuracy of your chatbot dramatically. Generative models, such as IBM Watson on the other hand, learn using Natural Language Processing and Natural Language Understanding, which makes them smarter and able to understand nuance and context to provide a more life-like experience. However, they can sometimes disappoint at the start of implementation due to taking queries too literally. They need time to actually learn.
Taking the plunge
So now that you’ve established what you want to do and have fed your chatbot with the data it needs to achieve this goal, it’s time for launch – and a host of more tasks. We will take a look at post-launch challenges in the third and final part of this series.
Looking to build a chatbots? You may want to talk to our IBM team.