Exploring The World of Chatbots

What if I told you the 1st time people thought of machines replicating human conversations was almost 70 years ago!

And that’s when the Turing Test was developed. The Turing Test was the 1st of its kind, it asks the question “Can Machines Think?”. Alan Turing proposed The Imitation Game. Here an interrogator will talk to a person, X, and a machine, Y, through a series of written questions and determine if X is a person or a machine. 

Even though the test has its flaws and saw a lot of criticism, the test is still in use today. Since then, this fascination of getting machines to answer our questions has only increased and today one facet of it is called “Chatbots”.

Recently, some updates were made to the International Data Corporation (IDC) Worldwide Semiannual Cognitive Artificial Intelligence Systems Spending Guide, which suggested that spending on cognitive and AI systems will reach $77.6 billion in 2022, that’s almost 3X the $24.0 billion forecast for 2018.

The largest and fastest-growing technology category goes to “Software”,  in which around 40% of all cognitive/AI spending with a five-year CAGR of 43.1%. Cognitive/AI spending will be focussed on 2 areas:

There are many advantages of implementing chatbots in your businesses. Some of these advantages are pretty evident like live chat, reduced costs, but let’s look at several other advantages of offering an intelligent automated self-service option. The way chatbots have stepped up and are helping businesses during COVID-19 to cope in terms of customer service and information sharing is amazing.

Full-Time Availability: Sometimes, you just want to speak to a customer service representative, but what if you need help at odd times, or at peak times when phone lines are jammed? In times like those, chatbots are available for your customer. 24*7 on all the channels your business needs.

Quick Response: Chatbots can recognize, understand and respond to any query or problem within seconds. They can help you find the ‘best match’ query even while you are interacting with a customer saving a lot of time and effort.

Here’s a report by Gartner that shows how a chatbot was able to answer questions within 5 seconds, while the average advisor took 51 seconds.

98% Accuracy: While being quick, it’s also important to be accurate. Accuracy is key to increase first-time call resolution rates and to ensure customers are happy talking to the chatbot whenever they have an issue. Advanced chatbots can solve 80% of the queries on their own with about 98% accuracy.

Compliant: Chatbots aren’t humans. They don’t make mistakes, even silly ones. Conversation tracking is easy, updates and changes can be rolled-out and everything audit-related can be maintained for compliance proposes where needed.

Scalable: One agent can only manage one customer, but one chatbot can handle millions of conversations simultaneously, all to the same high standard.

So we know that chatbots increase customer engagement, improve the brand experience, and deliver actionable insight. But still many chatbots have no idea what they are doing. Why?

It’s not the chatbot, it’s how you build it. We have found that chatbots fail for 4 major reasons:

When we are trying to learn a new language, we need to learn each rule and response, similarly, the machine learning system requires us to collect, select, and clean every single piece of training data. Things like relationships between words, phrases, sentences, synonyms, lexical entities, concepts, etc. come naturally to us, but they need to be ‘learned’ by a machine.

In a recent survey, 81% of respondents said that the process of training AI with data was more difficult than they expected. Well, nobody said it was going to be easy ;)

Not understanding the customer’s questions or behavior can be a big issue.

Say for eg, the same questions phrased differently can be easy for humans to understand, but are not that easy for chatbots. Again, the training set needs to include all such instances.

We live in a connected world. Companies have customers worldwide, so they need to offer support to their customers in different languages. This means chatbots have to be fluent in different languages, with the ability to learn more languages if needed. But that’s not it, the number of platforms, devices, and services is also varied, and chatbots need to support all of them too. 

Most chatbot development technologies require a lot of effort and more likely than we would like to accept complete rebuilds for each new language and channel that needs to be supported.  This leads to multiple disparate, solutions all trying to co-exist. This is not good for anyone.

Data is at the heart of any conversational AI system. Data can be used to personalize a conversation, improve the system, and deliver actionable insight. So, it’s important to leverage all these benefits of data while also complying with regulations.

GDPR in the EU is just the beginning of regulations in data, which are likely to increase in the future. Organizations need to figure out how to deal with storing the data, and also retrieving it or deleting it in a secure and auditable way.

One big problem is that many chatbot technologies restrict access to the conversational data generated between chatbot and customers in an attempt to comply with such regulations. This is a huge loss for businesses.

So, now we know how and why chatbots are failing. What can we do to make sure that chatbots actually deliver? 

AI chatbots actually do deliver the intelligent, humanlike experience that you are looking for. The majority of chatbots in the market today are obviously not based on AI. They do use some algorithms to understand your question and may come up with the correct answer, but if you go off the script, it’s gonna behave weird.

AI Chatbots or conversational AI systems can understand what a customer is saying, irrespective of how the question is phrased. They can also help with a lot more like filling forms, making recommendations, upselling, booking appointments, or even integrate with third-party or backend software like Robotic Process Automation (RPA), Enterprise Resource Planning (ERP) or Customer Relationship Management (CRM) systems to carry out further tasks.

The key to successful engagement is understanding the customer’s request and delivering a response that’s personalized and relevant to the individual.

In order to be successful in engaging with the customer’s request all you really need is to provide a personalized and relevant answer. For that you need to enable your AI chatbot with some capabilities:

This is more than just correctly interpreting the user’s request. It’s more about gathering information like geolocation or previous preferences and uses this information while delivering the answer.

Having some memory allows a chatbot to remember any important details that can be reused during a conversation or store anything they learn about a person. 

Imagine, you want to order dinner, you use the chatbot of some application, based on your previous orders, the chatbot can recommend similar cuisine, similar restaurants. Isn’t that amazing?

Sentiment analysis is extremely important. It enables a chatbot to understand how the customer is feeling, and what’s their mood like. This will in turn help them analyze the escalation levels and urgency of the situation. 

When your chatbot has its own personality, it allows customers to trust the chatbot and increases engagement rates. You can do this by using avatars or just in the way the chatbot talks.

This enables the user to veer off onto another subject, such as asking about payment methods while enquiring if a product is in stock. The chatbot should also then be capable of bringing the user back on track if the primary intent is not reached.

We know that dealing with all these patterns can be a little difficult. Building an AI chatbot requires much more effort than just supporting a linear conversation. But if you want to give the feeling of a real conversation, you gotta do what you gotta do.