closeup of robot handshake

C hatbots were predicted to revolutionise customer service across the internet, replacing the need for human customer service agents and to provide 24×7 accessibility that most companies could not afford.

In 2016, the Business Insider predicted that chatbots would reach adoption rates of 80% by 2020. In reality however, early chatbots were plagued with basic errors in their natural language processing (NLP) algorithms; they simply could not understand what humans were asking of them. The problem continues to exist to this day, with missed intents (“I don’t understand”) and false positives (“I think I understand”, but resulting in a wrong answer) typically making up 60-70% of open text responses today.

Despite recent advancements in the NLP, many chatbots and many other businesses do not use them at all due to personal experiences with chatbots.

This then begs the question: what is the current state of the chatbot market in 2021? More importantly, can chatbots deliver on their basic premise and deliver solid results in 2021 and beyond?

What is the state of the market? 

Many businesses started on their chatbot journey with the expectation that chatbots would not only revolutionise customer experience with their “always on” feature, but would also deliver significant cost savings to their business. The experience of companies that initially deployed chatbots over the last several years has been less than stellar however, leading to a slow-down of their chatbot development schedules. In some cases, companies have decommissioned their chatbots due to poor customer feedback. Many other businesses have actually put off plans for chatbots due to the experience of others. Recent research by bot performance experts, INTNT.AI, for example, suggest only 6.75% of the UK’s top insurance companies have adopted a chatbot despite the progresses made in recent years in the field of chatbot development.  This might be because a crucial element, the on-going training of the chatbot to suit the companies’ use case(s) are often neglected, leading to relatively poor results in customer engagement through such chatbots.

two toy robots on orange background

If chatbot technology has progressed, why are adoption rates so low?

The chatbot market has not grown as fast as early estimates expected. Gartner forecasted in 2017 that by 2020 the average person would have more conversations with a bot than with their spouse. TechCrunch even expected major websites to be shut down in favour of new conversational customer interactions through chatbots. 

Despite early growth forecasts being off the mark, the chatbot market is still robust. It was valued at $1.17 billion in 2018, is expected to reach $10.8 billion in 2026. Chatbots are expected to save the healthcare, banking, and retail sectors $11 billion annually by 2022. This expansion is being driven by large players in the natural language processing (NLP) space, including Microsoft, IBM Watson, Amazon, Facebook and Google, among others. With better NLPs, chatbots become more functional, more intelligent in being able to understand human communications. As use cases increase, so does the demand for more intelligent bots, bots that can be adapted and adopted for specific company use cases, thus the new moniker AI Bots.

What kinds of bots exist in the market today?

Currently, there are two distinct forms of chatbot: scripted and free-form AI bots. Scripted bots rely on the user selecting pre-determined questions that the bot has scripted answers to. This limits the functionality of the chatbot, and limits the customer service experience that AI bots can potentially offer, but serves basic queries well.

Free-form AI bots on the other hand, can respond to user-constructed queries and offer a more human-like interaction, but can be subjected to misunderstandings of the AI bot to free-form queries. A hybrid AI bot combines the best of both worlds, providing both a simple Q&A format when required, but having the flexibility to provide answers to any ad-hoc questions that the customer might have.

ai robot balancing boxes

What is the biggest problem with chatbots? 

Part of the reason why scripted chatbots are still useful is that free-form open text chatbots are often prone to mistaking the intent behind a customer’s query, yielding either no or wrong answers to customers’ questions.

Most NLPs are unable to correctly identify the intent and the object of a customer’s question in a consistent manner, and this often leads to end-user frustration. This is because human beings don’t always communicate consistently or enter their questions in a structured manner.

As Jude Tan, Chief Commercial Officer of INTNT.AI explains,

“The number of missed intents and false positives in an untrained bot typically hovers around 60-70%, and this can create frustration for the customer and limit the usability and cost-saving potential of chatbots if they end up having to transfer to a human operator when challenged beyond their understanding.”

How artificial intelligence is re-defining the bot market

Bots can’t train themselves quickly enough through AIML alone, especially in the case where the bot gets the question wrong and generates false positives (or wrong answers), which can cause further downstream issues (as bots “learn” bad answers).  Bots can theoretically be trained manually, by having customer service agents comb through 10s of thousands of questions in the bot’s chatlogs and validating the answers given by the bot – either confirming the answer or correcting the bot by giving the correct answer or creating a new answer. But this is a tedious and costly affair for companies who want to save cost.

Bot training companies like Intnt.ai have taken a different approach to this problem.  Using a blend of AIML and linguistic programming, Intnt.ai has been able to automate the bot training process to deliver a training regime on an automated basis, identifying missed intents and false positives on a 100% sampling basis – effectively doing in under an hour what would take 27 man-days to do on a manual basis (based on a chatbot with 40,000 questions) per month.

As Tan explains, “Most companies tend to eschew training their bots due to the immense amount of manpower required. Some take a sampling approach to the training, which is akin to trying to stop a flood with a bucket. Chatbots need to be trained on a continuous basis on a 100% sampling regime to continue to be relevant to their users”

This automated training approach ensures that the bot achieves a consistently high degree of accuracy, the overall customer experience but means that the bot is significantly less likely to need to divert the query to a human agent, allowing significant savings in customer service costs. In one study, it was shown that bot training reduced customer requests for transfers to human agents from 21% to 6% by maintaining bot response accuracy rates at 90% and above (from an original 35%).

robot and human hands high five

A new dawn?

While current perceptions of chatbots will be difficult for the industry to overcome, AI chatbots with proper training look set to change the landscape. Apart from the obvious time and financial savings that efficient and effective chatbots provide, they are also essential as the first point of contact to customers and in starting the conversational journey between customers and businesses. Customer growth and sales naturally follow.

There is also the potential of conversational AI bots to engage clients over social media, which is predicted to be a major area of growth for the chatbot market over the next few years. In this space, Juniper Research has predicted that chatbots will handle $112 billion dollars of eCommerce transactions through social media interactions by 2023. Again, the effectiveness of such bots rely heavily on continuous training, to keep the AI bot relevant to changes in product offerings and customers’ demands.  Towards this end, Intnt.ai has developed intelligent conversational AI bots across various social media platforms to serve this need.

Future outlook

Artificial Intelligence led solutions are leading the charge in chatbot development, at a time where the market is primed to explode. Training automation engines have the potential to power customer conversations, leading to better experience whilst reducing the need for human interaction through calls to call centres and live chats. AI-powered bots provide the 24/7 digital platform needed to engage customers in this new digital world–using social media to handle new conversational journeys and to drive sales.