Every technology that didn’t work well once didn’t do so not for the lack of innovation but because it failed to reduce friction without creating more. Although deemed the year of chatbots, 2016 saw quite a few chatbot failures in the form of delayed response, needing additional executive interventions, low or negative ROIs, and much more.
Since 2016, chatbots have gained momentum, however, quietly. With the assimilation of improvements seen in the fields of Natural Language Processing (NLP) and Artificial Intelligence (AI), chatbots have changed the face of conversational technology in just the last few years.
So what really is the hype behind chatbots? Why do chatbots fail? How can chatbots be successful? What makes them successful? And are chatbots really the future?
The Hype, The Downfall, The Growth:
When Facebook brought the messenger chatbot platform out into the world, the great benefits from chatbots held high expectations of increased sales, conversion rates, engagement rates, and more. With increased mentions and press releases everywhere, the hope spoke loud that Chatbots could be the next big disruptive technology. Facebook launched its own personal assistant, M, that created a great hype which eventually died down as human executives had to take over about 70% of the questions. The attempt with M was to automate all of its activities, however, the automation ability of M could only be taken up as high as 30%.
In addition to this, chatbots tended to make simple tasks like being able to find a recipe or finding some instruction a much longer and sometimes difficult process than it earlier was adding frustration into the chatbot experience. Most chatbots, having started off in specific industries had no data at this point that could be processed in high quantity to understand the flow of human language. With the lack of technology to support understanding context and complex queries, the chatbot hype slowly started to fade to the point where any talk of chatbots reached a deafening silence.
However, with the advances in Natural Language Processing (NLP) and Artificial Intelligence (AI) technologies, several industries have harnessed the lessons from these very failures in the past and improved them over time leading way to great Chatbot success story examples. Right from customer service bots to greatly-abled virtual assistants, Chatbot adoption is seeing an increased growth from 2016, today. Chatbots have quietly seeped into the technological normalcy of the day and made their space deemed essential in the last few years.
So what were the mistakes that wrote doom in the fate of Chatbot’s future and what helped them rise through?
- Specific vs. Vague Use-case Structure:
- Intent Overload:
- Limited Training:
- Lack of efficient chatbot design and constant improvement:
Failure: One of the major issues that led to increasing frustration in the chatbot experience was less capable automated workflows which tend to fail often and lead the user into loops of repetitive, useless jargon. This problem specifically occurs when you plan for a wide array of vaguely defined case structures built to accommodate everything, only to end up with a blurred clarity on what end goal is the ultimate goal for the user and the businesses.
Working with a diverging focus with the aim of reducing human intervention ultimately only creates the need for additional intervention added into the process causing resource waste. In addition to this, a combination of over-hyped expectations and insufficient NLP ability added to the mix of a vague use case structure leads to failure in answering even simple questions leading to extremely unsatisfactory outcomes.
Improvement: The best way to grow a chatbot would always be to start with a Minimum Viable Product (MVP), as with any other product launch. Defining a clear set of goals the chatbot should achieve and a defined ideal path of user journeys it should take to reach the said goals is required. Where productive conversations are the need of the day, defining niche use cases to respond to questions that your brand promises to answer gives your chatbot the edge and a strong definition of abilities. The strategic development of use cases with the knowledge of the contextual approach you’ve learned from your customers can take you on the road to success. With clearly defined use cases, automating simple tasks for your user’s convenience becomes easier.
Once the use cases are thoroughly developed, some energy will need to be focused on setting the right expectations. Losing the battle of brand promise vs. user expectation will result in decreased retention rates. Many chatbot customers who have had a bad first experience tend to never rely on the platform again. Users appreciate a guided, personalized experience. The extent through which the guidance and experience can be extended is something for which the businesses should define smart boundaries. Set the right expectations for your users, exceed them, and find space to grow in learning that chatbot development always has been and always will have iterative processes.
An intent is a motive that drives the user’s query – the user’s intention. Intents are assigned specific names in the backend. Each intent can be phrased in multiple ways and each phrase would be called an utterance. For example, a customer service chatbot may help a user with configuring their printer. The intent could be assigned the name ‘printerconfiguration’. The phrases “Can you help me configure my printer?”, “Printer configuration”, “How to configure my printer?” would all be utterances.
Failure: Speaking of the Minimum Viable Product in terms of a chatbot and niche use-cases, one of the major pitfalls that were seen in the earlier years was when niche businesses aimed to incorporate so many intents into the chatbot without considering the best strategy for elevating the customer experience and streamlining customer journey. This led to an incomplete response to common customer queries. The incorporation of each intent involved varying levels of complexity.
The attempt to overload intents to increase functionality is often met with decreased conversational experience. For one, it tends to take away the necessary focus that should be placed on responding to the simple questions, the actual use cases that require thorough automation. While there is no strict constriction on the number of intents a chatbot can handle, the current NLP systems may be limited to handling several 100s of intents.
Improvement: A bot that can answer any complex question that the user may pose while may seem to be the pinnacle of your chatbot’s success, forms a poor starting point. More often than not, the addition of a large number of new utterances and intents may not seem like the breaking point as it’s not possible to particularly notice the breaking till a performance decline is observed.
A strategical approach based on the list of common questions, the context of conversations in your industry and business, maybe some helpful tools in finding opportunities and ideas to structure productivity in the user journey.
Failure: Chatbot development is an interactive process that constantly needs to keep improving itself. Successful chatbots are always in development in the background as there is not one-time hit possibility with chatbots. For successful interactions with chatbots, while the initial plan required niche or industry-specific use cases, the problem arises in training the chatbot to respond to these questions amidst the clear lack of data for these scenarios. With the limited data available to train the chatbots, it’s possible more often than not for the chatbots to reach stagnancy without additional resources. Even the current approach to conversational AI may only be capable of taking a chatbot so far in understanding the complexity of human conversations.
Improvement: With the current state of NLP technology, one may not achieve the complete automated state of effortless chatbot conversations. Surpassing the barriers of limited training for new businesses may also seem daunting. While these are barriers that current technologies are trying to break through, they can be overcome if not entirely avoided.
Given that chatbot technologies have not still reached the point of completely utilizing the knowledge base provided to them, strategically placing escalation points and providing clear routes to the customer to access these escalation points can be one of the biggest saviors for chatbots. Chatbot developers can also include checkpoints and follow-up questions to make the most out of customer feedback that can eventually help in guiding the knowledge management processes. With thorough monitoring of conversations and context from the users, repetitive tasks that require escalation can be automated leaving the complex tasks for executives to handle.
Failure: One of the major downfalls created by the increased hype around chatbots in 2016 was that businesses expected quick results from chatbots. This expectation required accelerated chatbot development that missed out on so much essence of how the conversational experience is really built. With business processes trying to fit their ideas into off-the-shelf chatbots, potential customers ended up finding a weak, low-quality experience devoid of personality and came nowhere close to human interaction.
Improvement: Chatbot developers understand that the process of chatbot development continues beyond successful deployment. As much as efficient chatbot architecture and design are required for the chatbot to make any semblance of sense, how it improves over time is a result of careful understanding and monitoring of reports and analytics available through the chatbot’s conversations.
One of the issues in this area could be based on the specific industries and their differing needs in examining various aspects of the chatbot conversations for further improvement. Customized chatbot architecture enables one to report the essential details required to finetune the conversations and improve the knowledge base to answer seemingly complex repetitive questions, unlike the off-the-shelf chatbots that can be very rigid in the data monitored and filtered through for improvement.
There have been dramatically bad chatbot failures and there have been brilliant chatbot success stories, each of which is pushing the ceiling as to how far the advancing NLP and AI technologies can reduce the load on redundant human tasks and make the most out of the resources we have. The wide variety of industries, stories, channels, rules, and requirements in this process may make it look like an endless labyrinth of complex layers incapable of untangling.
However, all you really need is someone that can help you navigate these structures better so that you can create a tunnel-focus around your vision and efficiently deliver it as a productive tool to your audience. Right Angle Solutions provides exactly that. Through a multi-domain, multi-channel, intelligent approach intertwined with AI and NLP capabilities, we build custom-reporting chatbots that not only help you achieve your goals but keep growing through them. CONTACT US to understand how a sustainable, scalable chatbot version would look like for your business.