Why Chatbots Fail: Limitations of Chatbots
This article is an excerpt of “Chatbots: The Definitive Guide 2020”.
In this article we’ll cover the reasons chatbots fail and what to avoid when building your conversational AI chatbot strategy.
Chatbots Failing to Deliver
It’s claimed that chatbots increase customer engagement, improve the brand experience and deliver actionable insight to the business. So why are so many chatbots failing to deliver on their potential?
The answer lies in the restrictive nature of most chatbot technology. Few chatbots offer the rich, humanlike conversation needed to engage users, nor can they guide off-topic users back to the subject at hand. They can’t ask qualifying questions if clarification is required. And, they are not able to deliver over the different channels and languages by which customers want to communicate.
Add in a lack of intelligent interaction by the chatbot and confusion over data ownership and it’s no wonder Gartner expects that 40% of first-generation chatbot/virtual assistant applications launched in 2018 will have been abandoned by 2020.
The main issues can be categorized into four main areas:
A Lack of Training Data
It’s a common misconception that machine learning systems somehow work completely on their own, without any human supervision. This is not true.
Just as a linguistic based conversational system requires humans to laboriously craft each rule and response, a machine learning system requires humans to collect, select, and clean every single piece of training data, because using machine learning to understand humans takes a staggering amount of information. What comes naturally to us as humans — the relationships between words, phrases, sentences, synonyms, lexical entities, concepts etc. — must all 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.
For enterprises that don’t have a significant amount of relevant and categorized data readily available, this can be a prohibitively costly and time-consuming part of building conversational AI chatbot applications.
Build better voice apps. Get more articles & interviews from voice technology experts at voicetechpodcast.com
Poor Conversational Understanding
An even greater problem is the risk that the machine learning systems do not understand the customer’s questions or behavior.
In a linguistic based conversational system, humans can ensure that questions with the same meaning receive the same answer. A machine learning system might well fail to correctly recognize similar questions phrased in different ways, even within the same conversation.
There’s also the issue that pure machine learning systems have no consistent personality, because the dialogue answers are all amalgamated text fragments from different sources. From a business point of view, this misses the opportunity to position the company and its values through a consistent brand personality.
Ease of Creating Global Appeal
Organizations need to support their customers in different languages — a problem that will only increase over time. Hence, chatbots need to be fluent in many languages, with the ability to learn more when needed. But this is only part of the problem, because they frequently need to support a variety of platforms, devices or services too.
Most chatbot development technology requires a great deal of effort and often complete rebuilds for each new language and channel that needs to be supported, leading to multiple disparate, solutions all clumsily co-existing.
These solutions cannot reuse assets from the original build, nor can they surface the same solution through multiple devices and services.
Regulations Protecting Data
Data is at the heart of conversational AI, and is used to personalize the conversation, improve the system and deliver actionable insight to the business, so it’s essential that enterprises can reap the benefits while complying with regulation and legislation.
While GDPR is an EU regulation, the ramifications impact enterprises around the globe. It’s likely that regulation will increase throughout many countries in the future. For organizations, the challenge is not just in storing the data, but also in retrieving the information for export or deleting in a secure and auditable way.
Furthermore, many chatbot technologies restrict access to the conversational data generated, meaning businesses lose one of the key benefits to implementing a chatbot. Without this data, businesses are effectively blind to their customers.
Originally published on the Artificial Solutions website on November 27th.
