Point Break – how to ride the chatbot wave without hitting the wall

By Cathal McGloin

Cathal McGloin, CEO of ServisBOT discusses how industries can ride chatbot innovation wave.

"Chatbot deployments continue to grow as businesses realize their value for creating seamless user experiences and improving business results.  While plenty of chatbot success stories are being reported, there is a breakpoint phenomenon that is coming to light.

Chatbot Lingo:

When training a chatbot an ‘utterance’ is anything that a user asks. A chatbot ‘intent’ is the user’s reason for their question. For example, if a user asks, “What will it cost to add my 20 year old son as a named driver?” The ‘utterance’ is the whole sentence, while the ‘intent’ is to discover the cost of various insurance policies. When training a chatbot, ‘intents’ are given a name such as ‘ListPremiums’, ‘ShowPrices’, ‘ShowDiscounts’. ‘Entities’ modify intents. For example, “Will my car insurance go up next year?” might have the intent of ‘ListPremiums’ and the entity of ‘TimeDate’, ‘PolicyType’.

As companies expand the capabilities of their bots, adding additional intents and utterances, there is a potential for the bot to reach breakpoint.

Four Stumbling Blocks for Business Chatbots

For the team at ServisBOT it was important to get to the bottom of what contributes to successful user experiences with chatbots. In doing so, we also discovered why a chatbot might work really well up to a point and then suddenly start to deteriorate or fail. We identified four underlying problems.

Too much information?

When a company architects their bot solution, the issue of intents and utterances and how many a single chatbot can handle is often overlooked.

A chatbot may work perfectly well at the outset, but as intents and utterances get added over time, this can eventually impact the user experience.  Unfortunately, it’s impossible to define the exact breakpoint at which the bot performance starts to decline as it varies depending on the use case. Handling up to 100 intents and more, a single chatbot can come close to the limit that existing natural language processing (NLP) solutions can support.

Accuracy Drives Increased Intents

Another way in which a bot can be impacted is when the accuracy level needs to be raised. This may mean adding more variants of the intents. However, the increase in intents may reduce the capacity for other feature capabilities. Since language is fluid and has many nuances, aiming for accuracy can consume considerable bot capacity.

The Issue of Continuous Learning

A chatbot, once launched, cannot it be left to its own devices. It needs to be maintained as user interactions with it grow. As users interact with it new utterances may emerge that are not recognized by the bot. The bot needs to learn these missed utterances so that they can respond to them in future.

Building a chatbot is not like building an application. At any stage, a chatbot is tuned for a certain amount of data. Any changes can potentially disrupt the whole model. This a continuous task.

When a Single Bot isn’t Enough

When it comes to the different use cases that a business has for chatbots there are varying degrees of complexity. As companies become more sophisticated in their chatbot deployments, some are coming to the realization that one chatbot may not be sufficient to handle a single use case successfully.


Consider the example of a company building a chatbot to handle simple customer FAQs. Over time, they may decide to expand the bot’s capability, such as adding more personalization, managing different languages, adding small talk, handling context switching, or a combination of these. These additional capabilities can potentially erode the capacity that the bot has to deal with the original use case i.e. answering FAQs.

This breakpoint phenomenon has implications for how a business architects its chatbot solution to meet the requirements of the business use case.

Introducing a Multi-bot Architecture

An increasing number of businesses may experience the bot breakpoint in the next year. The good news is that there is a way to tackle this by adopting a multi-bot architecture with a central virtual assistant that coordinates and orchestrates multiple task-oriented bots. This approach helps overcome intent overload and subsequent deterioration of the chatbot experience.

While a single bot model works for many initial and less complex use cases, the way to think about a bot architecture is to think of bots as having different skills. Stringing together multiple bots will meet the needs of the use case without hitting the breakpoint of NLP limitations.

To address the breakpoint issue, we structure bots in a similar way to employees in an organization. Each person has a specific role and set of skills that they bring to the business. Similarly, each bot should be designed with specific tasks in mind. Just like a human, there is only so much a bot can learn, so it makes sense to train them for the skills that are specific to the business use case.

In a multi-bot architecture, individual skills can be managed using an enterprise virtual assistant, acting as a central brain that orchestrates and coordinates multiple bots that work towards fulfilling the business use case.

The virtual assistant blends independently managed bots into a unified experience, routing to the bot best skilled to respond to user requests.  It also monitors the ebb and flow of conversations and enables all bots to support language detection, translation, sentiment analysis, detection of protected health information/ personally identifiable information (PHI/PII detection), and escalation to a human employee, by centralizing these skills and making them available to all individually skilled bots.

Orchestration is then critical to navigating operations across multiple bots, executing both conversational and procedural logic. This multi-bot model allows businesses to scale horizontally with thin-sliced bots by blending together FAQs, business processes, and transactions.

Breaking free of the breakpoint in 2020

Natural language processing is opening up a whole new way to engage with customers and employees. However, these users expect more from chatbots than just question and answer capabilities and want to communicate using phrases that reflect how they think and speak. This may not match the utterances designed by the organization to train their bot, resulting in a rougher user experience, or a bot that reaches its limit and fails in its original task.   

Taking a multi-bot approach allows organizations to expand their chatbot horizon, delivering greater conversational and automation capabilities that deliver both a better experience for users and desired business results, without being limited by NLP.

For more information on business topics in the United States, please take a look at the latest edition of Business Chief USA.

Follow Business Chief on LinkedIn and Twitter.


Featured Articles

Amelia DeLuca, CSO at Delta Air Lines on Female Leadership

Driving decarbonisation at Delta Air Lines, Chief Sustainability Officer Amelia DeLuca discusses the rise of the CSO and value of more women in leadership

Liz Elting – Driving Equality & Building Billion-$ Business

Founder and CEO Liz Elting Turned Her Passion into Purpose and Created a Billion-Dollar Business While Fighting for Workplace Equality – and Winning

JPMorgan Chase: Committed to supporting the next generation

JPMorgan has unveiled a host of new and expanded philanthropic activities totalling US$3.5 million to support the development of apprenticeship programmes

How efficient digital ecosystems became business critical

Technology & AI

Mastercard: Supporting clients at a time of rapid evolution

Digital Strategy

Why Ceridian has boldly rebranded to Dayforce

Human Capital