Why customer service chatbots fail — and how AI can prevent the tragedy

“Hello! How may I assist you today?”

“I want to talk to your customer care representative.”

“Sweet! Welcome to Phoebe’s Orchids. Over 100000 happy customers trust us to bring their mornings alive with rich colours and fragrance. :)”

“Nice. Now if you could connect me with your support rep, that would be great. Thanks! ”

“We’ll be glad to walk you through some pretty choices.”

“Please connect me with a team member.”

“Awesome! You can explore our colourful gallery here! ;)”

“Speak to an agent?? I have got the wrong bunch.”

“Wonderful! Here is another bunch we think you’ll like very much.”

“Impossible!”

*customer logs off*

One of the most exasperating experiences of contacting a business is when an IVR system or a hardwired chatbot keeps running you around in circles with its stale, canned set of greetings. The predefined script that hardly veers off course, having been built for only very specific use cases, is a time-tested recipe for customer vexation. And what makes the whole shebang even more annoying is that people do not have an alternative to signing up for this trouble.

Industry veterans know a “trough of disillusionment” when they see one. The term represents an element of the ‘Hype Cycle’ first given by the US research firm Gartner and denotes a period of time when both the customer and the stakeholder interest thin out in a particular technology after it categorically fails to live up to its inceptive hype. The initial fascination may recoil soon over time and the technology may ultimately rise to measure up to its initial promises but not without first going through a major overhaul.

Chatbots today are charting their own similar troughs of disillusionment.

2016 saw the birth and immediate demise of Microsoft’s disastrous experiment, Tay. The AI-powered chatbot, who got corrupted and manipulated within a short span of 15 hours, only lived, as briefly as it did, to parrot harsh cynicism and racist hatred picked up from this world, into this world — courtesy the Twitter fraternity. More than being a sadly failed attempt at linguistic machine learning, it exposed the fragility of such pioneering AI experiments and that of the tech stalwarts themselves whose innocent advancements could suddenly turn into ignominious misadventures on an unforeseen dark bend.

Post that, Facebook debuted chatbots for Messenger, promising consumers a new way to interact with their favorite businesses that, in its nascent stages, had its own limitations. The following years did leave much to desire and imagination but the interest never dampened enough and the technology only found more ground to flourish.

Today, people can simply “chat” with virtual automated assistants across the apps and websites of their choosing to get LIVE cricket scores, know the weather, shop, update their account details, order food, book an appointment, and more. From playing a psychedelic rock number to checking the account balance, chatbots are already redefining how we interact with the world around us. It’s almost safe to say that the virtual assistants are fast becoming epitomes of service delivered better, cheaper, quicker. A Gartner research says that businesses will be looking at AI as their mainstream customer experience investment in the next couple of years. 47% of organizations across industries and verticals will use chatbots to automate customer care processes and around 40% will deploy virtual assistants.

As the predicted use cases for a chatbot grow, as per a Global Market Insights study, the overall market size for chatbots worldwide would be over $1.3 billion by 2024.

However, notwithstanding their luminous future, there are still ample challenges that need tackling for the successful implementation of the chatbot technology for customer support.

Businesses trying to leverage chatbots to achieve tangible benefits still don’t have enough confidence in their virtual assistants and instead find them a long way from delivering the desired impact. There are simple, preventable errors still plaguing the system like providing the customer with the wrong insurance policy option that often ends up costing the company an individual’s loyalty and can be devastating for the brand over the long haul.

So let’s dissect the most common mistakes that today’s chatbots are highly prone to making and understand how the problems can be solved:

1. Unable to understand the intent

Chatbots often misinterpret the requests and end up solving the wrong problem because they are not able to understand the right intent of the customer. Understanding ‘intent’ is critical for a chatbot to be able to actually and successfully cater to what the customer really asked for, so much so that ‘intent’ is really the building block for effective virtual assistants. However, many customers often find chatting with a bot to be a very taxing experience.

Let’s try to understand how that can happen. Think about what a customer service agent does as soon as a customer approaches them with a query. They authenticate the customer and then automatically (and subconsciously) begin digging deeper into what it is exactly that they want. For instance, if a customer says, ‘I have my sister’s wedding coming up so I’ll need to make arrangements.’ Now the support agent will instantly interpret those words in his mind and ask, ‘So you’re looking for a loan, I understand?’

Now you can imagine what a chatbot is trying to emulate when interacting with a customer. A virtual agent that is not wired to understand the intent but only the dialogue and the flows will find it hard to infer from the customer’s statement that they are looking for a loan. It might take a while to figure it out before getting on with actually solving the problem.

So the way most customers converse can flip out most bots and render them useless since real-life conversations don’t necessarily use any keywords or commands a bot might be proactively trying to read. A customer could just say, ‘ Hey! I need to borrow X amount of money’, without ever actually mentioning the word ‘loan’.

A bot that doesn’t have the ability to map those sentences and utterances to the customer’s intent will fail utterly and miserably at complex conversations.

2. Fail to interpret nuanced instructions

Owing to lack of conversational intelligence, chatbots often end up being tone-deaf to nuances of the dialogue and that leads to an inaccurate conversation. Most chatbots today are not capable of Natural Language Processing. They are concretely programmed to understand only a very specific set of instructions and falter when the ambit of conversation expands even a little too much. For instance, a smart virtual accounting assistant that wields a jargon-free finance vocabulary would stutter if given a shove outside of its territory of knowledge.

Even if the realm of conversational topics stays the same, many chatbots find it difficult to understand colloquial language or accents in the conversation and find themselves instantly rummaging for words at the first acquaintance of a local language with region-specific terminology. Most then end up responding with inappropriate messages or executing incorrect commands that can spur serious trouble for a business. On the other hand, multilingual AI-powered chatbots that are capable of learning new languages regularly are able to hold more natural and meaningful interactions with customers.

3. (Not so) Great expectations

Confusing the users into thinking that they are talking to a human when they are not can be a serious security threat to a customer’s data and in direct conflict with an organization’s compliance standards and privacy policies. Businesses need to be transparent with their customers about using bots for communication to nurture their trust and respect their loyalty.

A malignant code masquerading as a genuine human assistant is a ticking time bomb that could end up severely pilfering sensitive information. In such a case, companies that are not upfront with their client base about their usage of bots for communications are tangibly mishandling their confidence, setting the wrong expectations, and breaching code of conduct.

4. When ‘more data!’ turns into ‘!more data’

To perform one of its primary functions concerning predictive analytics, modern AI requires expansive and rich pools of data. AI’s deep-learning algorithms employ large quantities of data to train and equip themselves with reasoning capabilities for all possible use cases but are left flatfooted when the data set no longer accounts for the reality of the world we inhabit. Most AI systems today, for instance, the algorithms to predict the buying behaviour of shoppers or to predict the best strategies for investment, tank from want of enough data to train and evolve themselves through.

For applying AI to well-defined tasks, the increasing availability of dynamic massive data sets is indispensable to growing their computational power and turning them into a transformative tool for humanity’s prickiest problems. Chatbots that are not backed by day-to-day software engineers that absorb data, clean it, and send it off a cloud AI service are unable to make the right choices and succumb to failure eventually.

5. No hybrid chat

Chatbots that are incapable of sentiment analysis, fail protocol and prefer to completely take the reins in their own hands do more harm than good to the business. If only Phoebe knew that she was losing customers majorly because she had not had her chatbots programmed to transfer to an agent when the situation asked for it! Often chatbots can fail to solve complex queries that can then require an agent’s urgent intervention. Also, certain problems require more than just technical assistance; they need an agent’s empathy, reassurance, and time, which no chatbot but only a human being can offer.

The best bots have the ability for seamless escalation to human agents, with the entire context and conversation history intact, at any point in the conversation to ensure high customer satisfaction. The ideal platform is an out-of-the-box system with an easy-to-implement method for agent handover that reduces the time to go live and is able to resolve customer frustration instantly.

To make this a notch further, identifying a VIP customer and bypassing the chatbot altogether to immediately assign an agent for them can go on to show them how you and your business honour their loyalty and help keep relationships strong as ever.

Also, read about Why AI-powered Chatbots are superior to Click-based Chatbots