The rage for organizations consuming natural language solutions continues to swell as businesses recognize the potential value that chatbots may have in automating business processes, reducing operating expenses and impacting the bottom line. There seems no letting up in corporations’ appetite along this journey, despite bots’ generally failed history and recognized weakness in driving measurable results in terms of incremental lift and conversion.
The Shiny New Object
The attractiveness in a chatbot’s usefulness is hard to ignore. The utility in the bot is mesmerizing – they work 24/7, promise to acknowledge requests with natural language over all touchpoints and satisfy customer requests purely through automation. The reduction of ongoing cost to service these functions is extraordinarily attractive.
As with most modern, engineered technologies that promote, “faster, better, cheaper”, enterprises have jumped on the chatbot wave and a new automation category has spawned into yet another “must have”. Fortunately, there are many successful use cases to showcase. Yet, as is often the case, organizations find that achievement meets diminishing returns when their expanded use cases extend past the usefulness of the chatbots’ original intent. While the capabilities of chatbots continue to evolve and become more expert, the risk of reaching a breaking point is real, when expectations overstep usefulness and the bot’s utility wanes.
Reeling in the Years
Introduced over ten years ago, the first-generation chatbots to a huge fanfare initially, yet failed to meet expectations leaving a frothy crowd disappointed. There were well known catastrophes. Various reasons explained these first fiascos. The early chatbot implementations had limited workflow automation as the primary interest was concentrated solely on reducing workforce involvement and its related costs.
Yet, first and foremost, the early failures were tied to the general immaturity of natural language processing as most bot interchanges were forced to quickly transfer to a live resource who could comprehend and respond to the issue at hand, leaving the potential for manual reductions waiting for more. Not to mention the cost in terms of customer satisfaction by not going to a live resource first.
The primary use cases for bots in this first phase were driven by mostly customer services, ignoring the other lines of business. These implementations were limited in core skills that could do nothing more than read and respond to rudimentary customer service questions. As simple as it sounds, customers became more frustrated following their experience than before they started and became increasingly disenchanted. The business suffered.
Despite these setbacks, the combined advent of NLP, ML and with further process refinement brought more promise. Organizations pushed for easier wins with chatbots, looking both strategically and within for a more expansive inventory of potential simple and straightforward use cases.
These next-in-line chatbot projects began to permeate enterprises around 2015, supporting multilingual and contextual capabilities, targeting tasks in the process that once again were simplistic in nature. Yet simplicity does not necessarily track with commitment and resources. The extremely publicized Erica chatbot project undertaken by Bank of America was estimated to cost $30 million, taking 100 people and over two years to deploy. Similar projects such as Capital One’s Eno was equally well known for its size, scope and investment level. While not all projects are of this nature, large enterprise scale chatbots projects have seen to reach these levels of investment often.
Given this level of commitment by organizations of all industries, shapes and sizes, it is more than shocking to see this technology, embraced a few years ago as the remedy for the business to be failing in the marketplace. Many of these projects, now fit with the most sophisticated NLP and chatbot scripting tools are failing and leading the experts at Gartner to predict that 40% of chatbot projects started in 2018 will have failed by the end of 2020.
Reasons not excuses
Alas, all is not gloom and doom for the chatbot world. There is clear evidence of productivity gains, relying on the significant strides made in machine learning and automation. Yet, since natural language technology is still evolving, there are challenges in developing and maintaining strong conversational experiences of any real substance.
Compounding this issue is the empty canvas with respect to benchmarks and best practices and the lack of data supporting real business scenarios that would train these chatbots. Almost all chatbots are unique by industry and/or client, leading to more delays, frustration and discontent. But there are more real reasons, some of which are technical and others that are more reflective of both the human element and expectation of the consumer and the complexity of the engagement that stand in the way of success. XSELL believes these impediments are real and no level of NLP and automation will suffice in order to reach the level of real human conversations for at least twenty years.
THE TECHNICAL BARRIERS:
This is really a matter of sheer total volume of both the number of intents and utterances and what a single chat bot may manage.
One intent say in a banking example may be “what are the closing costs” that would render “Give All Fees” as the intent name, as the consumer’s intent is to understand the closing costs involved in a transaction.
Likewise, an utterance is what specific words the consumer says. Taking the closing costs example, the consumer might ask for this information several ways: “can I see the closing costs due”, “how much are the all fees”, “are you charging points” or even “what are all the costs involved in the closing and can I include them in the total loan?’ All of these sentences could be the utterance.
As you can see, just one intent can have several utterances associated with consumer engagements. Consider extrapolating the massive number of intent and utterance combinations for any single business and you can begin to learn how issues can surface and the chatbot performance declines and scaling issues emerge.
- One Size Does Not Fit All
The second cousin to Intent Saturation is the problem when trying to merge multiple use cases into one chatbot, or as it is referred to as One Size Does Not Fit All. Enterprises use of chatbots are of various degrees of complication and structure, accounting for both conversational and customer journey perspectives. Quite often. The complexity of an individual use case exceeds the support a single-bot solution is designed to manage. Since enterprises are implementing more chatbots inside their organizations, they begin to realize that one chatbot is not equipped to handle one particular use case or another successfully.
To illustrate, an enterprise might first roll-out a chatbot to handle simple FAQs in a first phase. Subsequently, the team recommends including transact functionality, across a workflow and journey, include some small talk and even context switching. Doing this can often erode the usefulness of the originally intended use case, leading the consumer away from using any of it.
- Squeezing the Golden Goose for Accuracy
Enterprises can limit the general utility of a well-intended chatbot when attempting to increase its accuracy yield, let’s say by increasing the number of intent variants. While capturing the desired result, pulling the yield lever places stress on features and future capabilities impacting the potential usefulness of the bot’s intent.
- Oil, Plugs and Data
As chatbots immerse themselves into the fabric of the customer interaction lifecycle there is a clear need to maintain and nurture them so it may react and respond to a new new and missed utterances. Unlike college basketball, chatbots are not a one and done game and cannot be left without a plan for care and feeding.
Human language is fluid and the rate of change can vary among customer segments and generations. Compared to other digital assets, the requirement to review and refresh conversation is the most critical of all. Consumer preferences and sentiment is in constant shift mode, as is business objectives and initiatives. There is nothing more potentially harmful to an enterprise’s image and brand than the bot which represents the face of the company. A big mistake is misbelieving a bot is similar to an application – no!. Be prepared to continuously maintain new sets of data at any stage throughout the bot’s lifecycle.
THE HUMAN ELEMENT BARRIERS:
- The Natural Unnatural Fit
Chatbots are not going away soon and as a matter of fact they are here to stay for a long time. Certainly, this technology is having a major impact on the customer service arena. Over the past couple of years, automated phone calls, social media, live chat, and instant messaging customer service strategies have become more the norm. That said, there is strong speculation that automation and chatbots could replace customer service agents.
Yet, bots are limited in its applicability. They are terrific when addressing simple, paternalistic and routine issues like triage, discovery questions, collecting contact information, running surveys, and maintaining a standard experience such as greetings, explaining game rules or giveaways, and handling FAQs.
In the chatbot’s defense, many of us rarely want to talk to a human — we would just like our basic question answered quickly. When is your bill due, how much will something cost – sure the bot can answer this. Yet, it is time to talk with a human being to learn the details of your overcharge. AI and natural language processing (NLP) technologies are just not ready for that task.
We have found that when used as a customer service solution, chatbots have become highly problematic and risky for both customer care and marketing teams concerned about positive customer experiences and brand perception. When chatbots or other automation are unable to address customer needs fully, it can appear that the brand is cold and out of touch, with customers beginning to question their brand affinity.
- Seeking the Human Touch
Humans seek human interaction and relationships on topics important to them. Attributing to the success of chatbots has been brands finding a way to equip bots with the ability to seamlessly transition from a chatbot to a human, and then once engaged, provide the best possible experience, with the highest level of expertise possible, as if the agent on the other side is the best ambassador in the company. In order to turn the aspiration into reality, the gifted agent must possess both customer-facing and empathy skills and strong domain and product knowledge while in flight and in the moment.
The truth is, chatbots aren’t that well equipped for that task. They are developed using decision-tree logic, relaying on keywords specific to the user’s input or words. Bots that combine both linguistic and natural language learning capabilities are a rarity to say the least.
Another huge challenge for bots is that to have any meaningful conversations in the first place, chatbots need to understand human context and tone, which of course it cannot. This suggests that customer service and sales organizations need to either constrain the bot to pre-scripted dialogues and responses or expect and accept errors.
Naturally then, bots can’t be positioned to be the go-to option for primary conversations as they lack the one most critical component: emotion. We have empathy. We know what frustration is and can reason and listen. It’s nearly impossible to effectively support or market products without some degree of EIQ. Therefore, for the time being, best in class customer communications requires live human involvement.
While companies are judged by its customer support and service, the act of the final sale is recognized as the real “moment of truth”, the inflection point where product, price and positioning meets customer demand – the ultimate measure of success for the organization as an ongoing concern. The engagement setting is critical in the sales environment of course, and Bain & Cop says “the businesses that successfully engage with their customers were able to increase the customer spend by 20% to 40%”.
- The Voice (and Chat) for Effective Sales
So, with sales events so incredibly critical to the lifeblood of the enterprise and with the proper use of automation, NLP and Machine Learning becoming increasingly more important towards ensuring best practices, cost containment and customer centric initiatives, how do best in class organizations go about optimizing the selling experience in account of the people, process and technology?
We believe this setting demands a highly specialized solution that blends the best of multi-channel (Voice or Chat) enablement that begins with the effective capturing of words that when synthesized appropriately, will signal customer intent that leverages the most effective use of machine learning and artificial intelligence, driving a series of recommended sales tips, guidance and suggestions that have proven to have the highest likelihood to perform. These sales tips are sent to the live agent UI for them to follow, representing the next best action, offer, product, response or campaign. These “next bests” speak the language of the agent as it relates to what that customer is predicted to favorably respond to.
To learn more about how XSELL can accelerate your agent-assisted sales, contact Rob Levine.