Part I: The State of Chatbots in 2023 – Lots of Dollars & Hype, Poor Results
Bloomberg Businessweek’s October 5th, 2022, headline said it all: ‘Even After $100 Billion, Self-Driving Cars Are Going Nowhere.’ In short, this article details the journey from the envisioned promise of our impending utopian Autonomous Driving Future – a breathtakingly efficient and safe ecosystem, free of traffic jams, accidents and traffic fatalities, and our daily Human driving rituals – to the delivered reality of something that looks more like a joint venture cooked up by Bernie Madoff, Adam Neumann, and Elizabeth Holmes. In other words, a $100 billion scam, with little to show for the effort.
For CX professionals working across all industries, the over-promised-yet-far-under-delivered predicament of our Autonomous Driving Future should sound eerily and uncomfortably familiar and parallel to that of their own elusive Great CX Hope: the all-knowing, customer-delighting, cost-slashing Chatbot. Chatbots (as a key focal point of broader-based Intelligent Automation strategies) have been a top CX investment priority for firms of all sizes for the last decade, with billions of dollars being thrown at the problem/opportunity. And while much has been (and continues to be) written about the evolution/advancement of chatbots, their cross-industry capabilities, and their potential for significant business impact – including glimmers of hope for/in specific industries and use cases – it is widely agreed that they are not living up to their promise in terms of capabilities, consumer sentiment and satisfaction, and/or return on investment (ROI) impact. A few stats paint this picture quite well:
So, against this backdrop and scorecard, is there still reason to believe in a vastly-improved Chatbot Future, and why should we even care anymore? As to the first question, we believe the answer is yes – and/but, as with any interesting problem/opportunity, realizing that Future requires (1) a new approach, and (2) realigned expectations that define success.
As to the question of why we should continue to care, the ROI potential in terms of both customer engagement/loyalty impact, as well as significant potential cost savings, are too large to ignore.
Part II: Why Chatbots Have Continued to Fail – And/Or at Least Underwhelm – Us
Before we jump into the way forward for delivering on a much brighter Chatbot Future and ensuing ROI, some techno-jargon-free context is helpful in understanding exactly why chatbots have failed to live up to their promise to date. We break this down into four (4) key areas – Design & Architecture, Maintenance & Upkeep, Knowledge Sources, and Expectations – with a brief explanation of each below.
1. Design & Architecture
Most chatbots are still designed and deployed using what we’ll categorize as the first two (2) generations of chatbot technology: (1) Rules- / Flow-Based, and (2) Intent-Based. In short, both generations are/were built on specific, pre-determined and pre-programmed ‘conversation’ flows, basically a series of ‘If-Then’ scenarios aimed at solving specific, predictable customer inquiries/issues. While Intent-Based chatbots offer(ed) consumers more proactive and pre-selected issue-/intent-specific options to choose from (e.g., ‘Reset password’, ‘Open New Account’, etc), they are/were still designed with the Rules- / Flow-Based taxonomy – basically a larger chatbot, with a collection of pre-determined flows to solve multiple customer inquiries.
At this juncture it’s also very important to note that these types of chatbots are actually QUITE EFFECTIVE* in handling simple, predictable customer inquiries – e.g., ‘Where is my order?’, ‘Reset my password’, etc. (*We will come back to this in the Expectations section below…) Their primary limitations and failure points, however, lie in their rigidity – in other words, they’re very easy to ‘break’ when a customer steps outside of the predetermined ‘rules’ and ‘conversation’ flow(s) that power the chatbot. As we’re sure most of us have experienced, even small deviations in word choice can run the bot into a wall – and, in turn, frustrate the customer. Significant advances in multiple technologies are now addressing these Generation 1 & 2 limitations under the banner of ‘Conversational AI,’ but these advances in and of themselves don’t eliminate the failure points. Which leads us to the next limiting factor, Maintenance & Upkeep…
2. Maintenance & Upkeep
Simply put, chatbots are traditionally difficult and time-consuming to build, and expensive to maintain. We’ll address this further in the Expectations section below, but this is the area that has likely contributed most to the disappointment, disillusionment, and frustration with chatbots’ failure to meet their positioned promise as the Great CX Hope. While Technology Sales Leaders have eagerly and aggressively positioned the eventual potential financial benefits of chatbots (e.g., increased revenue via improved digital self-service; decreased costs via elimination of interactions reaching the contact center), they have typically minimized or under-estimated the associated ongoing maintenance costs that fall outside the actual chatbot technology software fees (including data analysts, data scientists, data engineers, systems integrators, conversational designers, etc). Eleviant Technologies summarized this maintenance conundrum very well in an August 2021 article:
A key part of the Maintenance & Upkeep process to ensure any chatbot’s relevance, speed, and accuracy is ensuring connectivity to the most updated Knowledge Sources, our next limiting factor…
3. Knowledge Sources
Humans learn from a vast array of formal and informal sources and experiences. Machines (including chatbots) ‘learn’ in much the same way, albeit with much more limited informal contextual reasoning and association than Humans. (Refer back to the self-driving car article for numerous examples of this.) For our purposes in the advancement of effective and relevant chatbot performance broadly, the MOST important aspect of ongoing chatbot maintenance is access to, and/or integration with, updated Knowledge Sources within a business. In short, without access to updated and accurate content – the source(s) of truth, if you will – chatbots will fail quickly, and frustrate customers. And unfortunately, this is exactly where/why most chatbots fail: They are not accessing the most accurate, and/or all relevant Knowledge Sources, including CRM, CMS, KMS, FAQs, among others.
Beyond the customer frustration factor, this failure point then results in very significant increases in contact center volumes, as any contact center agent will tell you. This outcome is also risky, as Human agents are left to access the same multitude of disparate systems and knowledge sources that may or may not be updated/accurate. In the absence of a given agent’s significant experience and ‘tribal knowledge’ from handling numerous similar customer interactions, even the Human element could produce a poor outcome for the customer.
The final limiting factor and failure point for chatbots to date has simply been over-ambitious, misaligned, and mismanaged expectations around chatbots’ capabilities and business impact. The IBM prediction/proclamation we note above is but one example in a nearly endless sea of analysis on this topic – all well-meaning, but again short on details of the ‘how’ and ‘when’, as well as the constraints and conditions to make the Great Chatbot CX Hope a reality.
Part III: The Way Forward – Conversational AI Meets ‘Transformers’
As we noted near the top of this post, despite the somewhat rocky chatbot journey and results to date, there is reason to be very excited about a much brighter future ahead, and much improved delivery on chatbots’ original promise as the Great CX Hope – albeit with a new approach and reset expectations. Numerous advancements have been made in the specific technologies (Natural Language Processing, Natural Language Understanding, Machine Learning, Deep Learning, and Predictive Analytics) that comprise what we now refer to as Conversational AI, and these advancements allow us to break free of the rigid, rules-based, pre-determined conversation flows that have defined the first couple generations of chatbots – in favor of a much more dynamic, flexible, and Human-like experience to drive more elegant automation and digital self-service.
The critical technology component driving this new approach is what we refer to as ‘Transformers.’ In short, Transformers greatly reduce (and in some cases, eliminate) the key failure points noted above, and provide brands speed, ease, and scalability of self-service intelligence derived from the real source(s) of truth: their current content, regardless of where it lives (CRM, CMS, KMS, FAQs, websites, mobile apps, etc). By using this approach, next-generation, conversational, effective self-service can be launched in days, not months – which can either immediately complement a brand’s current-state automation/chatbot investments, or replace them. Importantly, we can and should think about (and power) brands’ holistic automation strategies, covering both customer-facing and agent-facing automation. As we consider the pressure of increasing contact center volumes, creating stress on already overwhelmed agents, we can use this approach and technology to create a sort of ‘Super Agent.’ Depending on the specific intents, a brand can conservatively shift 7-10% of your current contact center volume to self-service almost immediately, and deliver significantly decreased agent handle times (AHT) – which will significantly drive down operational costs, while simultaneously improving CX and revenue.