Chatbots are software that simulates conversation through voice or text or both. Since the first chatbot ELIZA, developed in the 1960s by MIT Professor Joseph Weizenbaum), the technology has evolved. In more recent years, two core technology – natural language processing and artificial intelligence – are furthering the use cases with more recent interests around conversational AT chatbots.
Fit for job
Ellen Campana, director at KPMG, wrote that chatbots are being used to relieve stress points but only if the need is scoped properly. She cited the use of chatbots during the pandemic period to augment customer service requirements and to solve new problems brought about by COVID-19.
Forrester VP and principal analyst, David Truog, observed that companies are using chatbots today primarily to automate some aspects of customer service.
“They are focusing on replacing simple human agent interactions (like "I forgot my password") with automated chatbot interactions. To some extent, companies are also using chatbots to assist with product exploration and selection, and for marketing (lead generation) purposes, but less than for customer service,” he continued.
Clare.AI & WATI.io co-founder, Bianca Ho said clients use chatbots for customer support, and marketing from small SMB to large enterprises - eCommerce, financial services and high growth startups.
“ECommerce brands add WhatsApp number to their online shopfront or social media. WhatsApp is used to get a product catalogue, ask questions about the product, and receive timely updates such as shipment notifications. This helps them to build trust with customers more effectively and accelerate sales,” she elaborated.
Sandeep Bagaria, chief executive officer, Tagit concurred and added that key use cases today are around enquiries and product/business information and some basic transaction capabilities, like view account balances, simple transactions.
COVID’s influence
The lockdowns and mobility restrictions imposed by COVID-19 is giving chatbots a new sense of purpose.
Ho observed that the adoption of digital tools has accelerated. Previously, going digital was nice to have. “With Covid-19, businesses of all sizes had to go digital to survive – adoption for remote collaboration tool is essential to keep business running.
“COVID meant there were more business disruptions i.e., government regulations, lockdown restrictions- making timely updates to consumers extremely important. Consumers' expectations have been high and will continue to be. It is also necessary for businesses to adopt automation to help ease the workload,” she continued.
Truog concurred adding that the pandemic has caused people to favour interacting with brands remotely via channels that require human staff such as by phone or live chat (i.e., text chat with human agents) and via automated channels such as websites and apps.
The former has strained companies' capacity in terms of staffing and the latter has tested the effectiveness of companies' technologies and often found them lacking.
“As a result, users have attempted to make more use of chatbots (because interacting using natural language instead of hunting through menus and buttons is appealing — if it works), companies have deployed them rapidly for their customers (and in some cases for their employees and partners, too), and technology providers have accelerated their efforts to provide chatbot development platforms, tools, and services,” he explained.
Bagaria explained that the accelerated use of chatbots has come primarily in trying to offload simple queries from the branch and call centres.
What has worked? What hasn’t worked so far?
Barriers to fully utilising chatbots remain, however. Campana said pop-up chatbots need a formal role in communication workflows to be effective, including a description of what they will do and what they will pass along to human teammates.
Truog concurred adding that the failings are due to a combination of technology immaturity, poorly defined scope, and lack of UX design skills (which help with — among other things — the scope definition problem).
“The technology will mature, but slowly, so the most fruitful area for companies to invest immediately to create better chatbots is in UX design, especially in a new discipline known as conversation design,” he elaborated.
“Unfortunately, the vast majority of chatbots are poorly designed and therefore are disappointing to users and underperforming for the companies that deploy them.”
Bagaria agreed and added that: “Often chatbots feel no better than searchable FAQ’s whereas the customer is expecting to have a conversation as they would with a human. It is a “chat” bot.
What needs to happen
Truog remarked that success occurs where the chatbot focuses exclusively on a narrow domain, states clearly what it's able to handle, and is created by teams with cross-disciplinary skills spanning development and design — in other words, teams whose skillsets include not only technology expertise but also UX (user experience) expertise.
For Ho, there remains a need for chatbots that are easy to implement – especially given the volatile business landscape, businesses need to see a fast ROI.
“Starting small is the best path forward. Focus on the business pain points and find the right technologies to address and measure the outcome,” she suggested.
Bagaria posited for success to occur requires a combination of both, the AI technologies are maturing and are better able to determine intents to provide appropriate responses. This, also, needs proper training of the chatbots and for organizations to be clear on the use cases for which the chatbot is suitable.
This is further complicated by language enablement and the challenges associated with language understanding. Typical “translation” of a language to English struggles to solve this problem. Combine this with further challenges around adding context to the query of the customer and the chatbot struggles with meaningful use cases over and above the simple tasks they perform today.
Metrics of the effectiveness of chatbots
For Ho it comes down to efficiency – time saved from automated conversations, possible manpower saved – and effectiveness – increase in CSAT, revenue from chat experiences.
For Truog user satisfaction is the most important metric and some companies survey users after a chatbot interaction. But he cautioned that this is often misleading because users may (and often do) abandon the interaction before task completion because most chatbots' capabilities are so disappointing, at this early stage in their evolution. This tends to cause inaccurate, misleading measurements ("our chatbot is going great!" when in fact it is not).
“For now, it's best for companies to focus on reducing the rate of misunderstanding or non-understanding by examining the non-completed interactions and reviewing the back-and-forth dialogues that took place in those interactions to troubleshoot and improve the bot.
“This typically involves reviewing (e.g., weekly) user utterances to which the bot was not able to respond appropriately and using human expertise to map them to the correct intent to prevent recurrence of that particular misunderstanding (or non-understanding). Some companies refer to this set of utterances as the "confusion bucket" or the "IDK bucket," he explained.
Bagaria summed it up as completion rates, human takeover, chat volumes and customer satisfaction scores.
Tips for when shopping for a chatbot
Ho suggests starting from the outcome and working backwards – look at the use cases that they have implemented.
“Not only looking at the costs of implementation but also, aware of the resources that are necessary from your internal team. Who needs to be involved, how much time, whether they have the capacity or necessary skillsets,” she added.
Truog recommended starting the process without using any form of machine learning AI.
“Start by building a simple chatbot based on a deterministic decision tree and get that working before moving on to incorporating ML technologies as the next phase. This is important because most companies do not yet have in place the skills and processes for using ML initially.
“There is a lot of learning to do upfront about conversation design and understanding user needs and behaviour patterns before advancing to using ML.
“It's also crucial for companies to make sure the chatbot platform vendor they select recognizes the importance of UX design for chatbot success and provides tools and professional services to help with that aspect of chatbot creation, deployment, and ongoing evolution,” he concluded.
Recognizing that not everyone can afford to build in-house teams, Bagaria posited to look for vendors, that have deployed chatbots in similar industries as domain knowledge is equally important as technical knowledge.
