A Guide to Effective Customer Service Chatbots

Highlights

  • Contact center chatbots are well suited for monotonous and repetitive cases. Research by IBM shows up to 80% of routine customer support questions can be answered by a bot.
  • Contact and call centers can move high volume, low complexity cases to bots and thereby increase accessibility and decrease cost.
  • Despite broad adoption, customer research shows that most chatbot projects fail. The top two reasons for failure include an overly broad scope and unintelligent bots that do not satisfy customers. These failure points can be remedied with chatbot best practices.
Johnny Grow Revenue Growth Consulting

A chatbot, often just called a bot, is a software application that uses online text or speech to simulate conversations with people. It uses natural language processing (NLP) technology to converse (the 'chat') and process automation to fulfill requests or complete tasks (the 'bot'.)

In a customer support context, these bots are trained on customer support transcripts, governed by business rules, often operate pursuant to deterministic decision trees and are powered with NLP and artificial intelligence (AI).

Virtual assistants, often called virtual agents, are closely related to chatbots. They also simulate interactive human conversations using AI and NLP. Their conversational interfaces are most often designed to assist users with questions, information retrieval, decision making, problem resolution and task completion.

Virtual agents have become common place in our personal lives. Popular virtual agents include Amazon Alexa, Apple Siri and Google Home.

So What's the Difference between Chatbots and Virtual Agents?

Sometimes customer service managers are unclear in the differences between bots and virtual agents.

Chatbots tend to be text-based, trained with structured data and designed for limited and specific business processes.

Virtual agents tend to be audio based, trained with a mix of structured and unstructured data and are designed for broader use cases. They are natively built on AI and natural language processing and apply sophisticated speech analytics to create more cognitive and conversational platforms.

However, the bots are increasingly adopting AI and NLP to become cognitive and integrating with Robotic Process Automation (RPA) to assume more sophisticated task processing.

The two technologies are converging, and the subtleties are becoming differences without distinction.

The Business Benefits

Bots are seeing tremendous adoption because they can simultaneously deliver customer benefits and significant cost reductions.

Analyst firm Gartner advises that most customers now demand the ability to manage their relationships with brands without engaging a person. Repeated research from many sources shares that most customers prefer to solve product problems without contacting the call center. They prefer self-service options that they can manage on their own timeline and without being tied to a linear conversation. This is a trend that will only grow as Millennials become the majority of the workforce.

Chatbot benefits fall into four categories.

  1. They improve customer satisfaction. Bots are available on-demand, 24 by 7. They improve customer experiences by eliminating wait time, accelerating resolutions (i.e., decreasing speed of answer (SoA) time) and delivering a consistent level of service.
  2. Bots improve the agent experience. They assist agents with guided service delivery and AI-based Next Best Action recommendations. This type of staff enablement creates a synergistic combination of agents and intelligent bots working together to do more than either can do alone. The bots absorb monotonous and repetitive tasks to free up agents to focus on the customer experience. They can aid agents by automatically authenticating customers, ingesting customer supplied data, accessing legacy systems to retrieve helpful information, populating fields in forms and a host of other tasks that otherwise require manual labor. They allow the agent to focus on the customer and deliver the most important and fulfilling piece of the resolution experience.
  3. They lower the cost to serve. Contact centers move high volume, low complexity cases to bots and thereby reduce labor costs. Common use case examples include transaction inquiries such as order fulfillment and package tracking updates, support requests such as inventory availability or return policy, cases to process claims such as credit memos or return merchandise authorizations, and repetitive questions such as business hours.

For example, bots can quickly query ERP systems with integration to distribution carriers to determine where a product is and when it will be delivered. This function alone creates a big savings as this type of inquiry accounts for 10 to 12 percent of cases for consumer product goods, retail and inventory carrying companies.

They bring efficiencies and scale to the contact or call center. They can manage a virtually unlimited number of simultaneous customer interactions. That flattens peak periods and reduces overall case volume. Managing unanticipated peak volumes is especially useful for occasional fluctuations such as company downtime that impacts production capacities or crisis scenarios such as the Covid pandemic.

Even when visitors are escalated to live agents, they can determine the types of cases to forward to live chat, which is another channel that offers case concurrency. Live chat and messaging apps keep customers in a text-based communication, so agents can engage multiple customers concurrently and the support organization can achieve agent economies of scale. Agents in chat typically handle 2 to 5 simultaneous customer sessions unlike calls where agents are limited to one customer at a time.

Customer Service Channels
  1. Bots accumulate valuable customer Intelligence.
    Customer intelligence enables contact or call centers to deliver relevant, personalized, contextual and predictive customer experiences. Customer intelligence is a company's most valuable intellectual property. Yet for most support organizations it remains just a good idea. Bots collect unconstrained and unfiltered customer input in the forms of questions, problems, requests and frustrations.

Their transcripts identify customer friction points and where those sources of frustration occur along the customer journey. They capture digital footprints, customer sentiment and communication dialogue such as what customers want, what products they struggle with and why. This content creates customer insights that feed customer support processes and contact center analytics. Mature contact and call centers harvest chatbot data as part of their Voice of the Customer (VoC) program and feed that data to the CRM 360-degree customer view.

Why Customer Service Chatbots Fail

Despite widescale adoption, customer research and the experience of anybody who has tried to use them, shows that success remains elusive. The reality is that most contact center chatbot implementations focus on the technology and not the customer experience. That results in a degraded customer experience, low customer satisfaction scores and ultimately a failed implementation.

The top two reasons for chatbot failure include an overly broad scope and dumb bots that fail to provide relevant and accurate help.

Dumb bots are sometimes more graciously called foundational bots. They are built on a purely deterministic configuration that scans text for general keywords and delivers common phrase responses extracted from a file or database.

Many times, they are just page pushers, meaning instead of an answer, they deliver a page of links to webpages or knowledgebase articles that may or may not have the answer. Page pushing is easy to implement but research shows it mostly fails.

Customers don't have the time or patience to search through more information. They expect answers to be delivered. Dumb bots normally fail because their accuracy is low, their responses are most often incomplete and any dialogue beyond the initial question lowers relevance, and often loses communication fidelity.

Smart bots, sometimes called cognitive bots, are built with AI and machine learning. Most are self-learning, so they improve with each interaction. Implementing AI can seem complex but when adopting the AI tool in your CRM software the process is made much simpler.

A Better Way – Contact Center Chatbot Best Practices

When planning implementations with clients I can't tell you the number that previously had a failed POC (proof of concept). More often than not, the client viewed the bot builder as relatively simple so began a technology focused deployment that didn't adopt user experience best practices, targeted customer use cases (and their many variants) or measurable outcomes such as increased customer satisfaction (CSAT).

To do better, here are some chatbot best practices that we've applied many times over.

  1. Start small and take a phased approach. It's a good idea to start with a narrow scope of use cases best served with bots. Remember, this technology is best suited to automate low-variability, high frequency requests. Smartbots can step up to moderate complexity cases. Achieving customer satisfaction is much easier with highly practical and well-bounded use cases.
  2. Choose a small set of use cases based on case history. Don't try to anticipate what customers might ask. Instead, tabulate case history to see their actual requests.
  3. Recognize that the user interface (UI) and user experience (UX) are essential to customer adoption and satisfaction. Most successful bots follow the same user-centered design principals as social technologies.
  4. Keep within your CRM technology stack if possible. Popular contact center CRM systems such as Microsoft Dynamics 365 with its PowerApps Chat-bots and Salesforce Service Cloud with its Einstein offer purpose-built solutions with extensible toolkits. By using tools within your CRM software, you eliminate yet another contact center point solution, leverage your existing technology architecture and avoid system integration.
Chatbot Architecture
  1. Use AI to make your bot smart. This will require AI technology and training models to move beyond linear decision trees, rudimentary keyword replies, static dialogues and hand programmed responses.
  2. Don't release before a thorough MVP and user acceptance test. The bot is ready when it can pass a comprehensive UAT or the Turing Test, which is when a test evaluator cannot reliably tell whether the text interaction is with a person or machine.
  3. Make sure your bots offer on-demand escalation to agents. Customer support representatives must be an immediate backup when the technology fails to determine intent or deliver accurate responses. Upon transfer, the app should also pass the content to the agent. Finally, all escalations should be reviewed to see if something went wrong or how mistakes can be prevented in the future.
  4. Each automated customer experience should be followed with an automated CSAT survey. This customer feedback will be your richest source of information for refinements or improvements.

Following these best practices we've found bots can receive positive average CSAT scores and quickly deflect 15 to 25 percent of inbound customer inquiries. It's also been our experience that clients realize about 14 percent queue abandonment rates however, quick diagnosis and course corrections steadily reduce that figure.

Using the customer service chatbot best practices, bots can receive positive average customer satisfaction (CSAT) scores and quickly deflect 15 to 25% of inbound customer inquiries.

Click to Tweet

One more thing, you will need to adjust your contact center performance measures. Because bots siphon off the most common, repetitive and simple customer inquiries, agents are left with the more challenging and time-consuming cases. For example, average handle time (AHT), speed of answer (SoA) and frequency of tiered escalations will all increase as agents are left with the more complex cases.