How to build a meaningful AI customer self-service application

How to build a meaningful AI customer self-service application

Apr 16, 2021


Not too long ago, allowing customers to access information and perform transactions by navigating through a series of menus from a phone keypad was seen as cutting-edge. IVR was quite the innovation! Since then, there have been many technological advances motivated by customer demand. Touch tone menus gave way to applications that let us speak our choices, and now self-service applications have further evolved to handle free human expression, either by text or by voice, in the form of chatbots and voicebots.

And while the promise of automated chat is revolutionary, building these applications without exposing their limitations remains a challenge. Touch tone’s innate limitations are obvious as there is only so much you can achieve with a twelve-digit keypad. The challenge with chatbots and voicebots is imagining how to best leapfrog past a limited set of menus.


How can you design and build an automated customer self-service application that provides meaningful service in a medium that encourages customers to say things unpredictably?


Understand Your Customers

Everything starts and ends with understanding your customers. What do your customers want to accomplish? Do they want to redeem their fidelity points, or see how many they have accumulated? Do they want to report a lost or stolen card, or want an explanation for an unexpected charge?


Use Real-World Interactions

Start building your app using real-world sample interactions, so that it understands pertinent interactions. Start small and grow as your interactions grow.


Feed Your App

Once live, you will need to keep growing what your app can understand as it will be exposed to unpredictable real-world conversations. Continue to incorporate user data into your app.


Review the Data

Expand your chatbot’s scope by analyzing and incorporating the questions users are asking. By nurturing your data library, you will be able to expand your chatbot’s self-service offerings and propose new ways for customers to achieve their tasks for customers to do what they need to do.


Clarify the Message

Develop strategies for handling misunderstandings.

  • Be explicit when prompting.
  • Echo back user questions to ensure comprehension.


Make Agent Transfers Seamless

Make sure that whatever the app has learned about your customer gets sent to the right agent. Especially during tense exchanges where an issue has been escalated, and the customer now expects to speak with a real person. Don’t make them repeat all the information they’ve already provided to the bot!



Take Care of Your Brand

A chatbot may be the customer’s first interaction with your company. We all know about first impressions, so make sure your chatbot embodies the way you want your company to be perceived.

Often, we focus too much on technology, and forget that the technology should represent your brand. Think of it as user engineering – tailoring the way your chatbot interacts with your customers in a way that best represents your company.

Consider how an online store for a high-fashion boutique wants to look and feel different from an app that redeems gas station cardmember points. Contrast that with how a health-services organization defines its role. The health-services enterprise may emphasize compassion and responsiveness, the fashion store may direct customers to complementary accessories, and the gas station may promote their early-bird coffee and muffin special. Messaging and tone of each company will be very different. The chatbot must also be consistent to the brand, otherwise users will become disoriented.


Managing Scope with Real-World Data

Before making the decision to build an AI chatbot, you must stake out a scope. What problems do you need to solve? Is your contact center overwhelmed by trivial interactions that can be automated, such as opening hours and locations? Do customers need to book appointments or reschedule them? Maybe they need to report a power outage or inclement weather. Perhaps there is a frequent need to make  corrections to a recent order, or to make a payment? Start by knowing what problem needs solving, and then refine, refine, refine as much as you can.

When developing a conversational AI tool, data is key. The more your data library is fed, the better the chatbot will be able to accurately respond to customer inquiries and deliver a flawless, human-like experience. One way to gain a clearer picture is to gather real-world data.


There are several ways to obtain real-world data:

1.  Use data already collected that meticulously classifies the purpose of every customer interaction.

2.  If customer interactions have not been classified, incorporate a plan using your staff to capture this data for a specific period of time.

3.  Instead of manually gathering this data, you could instead deploy an application – using speech or plain text – that simply asks “How can we help you?” and records the answers. A team can then be dedicated to classifying these recordings.


Transcribed and classified, these customer responses become the core that feeds the natural language engine at the heart of the chatbot. There’s no need to predict what users will say when you have real interactions forming your base. Nothing beats real-world data!


Real-World Data Feeds Machine Learning

Once your chatbot is deployed, real-world data continues to play an important part in tuning the application for both improving recognition and refining scope. Contrary to the classical approach used with language recognition, more recent systems use machine learning to leverage vast amounts of real-world data to guide what it will recognize.

By closely monitoring recognition successes and failures, machine learning algorithms provide relatively straightforward ways to re-classify its failures of understanding to join with existing user intentions and thereby organically grow its ability to understand human expression. To start, this requires careful monitoring and curation by people trained in this art, but after this initial period these applications largely teach themselves. The results of these techniques have revolutionized how machines understand what we say.

Likewise, the scope of an enterprise’s chatbot can be refined and expanded to meet new opportunities by keeping track of what users ask for. Maybe they are asking for services not yet offered through automation. By identifying them, companies can evaluate whether some of them provide an opportunity to deal with additional subjects through the chatbot.


A Balanced Mix of Customer Services

Chatbots and voicebots do not live in isolation. Although they may be your customer’s first touchpoint, they form only one part of a balanced mix of customer services. Some of these are face-to-face, others by phone or automated in a web browser. Regardless, it is important that your brand is reflected in all communications, and that they receive a consistent customer experience across channels. From speaking to Holly in Accounting on the phone to chatting with the company chatbot, the experience should be seamless.



Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

blender bitcoin bitcoin mixer bitcoin blender blender io cryptomixer