AI 101: The Basics of Automation for Customer Support


Share on LinkedIn

While you might not know much about artificial intelligence yet, experts predict that the use of machine learning technology will double in 2018 compared to 2017, and then double again by 2020. This enigmatic technology — seemingly lifted from science fiction movies — is already a staple of many industries, but 2018 is the year AI will become a part of everyday life.

So, what is AI, and why is the New York Times calling it one of the most significant tech trends of 2018?

Simply put, AI is the ability for a machine to mimic human behavior. Most AI systems leverage what is known as machine learning — a discipline of computer science which focuses on teaching computers to learn from data. Machine learning allows computers to study examples, recognize patterns, and incrementally learn, without explicit human instruction.

This might sound abstract at first, but there are countless real world applications for AI. Today, forward-thinking companies in manufacturing, healthcare, and — for our purposes in this guide — customer support, are already harnessing AI as a means of augmenting employees and speeding up day-to-day processes.

The most common applications for AI in customer support can be understood in three broad categories:

Natural Language Processing (NLP) : The ability to understand customer intent by interpreting words as if the computer were a human. Through AI, a computer can adopt the mindset of a human, read between the lines, and extrapolate meaning from colloquialisms, slang, and language nuances.

Dialog Management/Conversational AI: A computer can thoughtfully conduct conversations with customers or support agents and can suggest responses based on intent.

Speech Recognition: This is recognition and translation of the spoken word. Speech recognition transcribes speech into text so it can be analyzed using the techniques outlined above.

Thanks to these three AI methodologies, customer support teams have experienced significant improvements in agent productivity and CXM efficiency.

Many CXM managers wonder how they can integrate machine learning into their system. Well, it’s a lot easier than you might think. But, first things first: we need to clear up what AI is and what AI is not.


There is a common misconception that AI and Chatbots are the same thing. While it’s easy to confuse them in the context of customer support, chatbots are just one aspect of AI implementation specific to the chat channel.

Chat is gaining popularity due to the perception of lower costs, ability to automate with bots, and, most importantly, catering to millennial customer communication preferences.

A Chatbot is a software designed to simulate conversation with human users. It is a virtual assistant that communicates with humans through instant messaging interfaces and text messages. The bot is an automated means of communicating with users.

Chatbots combine AI technologies — Natural Language Processing and Dialog Management — which allows customers to ask questions naturally and have the bot resolve simple issues or carry out simple tasks. Chatbots are designed to quickly and efficiently understand intent and resolve common informationrelated questions such as password reset, account access, and troubleshooting.

One of the biggest drawbacks to Chatbots is that they are expensive to integrate. Chatbot technology must integrate with existing software, and this can get pretty expensive. This software cannot simply be tacked onto a CXM — Chatbots require exhaustive, pricey integration in order to perform their duties.

Additionally, beyond tasks such as order tracking, billing, refunds, and other simple interactions, Chatbot utility quickly falls off (unless they are trained and integrated with line of business systems). Anyone who has braved what can feel like an endless loop of text prompts knows that Chatbots simply lack the contextual knowledge to effectively account for all the complexities of language. And here’s why: without conversational AI, Chatbots lack the ability to learn and adapt to customer needs.

The term Chatbot is not interchangeable with AI. Rather, more advanced AI is the solution to Chatbot inefficiencies.

The adoption of AI is inevitable — the question isn’t if you will incorporate AI, but rather, when. According to Gartner, Inc, “by 2020, AI technologies will be virtually pervasive in almost every new software product and service.” Early adopters of this game-changing technology stand to significantly improve day-to-day workflows, break away from the competition, and unlock the true potential of their CXM. But how?

It takes a collaborative approach between support organizations and the Intelligent automation provider to understand the decision trees and internal processes for automation that goes beyond simple content-based deflection.

The best way to incorporate AI into your CXM is through a 3rd party AI partner who can offer an AI model that mimics the decisions of your best agents based on historical responses. The most effective AI systems study your CXM’s history of inputs and builds a correlation model. AI then learns the ins-and-outs of your CXM, which allows it to offer specialized insight based on historically-accurate responses. Here’s how AI begins to impact customer support:

• Observe: The first step is simple: AI silently observes how agents respond to questions, compares the results, and incrementally learns how to best serve your CXM. This is often broken into two steps: data collection and training.

• Prediction score: After a period of studying, AI can start offering “mock responses” which are used for testing alone. These scores range from 0 – 100 percent and represent how confident the AI system is with the demo response. Any response exceeding 70 percent is a good candidate for complete automation. Anything under that threshold may not be ready for complete automation, but may qualify for AI augmentation such as recommended response (more on that in the next section).

Essentially, every resolved ticket is an opportunity for an AI system to learn and adapt. After a period of studying, and once machine learning has reached a predetermined confidence threshold, it is time to start implementing AI solutions within your CXM.

As you can see, AI for customer support is a process — it is not as simple as flipping a switch and suddenly all tickets can be automated. There are shades of AI, and this subtlety is lost in the overall conversation about machine learning. Recognizing the variety in AI solutions — options which might not include complete automation — is important to uncovering significant ROI.

There are several AI solutions — be that automation or agent augmentation — and each option improves CXMs differently:

Triage: This is the first logical AI implementation for most CXMs. It is easy for an AI algorithm to learn ticket intent and route tickets to the correct agent. Plus, this is a great way for machine learning to study a variety of customer inquiries. AI identifies patterns in ticket classification and predicts which agent group is best equipped to handle the customer inquiry. Natural Language Processing is well equipped to detect spam, so you can rest assured all tickets are genuine. There’s no sense in wasting agent energy on repetitive tasks like triage, so this is usually a good inroad into AI for customer support.

EXAMPLE: Expedia® routes over half a million tickets a month — without any human oversight. Most businesses with high volumes of tickets/cases use rules-based automation, but this method has limited success and requires significant human interaction. With the help of triage technology, Expedia automatically routes tickets at an accuracy rate of over 90 percent.

Recommended responses: Once acquainted with the nuances of your CXM, AI can begin offering agent suggestions known as recommended responses. AI provides the top three macros and templates based on historically-proven ticket resolutions. This helps guide agent interactions, and macros can be personalized according to customer need.

EXAMPLE: Thumbtack® is an online service that millions of customers use everyday to find help with specialized tasks — they hire professionals such as plumbers, handymen, piano teachers, etc. Thumbtack is able to handle over a million requests a year, with only a small pool of agents, because they use a recommended response service. Thumbtack agent productivity improves by over 30 percent when they use a recommended response.

EXAMPLE: A leading gaming provider, which receives 200K tickets per month and has 400 agents, implements over 1000 macros to respond to 70 percent of their tickets. They saw a 66 percent increase in agent productivity (50 tickets/day rather than 30) by using the recommended response service.

• Ticket Automation: Automating responses with zero touch from agents is the ultimate goal for most businesses as there is no human cost attached to handling the ticket. After a period of studying tickets, AI can begin automatically responding to tickets. Refunds, password resets, package tracking, and other common tickets are prime for automation. Once the AI system has reached a predetermined confidence threshold (usually 70 percent or more) it can begin automating tickets, relieve agents from repetitive tickets, and enable them to spend more time offering individualized customer care.

EXAMPLE: Groupon® is an e-commerce platform where millions of customers worldwide shop for great deals. Groupon support tickets range from simple inquiries about products, services, and policies to complex questions about coupon redemptions, rescheduling appointments, cancellations, and refunds. This company uses artificial intelligence to automate responses for complex scenarios such as refunds. The AI system processes a refund request without human involvement by retrieving order details and automatically applying relevant business rules as per the refund policy

While zero-touch ticket automation is the gold standard, significant ROI can be uncovered throughout the entire spectrum of AI. The variety of solutions outlined above illustrate that, as long as you have a decent ticket history for machine learning to study, there is an appropriate AI solution for every CXM.

The most important consideration when choosing an AI solution for customer support is to evaluate the turnkey nature of the offering. Most AI platforms can take anywhere from three to six months of learning before any automation can be implemented.

That’s too long of a learning curve.

Not only does this slow timeline delay realization of the benefits of AI, but it puts a huge burden on the customer support team. Months of implementation will overload team members with incomplete, sluggish processes. CXMs with slow AI integration generally experience all of the delays of a new technology with none of the benefits. An AI solution that requires several months of implementation is effectively offering one step forward, and two steps back.

Keep your eyes peeled for a solution that is turnkey — where the AI solution can enable automation within 30-60 days.

Pradeep Rathinam
Pradeep Rathinam (Paddy) is the Chief Customer Officer (CCO) at Freshworks. He leads a global team of customer experience employees that span customer success, professional services, and customer service. He founded AnsweriQ, an AI-based customer service automation company, which was acquired by Freshworks in 2020. Previously, he was the CEO of Aditi Technologies and spent over a decade at Microsoft as general manager of the Independent Software Vendors business. Pradeep holds a business management degree from Delhi University.


Please use comments to add value to the discussion. Maximum one link to an educational blog post or article. We will NOT PUBLISH brief comments like "good post," comments that mainly promote links, or comments with links to companies, products, or services.

Please enter your comment!
Please enter your name here