How deep learning drives better customer experience

Published November 4, 2020
Last updated November 4, 2020

Guest blog from Pyry Takala, CEO, TypeGenie

The generational shift from Millennials to Generation Z is fundamentally changing customers’ expectations in customer care. Not only does Gen Z love to solve problems on their own in the first place. They expect customer service 24/7, and they want it immediately and highly personalised.

Keeping up with these expectations is a struggle for many customer support managers. Sure, you can pre-write answers to standard questions, and present them as knowledge articles, canned responses, or chatbot flows in your support centre. But this approach clearly has limits. Adding knowledge articles to cover the top 20 FAQs is extremely valuable for clients. Adding answers to the next 200 questions, however, might not add the same value. And while customers want quick solutions, above all they want to get the answers they need and are not necessarily satisfied with template answers.

This is going to change: breakthroughs in deep learning technology can help meet these increasingly challenging expectations.

Personalization at lower cost

Artificial intelligence was a slowly developing field in the early 2000s, with very limited adoption in customer service. However, following research breakthroughs in the field of artificial neural networks in the late 2000s, this started to change. By June 2016 Scientific American saw artificial intelligence “...finally catching up to its early promise, thanks to a powerful technique called deep learning.” Over the last four years, deep learning has made further rapid advances. Deep learning is what allows seamless speech recognition and image recognition across almost every possible industry today. It’s the sophisticated pattern-recognition technology that enables self-driving cars, new cancer diagnostic tools, or empowers the latest Google Translate. With it come also the possibilities to automate simple tasks of a customer service agent and to personalise customer interactions, making it one of the most promising advances for customer support.

 

Deep learning outperforms standard responses in terms of personalization and cost.

Writing and maintaining thousands of FAQs would be a daunting task for a human. Because of breakthroughs in deep learning, machines can help craft new responses on the fly. The real value, however, goes beyond that: Unlike templates, deep learning can adapt to the situation, for instance adjusting the tone of voice and style to the customer. This goes hand in hand with a continuous increase in quality and drop in support costs.

Deep learning is not perfect, though. It works best when used to augment the service agent’s skills, not to replace them.

Deep Learning in practice

2021 will not yet see fully automated contact centers, but the first innovative support teams are already benefiting from deep learning today. It’s estimated that up to 40% of customer contacts could be fully self-serviced. Gartner estimates that by 2022, 85% of customer service interactions will start with self-service, up from 48% in 2019.

Deep learning eliminates repetitive typing

One of the most time-consuming activities of a customer service agent’s work is typing. Answers are not the same, but they do have repetitive parts. This is where deep learning is already used by innovative service teams. Let’s look at three examples:

JustPark, a car parking app with 3.5 million users in the United Kingdom, is using deep learning capabilities to faster resolve parking issues in their customer support centre. This can be parking space bookings, credit card payment errors, or questions on vehicle detail changes that have one thing in common: They tend to occur frequently and lead to repetitive typing. Deep learning-driven typing technology helps make a human agent more efficient by proposing personalised sentence completions, shortening response times along the way while freeing up agents’ time for more complex, value-adding work.

Los Angeles-based online clothes retailer Sole Society is using the same technology to resolve delivery issues faster. Up to 450 sentence completions per day help their relatively small support team to cope with customers’ requests.

In a similar way the Dutch logistics company, SendCloud, is seeing deep learning-assisted responses driving the productivity of their teams in Dutch and a variety of other languages.

Deep learning improves service quality

Besides increasing productivity and speed for typing, response quality is a different sort of challenge, especially when you work with outsourced teams. As deep learning has improved sentiment detection accuracy, a Quality Assurance manager can now identify within seconds conversations with unhappy customers for further action.

 

A sentiment classifier by Sentient Machines highlights positive and negative phrases in past customer service conversations. This allows the QA manager to give very precise feedback to a service agent. Credit: Sentient Machines

Preventing quality issues before they occur is even better. Language correction app, Grammarly, for example, helps service teams to optimise grammar and writing style while they are in the process of writing.

2021: Auto-triage, template recommendations, and more

These trends are only first glimpses of what is to come. In 2021, we will see deep learning support being rolled out by firms across the entire workflow of service agents. We at TypeGenie expect three major trends:

Auto-triage and classification will propel ticketing system automation to the next level.
Today, tickets are typically distributed based on customer-chosen categories or in a “round robin” fashion. This means that service agents need to do a lot of context-switching between tickets.Auto-triage or automated ticket triaging, however, involves automatic analysing and directing support tickets to the right agent, and sorting tickets in order of priority, topic or urgency.

Managing the customer support queue with template recommendations: Are you using canned responses as predetermined answers to common questions? Finding the correct canned response can often be a challenge, especially for new agents. AI will start surfacing the most relevant template when one is available, saving agents time and helping new service agents to become aware of the existence of various templates.

Identifying your FAQ gaps: Deep learning will be identifying the areas where customer questions are not sufficiently covered and suggest where you should add a knowledge article. With new articles covering these gaps, more requests can be deflected.

All of these improvements will help drive higher efficiency in support centers and save agents time, but perhaps more importantly provide speedy responses, contributing to delivering a service that meets the expectations of Generation Z.