8,900+

American music retailers in 2020

$4.5 billion

2021 musical instrument & supply stores market size

1.4%

expected market growth in 2021

clients_ThermoFisherLogo

EXECUTIVE SUMMARY

To help a leading music retailer provide a better experience for their customers, we build a call center chatbot for them that can answer frequently asked questions.

GOAL

How can we create a chatbot experience that is as good, or better, than speaking to a live employee?

RESULTS

Using Amazon Lex, the team build a fine-tuned chatbot that could answer frequently asked questions about store locations, store hours, and more.

Background

A leading American music retailer wanted to improve their customers’ experience by providing a call center chatbot that could answer frequently asked questions. This way, customers could get fast and reliable answers without having to try to reach live employees at physical stores.

The Challenge

The team faced a few challenges in building this chatbot. First, Lex had trouble recognizing city names, so the team needed to build a custom slot type and code for appropriate fallback logic. The team also needed to figure out how to get the system to retain the information that was given at the beginning of the conversation, and have it readily available throughout the call. Lastly, the project required the system to include a lot of intents, and the team had to adjust to a multi-bot architecture to ease the load on each bot, and keep accuracy high.

The Solution

The first thing the team needed to do for this project was refractor the existing connect contact flows so that they were more modular. They were originally interdependent, so if one flow had a bug, the entire flow might experience major repercussions. Next, the team tuned the Lex bots to understand popular US cities, and the cities were store locations were available. The team also had to make sure that all of the data that the customer gave the interactive voice response (IVR) was remembers throughout the call, so the customer wouldn’t have to repeat information they already gave. Lastly, the team also made sure that the bot asked guided conversational questions that would give the customer confidence in responding to the bot without having to take a long pause before answering.