AWS Partner Network (APN) Blog
Leveraging Amazon Transcribe and Amazon QuickSight to Extract Business Intelligence from Call Center Data
By Ajay Shah, Solutions Architect – AWS
By Sukhmani Gill, Director, AI/ML – SourceFuse
By Vaidant Singh, CMO – SourceFuse
 |
| SourceFuse |
 |
Regardless of industry, and despite modern online methods of communication, call centers continue to receive an overwhelming number of phone calls.
Many organizations record calls which are potential gold mines of rich insights about customer satisfaction, customer churn, competitive intelligence, service issues, agent performance, and campaign effectiveness.
However, the sheer volume of phone calls exceeds a contact center’s ability to review and analyze them in order to glean those valuable insights. That means only a small fraction of calls, reviewed manually using unsophisticated analysis, are relied upon to see the bigger picture.
In this post, we’ll highlight how a healthcare customer reviewed and analyzed 100,000 calls received by a call center at its peak during the pandemic. Although each call was recorded, the audio files were simply stored and not reviewed for quality or outcomes. High-level details were available—such as quantity and duration—but there was low visibility on call quality, and no insights on resolution of customers inquiries.
The customer engaged SourceFuse as an AWS Advanced Tier Services Partner with the AWS Healthcare Competency to develop a solution that would overcome core challenges:
- High operational costs associated with high call volume.
- Less than satisfactory customer experience and agent performance.
- Inability to measure campaign effectiveness using incoming calls.
We will describe how SourceFuse used custom microservices development to design a solution. Leveraging various AWS services, SourceFuse was able to automatically process and transcribe calls, and provide meaningful business insights via engaging dashboards.
We’ll also demonstrate how to overcome specific challenges relating to speaker recognition, nuanced language, and channel formatting, along with potential next steps to further improve the qualitative analysis.
Developing the Solution
During the initial discovery phase of this project, SourceFuse found that of 106,170 calls reviewed, 20% were < 1 minute long while 30% were dropped or noise-disrupted and only 10% were in English.
The proof of concept (PoC) project involved identifying and proposing solutions to overcome three main challenges:
- Channel: The calls were recorded on a single channel, rather than a stereo file (where the left channel is typically the first speaker and the right channel is the second speaker). In order to provide call center specific transcription, both agent and customer should be recorded in their own channel.
.
SOLUTION: Since it could not split the call recording using channels, SourceFuse implemented the Amazon Transcribe API, which identified who was talking using speaker diarization, and transcribes speech-to-text.
- Language: The customer has diagnostic centers spread across the country; hence, incoming calls were in multiple languages. In these scenarios, the customer either speaks in the local language or may switch between English and a local language.
.
SourceFuse investigated the use of Amazon Translate to transcribe calls in local languages and translate to English. It also explored Amazon Transcribe to support automatic language detection, using a custom language model to detect the most dominant language used. In both cases, the overall non-English call transcription accuracy was low, once language was identified and translated.
.
SOLUTION: For the purposes of the PoC, only calls where the customer and agent conversed in English were manually identified and used in the dataset.
- Speaker: Due to the channel issue, the transcription could only identify speakers as Speaker 1 and Speaker 2. That meant it was impossible to automatically determine which speaker was the call center agent.
.
SOLUTION: At a later stage, SourceFuse would custom-build a speaker classification model to better identify the speakers.
The resulting PoC process was as follows:
Figure 1 – Analysis process.
As well as Amazon Transcribe, SourceFuse implemented Amazon QuickSight for presenting actionable insights and business intelligence (BI) via interactive and appealing dashboards.
SourceFuse also leveraged AWS Lambda for serverless, event-driven computing, without the need to provision or manage the customer’s infrastructure.
Automation of Qualitative Analysis
Using this software, the customer was able to automate the process of checking every single conversation, rather than sampling just 1-2% of calls.
Analysts would no longer have to spend hundreds of hours listening to calls, eliminating the risk of erroneous inputs when manually logging results. Artificial intelligence (AI) does the hard work and identifies any further areas of improvement.
The dashboard enables the quality analyst and management to understand the underlying theme of the calls, topics of conversation, key queries, and view emerging category trends.
Figure 2 – Call volume trends and call topics.
The solution offers the ability to review the overall sentiment of the call along with the sentiment based off the agent and the customer. This helps filter calls where customer interaction was not up to the mark to further identify areas for agent training and help provide a better customer experience.
Figure 3 – Sentiment analysis.
Gain Valuable Market Intelligence
The automated call review pipeline enables the business to review 100% customer conversations to find propositions that offer the most value to customers.
By gaining valuable market intelligence and better insights, businesses can provide enhanced guidance to the agents, leading to more effective conversations. This helps revolutionize engagement rates and minimize churn.
Figure 4 – Market intelligence.
Lower Operational Costs
The solution transcribes 100% of recorded calls to automatically discover and analyze words, phrases, categories, and themes.
With speech analytics, you can:
- Enhance contact center performance with insights to reduce agent handle time and repeat calls.
- Discover customer insights regarding satisfaction, business issues, competitive intelligence, and marketing campaigns.
- Reduce churn by discovering root causes and predicting at-risk customers via your contact center recordings.
- Improve quality monitoring by reviewing large samples and specific call types.
For large contact centers in particular, minimizing wrap time is crucial as it reduces menial work when agents could be talking.
SourceFuse’s solution performs much of the job automatically; for example, by assigning a call category without the need for entering data manually. This helps to eliminate potential for human error from the information collection process.
Building Better Support
With the power of speech analytics, building a better ecosystem for call centers can:
- Enhance customer experience and reduce churn.
- Train and motivate agents.
- Improve compliance and quality monitoring.
With speech analytics, the exact points in each script that are underperforming or causing calls to be dropped can be pinpointed, enabling more effective optimization.
At the same time, the delivery of scripted material can be analyzed, which isn’t normally possible without manual quality assurance. This empowers companies to provide more constructive feedback to call center agents, to get the most beneficial results from pre-prepared content.
Pay Per Usage
Deploying and running applications generally has significant associated costs. However, this solution was designed to have minimum impact on the current infrastructure, and costs now scale on demand.
Improve Customer Experience
While having the ability to transcribe and analyze huge call volumes is powerful, being able to develop an emotional profile of your customers is even more relevant.
With better insights comes more effective conversations, eliminating the need for post-call surveys to monitor customer satisfaction. Knowing what to say and how to say it can revolutionize engagement rates and customer loyalty.
Managing Compliance
Speech analytics can ensure call center agents are meeting your organization’s obligations with regards to regulatory compliance and cold-calling policies. For example, ensuring agents have let the customer know who they are speaking to and why at the beginning of each conversation.
Leaving nothing to chance, speech analytics software can check each and every call for potential breaches.
Conclusion
For this leading diagnostics organization experiencing extremely high call volumes, high operational costs, and wanting to improve customer satisfaction, overcoming these challenges required a unique solution.
SourceFuse was able to custom build a solution and carry out a successful proof of concept, incorporating microservices and AWS tools and services, to automatically analyze and present meaningful business intelligence and actionable insights from qualitative data.
The end result was a bespoke solution that reduced operational costs while enhancing both the call handling agents’ and customers’ experience by providing better support and compliance.
To learn more, contact SourceFuse or read similar use cases.
.
.
SourceFuse – AWS Partner Spotlight
SourceFuse is an AWS Competency Partner that is transforming the way today’s most successful companies develop breakthrough roadmaps leveraging cloud-based technologies.