The Challenge

Our client, a mid-sized enterprise operating in the healthcare industry, faced inefficiencies and rising costs due to a reliance on manual transcription call processes. The quality of their prior automated transcription system was low, achieving only a 5% no-touch transcription rate, requiring substantial human intervention to maintain the high-quality results their analytics products demanded. As data volumes grew, they needed a modern to boost automation efficiency without compromising human-level transcription quality.

The Solution

NLP Logix developed a cloud-native service that leveraged multiple deep neural networks that handled the audio transcription process and developed architecture to ensure it can scale elastically as the client’s volume fluctuated.

• Multi-Neural Networks Approach: Multiple neural networks collaborated to transcribe audio, with outputs refined through an ensemble model and post-processing rules.
• Confidence Scoring: High-confidence outputs were processed automatically, while low-confidence results were routed for manual review.
• Scalable Architecture: An event-driven design ensured the system could handle fluctuating workloads seamlessly.
• Drift Monitoring: Periodic human review of high-confidence transcriptions maintained long-term accuracy and generated new training data.

The Results

By adopting this AI-powered solution, the client improved efficiency, reduced costs, and reinforced their reputation for delivering timely, high-quality insights in the healthcare sector.

• The no-touch transcription rate rose from 5% to 68%, reducing reliance on manual intervention.
• The system consistently matched or exceeded human-level transcription quality.
• Reduced manual workloads and faster transcription turnaround times enhanced operational efficiency.
• The flexible architecture supported variable workloads, ensuring consistent performance.

Additional Case Study Details

Introduction

Our client is a mid-sized enterprise operating in the healthcare industry. Their core business focuses on delivering a range of analytical products and services to meet the diverse needs of the healthcare sector.

To maintain their competitive edge, our client recognized the need to modernize their approach to audio transcription. They sought a solution that could significantly boost efficiency without compromising the superior quality their customers had come to expect. They also wanted a solution that was flexible enough to be able to easily integrate new transcription technologies as they become available.

NLP Logix was engaged to develop an innovative AI-powered solution to address these challenges. The project aimed to dramatically increase the no-touch transcription rate from 5% to 68%, while maintaining or exceeding human-level transcription quality. This case study explores how NLP Logix developed cutting-edge AI technology to not only meet but exceed the client’s expectations, resulting in a transformative impact on their operations and cost structure.

Goals and Objectives

• Increase the no-touch transcription rate from 5% to 68%
• Decrease the amount of time required for transcriptionists to complete a file
• Maintain or exceed human-level transcription quality

Requirements

To provide accurate analytics and alert their customers, our client required a very high level of accuracy from their audio transcription provider. All low-confidence automated transcriptions needed to be reviewed by a human and corrected. To handle the correction process, our client had a large workforce listening to recorded audio files and manually transcribing them into raw text that was stored in a database for further analytic processing.

The quality of the prior system’s automated transcriptions was low, and as a result, the overall system was inefficient. The client was paying transaction fees to the transcription software as well as paying for the labor to do manual transcription. With both cost components on the rise, our client needed to find a new, more efficient approach.

Solution

The solution outlined below was developed to achieve the client’s goals and objectives.

Algorithm

The solution NLP Logix developed was a cloud-native service that leveraged multiple deep neural networks that handled the audio transcription process.

The text output from each of the networks was merged by an ensemble model into the final text, and post-processing rules that handled domain-specific edge cases were then applied.

The text was given an overall confidence score, representing the likelihood the transcription was correct based on individual transcription confidences and inter-model agreement. If this score was high enough, the text would be automatically processed without intervention from a human processor. If the transcribed text was too low for automated processing, the call audio, along with the text, was routed to a human for manual review and correction.

Scalability

Besides the core transcription algorithm, NLP Logix designed, developed, and tested the architecture to ensure this service can scale elastically as our client’s volume fluctuated.

To achieve this, the algorithm is hosted within an event driven architecture that allows additional instances of the algorithm to be scaled up and down automatically based on the call volume.

Drift Monitoring

To ensure the end-to-end process continues to perform at acceptable levels, NLP Logix designed a health monitoring process where the existing human workforce would be routed a small set of sampled audio files for manual transcription, even though they had a high confidence score.

This was done so the overall system could be monitored for drift and to ensure additional training data for future model iterations could be collected along the way.

Outcome

The implementation of NLP Logix’s AI-powered audio transcription solution yielded significant improvements for our client. By leveraging multiple deep neural networks, an ensemble model, and post-processing rules, the system dramatically increased the no-touch transcription rate from the initial 5% to exceeding the 30% target to 68%. This substantial boost in automation efficiency meant that a much larger portion of audio files could be processed without human intervention, leading to faster turnaround times and reduced labor costs for the client.

The solution maintained or exceeded human-level transcription quality, meeting one of the primary objectives of the project. The confidence scoring system ensured that only high-quality automated transcriptions were processed without review, while lower-confidence transcriptions were still routed to human transcriptionists for manual correction. This approach, combined with the drift monitoring process, allowed the client to maintain the superior quality their customers expected while significantly reducing the workload on their human workforce.

As a result, the time required for transcriptionists to complete each file decreased, further enhancing operational efficiency. The client realized substantial value through reduced transaction fees for automated transcriptions and decreased labor costs, all while improving their ability to deliver timely, accurate insights to their healthcare sector customers.

View more of our case studies to see how our team continues to push automation and technology efficiencies. Contact us to learn more on how we can help your business.