HomeCustomersAirCall
Customer Stories

AirCall’s migration to scalable transcription with Deepgram and AWS

TL;DR

  • AirCall provides over 19,000 businesses with a cloud-based phone system software that optimizes the customer experience and improves remote communication.

  • To process calls, AirCall initially implemented a solution using OpenAI’s Whisper, but they quickly found limitations in their transcription models.

  • Deepgram offered the key features Aircall needed for their platform, including real-time transcription, enhancedaccuracy, faster processing, and cost efficiency.

  • Thanks to the clarity of Deepgram’s documentation, the availability of their applied engineering team, and the intuitiveness of their API, AirCall found the migration from Whisper extremely straightforward and no longer had to commit expensive developer resources to a suboptimal solution.

The Landscape

Since 2014, AirCall has been perfecting the technology that helps businesses communicate with their customers as smoothly and clearly as possible. Specifically, they’ve engineered a cloud-based phone system and call center software designed to facilitate seamless voice communications for businesses. It provides features such as virtual phone numbers, interactive voice response (IVR), and integration with other business tools like CRM systems (Customer Relationship Management).

The Challenges

Challenge 1: Addressing the Need for Scalable, Accurate, Real-Time Transcription

As of January 2025, AirCall processes calls made by over 19,000 businesses each day for customer support, sales, and general communications. Such businesses include Vuori, 1Password, UntuckIt, and Pipedrive.

To aid with this high volume of call processing, AirCall turned to OpenAI’s Whisper—an open-source, speech-to-text AI. They built an in-house transcription solution powered by OpenAI’s models. However, Whisper turned out to be extremely limited. Specifically, it did not meet their needs for accuracy, cost, multilingual capabilities, scalability, and latency.


Most importantly, Whisper could not transcribe calls as they were happening; rather, Whisper could only transcribe pre-recorded audios.

Challenge 2: Migrating from Whisper to Deepgram

The process of migrating a system from one dependency to another can become extremely laborious for engineers. In AirCall’s case, their developers would have to endure migrating from Whisper to another transcription provider. Note: AirCall ended up choosing Deepgram.

Given the simplicity of Deepgram’s API, its seamless handling of large amounts of data, and its team of Applied Engineers ready to troubleshoot at a moment’s notice, AirCall’s move away from Whisper occurred extremely smoothly and quickly. Details on this transition are outlined below.

The Solutions

Taking a Workload off AirCall’s Engineering Team: Deepgram + AWS Marketplace

Due to Whisper’s constraints, AirCall explored new transcription providers that could offer better performance as well as features like entity recognition and custom vocabulary/keyword support. Consequently, Aircall evaluated Deepgram as a potential new provider for transcription during their RFP process.

After a thorough examination, Aircall chose to partner with Deepgram to address their needs in the areas of speech analytics, AI summaries, agent assist/copilot, and AI Voice Agent. 

This new Deepgram-hosted solution took a sizable amount of work off the Aircall’s technical team. Before, AirCall would’ve had to dedicate the entire engineering team to manage and maintain the limited, Whisper-fueled transcription solution. Today, however, the company can simply dedicate a mere ½ - 2 full-time employees to managing the Deepgram-powered solution.

Making Migration as Quick and Painless as Possible

Because of the open-source nature of Whisper, AirCall had no customer support when things went wrong.

On the other hand, Deepgram’s applied engineers remain readily available and extremely familiar with the migration process. Thus, the task of moving batch workloads off of Whisper and on to Deepgram becomes straightforward. And when coupled with the resources of AWS Marketplace, plans to cut traffic over to Deepgram ran even more smoothly. For example, AirCall hosts this solution on their AWS VPC which allows them to maximize control and minimize latency.

Because of Deepgram’s efficiency, clear documentation, and strong technical team, AirCall’s engineers got to focus on roadmap items that are actually differentiators, as opposed to maintaining a solution that (1) isn’t best-in-class, and (2) prohibits pursuing high value use cases due to lack of real-time streaming support.

As a result, AirCall rolled out their new agent assist copilot product which they were unable to do with Whisper.

The Benefits

Integrating Deepgram's Advanced Speech Recognition

AirCall successfully leveraged the AWS Marketplace to streamline procurement and optimize their committed AWS spend by using credits for Deepgram’s transcription solution, which outperforms competitors in speed, accuracy, and cost. Key benefits include:

  • Accuracy: 30% lower word error rate (WER) than competitors on average

  • Speed: Up to 40x faster inference time, supporting real-time applications

  • Cost: 3-5x lower cost than the competition

  • Developer experience: Fast and easy to implement with robust developer resources

  • Future-proof: Comprehensive platform that also includes APIs for voice synthesis, spoken language understanding, and AI voice agents

  • Enterprise-grade: Security, scalability, and reliability from AWS that companies depend on for mission-critical applications

Try Deepgram for free with our API Playground

Test your own audio files or quickly explore its capabilities with our pre-recordings. Try it now for a seamless audio API experience!

Go to API Playground