Intact Financial accelerates the call auditing process by 15x with Amazon Transcribe

As the largest property and casualty insurer in Canada, Intact’s customer service calls contained a wealth of untapped data. The company’s vision was to apply AI to its call auditing process and automatically derive data-driven insights that would improve customer experience through faster, more effective service. Intact planned to use these insights to refine customer service scripts, provide coaching opportunities for agents, and improve processes around the handling of calls throughout different business lines. It could significantly improve agent efficiency by off-loading repetitive issues to a chatbot and automatically notifying a supervisor when a challenging interaction required assistance. Additionally, Intact hoped to leverage the data to analyze customer feedback, identify potential product issues or service gaps, and uncover new business opportunities.

The Intact Lab sought to build the foundation of CQ around a speech recognition service that would automatically convert all the customer calls into text for further analysis. However, most options in the market couldn’t support Canadian French. After evaluating several alternatives, the team chose Amazon Transcribe, an automatic speech recognition (ASR) service that makes it simple for developers to add speech-to-text capabilities to their applications. “We scanned the market, properly evaluating and testing the different providers out there,” says Etienne Brouillard, AI architect specialist in the Intact Lab. “We found Amazon Transcribe to be 10-15% more accurate in handling both English and Canadian French calls compared to other ASR vendors.” With Amazon Transcribe at the core of CQ, Intact could accelerate the call auditing process to gain business intelligence across a larger swath of customer calls.

Working alongside an AWS solutions architect, Data Lab built a secure, serverless architecture for CQ. First, CQ takes a recorded call stored in the Amazon S3 bucket and sends it to Amazon Transcribe to convert into text. Intact then runs various ML models to identify call components, conduct sentiment analysis, and quickly determine call intent, outcome, and whether further review is needed. Intact can now boost transcription accuracy with Amazon Transcribe’s automated language identification and multilingual transcription features which accurately transcribe the text in the language used even as the speaker switches between languages. In 2023, Intact used Amazon Transcribe’s personal identifiable information (PII) redaction feature to ensure data privacy by automatically detecting and removing sensitive data such as customer names and credit card information. They also benefit from AI-powered summarization that boosts agent productivity by reducing agent after call work.

The CQ dashboard visualizes call quality trends, metrics, and actionable business insights. The user interface also provides a powerful search tool that allows quality analysts to quickly find and analyze calls based on specific metrics or components, such as introductions, questions, quotes, and topics. For example, Intact analysts can dissect calls to identify where agents spend most of their time, the amount of silent time during calls, and average handling time—metrics that provide opportunities for data-driven agent coaching. “We can devise new mentoring and coaching strategies to improve scripts and train new agents,” says Brouillard. “That means less time spent on the phone for our customers, which is a good thing.”

Furthermore, Intact created an internal meter to allow them to predict transcription costs based on per minute charges. This includes variable pricing for usage at varying thresholds. The solution helps Intact to forecast expenses reliably, predicting costs down to the minute.

Intact continues to deploy the CQ suite across an increasing number of business units and use cases. “With Amazon Transcribe, we are able to use the same infrastructure, the same solution, and the same functions to deploy the product in different contexts with different AI models based on the same call transcript foundation,” says Brouillard. “We can evaluate calls for claim agents, underwriters, brokers and even other affiliated companies of Intact. Although these business lines have specific models and use varying on-premises or cloud-based call center solutions, all have integrated the CQ suite seamlessly. CQ has scaled to handle 800 calls simultaneously and has processes 20,000 calls within 4–6 hours, an improvement in speed of 15x.” Intact also experienced a lower word error rate in English with about a 6% improvement year over year.

Quality auditors are now 65% more efficient, and the agent coaching insights garnered from the calls have reduced agents’ call handling time by 10%.

As a next step, the company aims to process calls using Amazon Transcribe streaming, which transcribes audio in real time. Based on the conversation, a virtual agent will deliver relevant information and recommended answers to the agent as needed. Additionally, Intact plans to use generative AI to further improve agents’ interactions with customers.

“Working alongside AWS, we have always had experts and tools within reach so that we can plan and design with ease,” says Brouillard. “We are confidently building secure AI solutions on AWS today to improve our call quality and continue raising the bar on customer experience.”