Intact Monetary Company is the main supplier of property and casualty insurance coverage in Canada, a number one supplier of worldwide specialty insurance coverage, and a pacesetter in industrial strains within the UK and Eire. Intact confronted a problem in managing its huge community of buyer help name facilities and required a workable resolution inside 6 months and long-term resolution inside 1 yr. With as much as 20,000 calls per day, the guide auditing course of was inefficient and struggled to maintain up with growing name visitors and rising customer support expectations. High quality management brokers needed to manually decide calls to audit, which was not a scalable resolution. To deal with this, Intact turned to AI and speech-to-text know-how to unlock insights from calls and enhance customer support. The corporate developed an automatic resolution referred to as Name High quality (CQ) utilizing AI providers from Amazon Internet Companies (AWS). The implementation of CQ allowed Intact to deal with 1,500% extra calls (15 occasions extra calls per auditor), scale back agent dealing with time by 10%, and generate priceless insights about agent conduct, resulting in improved customer support.
Amazon Transcribe is a totally managed computerized speech recognition (ASR) service that helps builders add speech-to-text capabilities to functions. It makes use of deep studying to transform audio to textual content shortly and precisely. On this put up, we reveal how the CQ resolution used Amazon Transcribe and different AWS providers to enhance essential KPIs with AI-powered contact heart name auditing and analytics.
This allowed Intact to transcribe buyer calls precisely, prepare customized language fashions, simplify the decision auditing course of, and extract priceless buyer insights extra effectively.
Resolution overview
Intact aimed to develop an economical and environment friendly name analytics platform for his or her contact facilities by utilizing speech-to-text and machine studying applied sciences. The purpose was to refine customer support scripts, present teaching alternatives for brokers, and enhance name dealing with processes. By doing so, Intact hoped to enhance agent effectivity, establish enterprise alternatives, and analyze buyer satisfaction, potential product points, and coaching gaps. The next determine exhibits the structure of the answer, which is described within the following sections.
Intact chosen Amazon Transcribe as their speech-to-text AI resolution for its accuracy in dealing with each English and Canadian French. This was a key think about Intact’s choice, as a result of the corporate sought a flexible platform able to adapting to their various enterprise wants. Amazon Transcribe provides deep studying capabilities, which might deal with a variety of speech and acoustic traits, along with its scalability to course of anyplace from a number of hundred to over tens of 1000’s of calls day by day, additionally performed a pivotal function. Moreover, Intact was impressed that Amazon Transcribe might adapt to numerous post-call analytics use instances throughout their group.
Name processing and mannequin serving
Intact has on-premises contact facilities and cloud contact facilities, so that they constructed a name acquisition course of to ingest calls from each sources. The structure incorporates a totally automated workflow, powered by Amazon EventBridge, which triggers an AWS Step Capabilities workflow when an audio file is uploaded to a chosen Amazon Easy Storage Service (Amazon S3) bucket. This serverless processing pipeline is constructed round Amazon Transcribe, which processes the decision recordings and converts them from speech to textual content. Notifications of processed transcriptions are despatched to an Amazon Easy Queue Service (Amazon SQS) queue, which aids in decoupling the structure and resuming the Step Capabilities state machine workflow. AWS Lambda is used on this structure as a transcription processor to retailer the processed transcriptions into an Amazon OpenSearch Service desk.
The decision processing workflow makes use of customized machine studying (ML) fashions constructed by Intact that run on Amazon Fargate and Amazon Elastic Compute Cloud (Amazon EC2). The transcriptions in OpenSearch are then additional enriched with these customized ML fashions to carry out elements identification and supply priceless insights reminiscent of named entity recognition, speaker function identification, sentiment evaluation, and personally identifiable data (PII) redaction. Common enhancements on current and new fashions added priceless insights to be extracted reminiscent of cause for name, script adherence, name final result, and sentiment evaluation throughout numerous enterprise departments from claims to non-public strains. Amazon DynamoDB is used on this structure to manage the boundaries of the queues. The decision transcriptions are then compressed from WAV to an MP3 format to optimize storage prices on Amazon S3.
Machine studying operations (MLOps)
Intact additionally constructed an automatic MLOps pipeline that use Step Capabilities, Lambda, and Amazon S3. This pipeline gives self-serving capabilities for knowledge scientists to trace ML experiments and push new fashions to an S3 bucket. It provides flexibility for knowledge scientists to conduct shadow deployments and capability planning, enabling them to seamlessly change between fashions for each manufacturing and experimentation functions. Moreover, the appliance provides backend dashboards tailor-made to MLOps functionalities, guaranteeing easy monitoring and optimization of machine studying fashions.
Frontend and API
The CQ software provides a sturdy search interface specifically crafted for name high quality brokers, equipping them with highly effective auditing capabilities for name evaluation. The appliance’s backend is powered by Amazon OpenSearch Service for the search performance. The appliance additionally makes use of Amazon Cognito to supply single sign-on for safe entry. Lastly, Lambda capabilities are used for orchestration to fetch dynamic content material from OpenSearch.
The appliance provides pattern dashboards custom-made to ship actionable enterprise insights, aiding in figuring out key areas the place brokers allocate their time. Utilizing knowledge from sources like Amazon S3 and Snowflake, Intact builds complete enterprise intelligence dashboards showcasing key efficiency metrics reminiscent of durations of silence and name deal with time. This functionality allows name high quality brokers to delve deeper into name elements, facilitating focused agent teaching alternatives.
Name High quality Pattern Dashboard
The next determine is an instance of the Name High quality Pattern Dashboard, exhibiting the data obtainable to brokers. This contains the flexibility to filter on a number of standards together with Dates and Languages, Common Deal with Time per Parts and Unit Managers, and Speech time vs. Silence Time.
Outcomes
The implementation of the brand new system has led to a major enhance in effectivity and productiveness. There was a 1,500% enhance in auditing velocity and a 1,500% enhance within the variety of calls reviewed. Moreover, by constructing the MLOps on AWS alongside the CQ resolution, the staff has decreased the supply of latest ML fashions for offering analytics from days to mere hours, making auditors 65% extra environment friendly. This has additionally resulted in a ten% discount in brokers’ time per name and a ten% discount of common maintain time as they obtain focused teaching to enhance their buyer conversations. This effectivity has allowed for simpler use of auditors’ time in devising teaching methods, enhancing scripts, and agent coaching.
Moreover, the answer has supplied intangible advantages reminiscent of extraordinarily excessive availability with no main downtime since 2020 and high-cost predictability. The answer’s modular design has additionally led to strong deployments, which considerably decreased the time for brand spanking new releases to lower than an hour. This has additionally contributed to a near-zero failure fee throughout deployment.
Conclusion
In conclusion, Intact Monetary Company’s implementation of the CQ, powered by AWS AI providers has revolutionized their customer support method. This case research serves as a testomony to the transformative energy of AI and speech-to-text know-how in enhancing customer support effectivity and effectiveness. The answer’s design and capabilities place Intact properly to make use of generative AI for future transcription tasks. As subsequent steps, Intact plans to additional use this know-how by processing calls utilizing Amazon Transcribe streaming for real-time transcription and deploying a digital agent to supply human brokers with related data and beneficial responses.
The journey of Intact Monetary Company is one instance of how embracing AI can result in vital enhancements in service supply and buyer satisfaction. For purchasers trying to shortly get began on their name analytics journey, discover Amazon Transcribe Name Analytics for reside name analytics and agent help and put up name analytics.
Concerning the Authors
Étienne Brouillard is an AWS AI Principal Architect at Intact Monetary Company, Canada’s largest supplier of property and casualty insurance coverage.
Ami Dani is a Senior Technical Program Supervisor at AWS specializing in AI/ML providers. Throughout her profession, she has targeted on delivering transformative software program improvement tasks for the federal authorities and enormous corporations in industries as various as promoting, leisure, and finance. Ami has expertise driving enterprise development, implementing revolutionary coaching packages and efficiently managing advanced, high-impact tasks.
Prabir Sekhri is a Senior Options Architect at AWS within the enterprise monetary providers sector. Throughout his profession, he has targeted on digital transformation tasks inside giant corporations in industries as various as finance, multimedia, telecommunications in addition to the power and fuel sectors. His background contains DevOps, safety, and designing and architecting enterprise storage options. Moreover know-how, Prabir has at all times been captivated with enjoying music. He leads a jazz ensemble in Montreal as a pianist, composer and arranger.
