Proving the Business Case for the Internet of Things

AWS machine learning improves contact centres

Steve Rogerson
July 29, 2020

Amazon Web Services (AWS) has added machine-learning capabilities to its contact centre service making it easier for businesses to identify customer issues and trends, search call transcripts, and improve agent performance.
Called Contact Lens, it is a set of capabilities for Amazon Connect enabled by machine learning, which gives contact centres the ability to understand the sentiment, trends and compliance of customer conversations to improve their experience and identify crucial feedback.
Amazon Connect is a fully managed cloud contact centre service that helps companies of any size improve customer service at lower cost and is based on the same technology that powers Amazon’s customer service.
Companies such as John Hancock, Capital One, Intuit, GE Appliances, Square, Fujitsu, Mutual of Omaha and Dow Jones use Amazon Connect to run their contact centres, scaling to thousands of agents.
With Contact Lens, contact centre supervisors can discover emerging themes and trends from customer conversations, conduct fast, full-text search on call transcripts to troubleshoot customer issues, and improve agents’ performance with call analytics, all from within the Amazon Connect console.
Coming later this year, Contact Lens will also provide the ability for supervisors to be alerted to issues during in-progress calls, giving them the ability to intervene earlier when a customer is having a poor experience. Contact Lens requires no technical expertise, and getting started takes a few clicks in Amazon Connect.
Contact centres are often the only personal connection that a customer has with a company, and the experiences these customers have interacting with agents can have a profound impact on customer trust and loyalty. Contact centres field large volumes of customer calls every day, resulting in millions of hours of recorded calls. These conversations contain valuable customer feedback, but given the volume, companies struggle to extract and analyse this information in a timely fashion, if at all.
Most companies that try to get value from these data use legacy contact centre analytics offerings, but these technologies can be expensive, slow at providing call transcripts, and lack the required level of transcription accuracy, all of which makes it difficult to detect customer issues quickly and provide precise feedback to customer service agents and supervisors.
Existing contact centre technology also lacks the ability to provide real-time analytics on in-progress calls, which prevents supervisors from identifying and helping frustrated customers before they hang up. As a result of these challenges, many organisations face high levels of customer churn, long hold times, agent turnover and regulatory fines.
Contact Lens helps contact centre users address these problems by providing fully managed, machine-learning-powered analytics capabilities within Amazon Connect, with no coding or machine-learning experience required. It uses accurate machine-learning technology to transcribe calls, and automatically indexes call transcripts so they can be searched from the Amazon Connect console.
Machine learning also makes it easier for supervisors to search voice interactions based on call content such as customers asking to cancel a subscription or return an item, customer sentiment such as calls that ended with a negative customer sentiment score, and conversation characteristics such as talk speed, long pauses, or customers and agents talking over one another.
By clicking on the search results, supervisors can view a contact detail page to see the call transcript, customer and agent sentiment, and a visual illustration of conversation characteristics, and use this information to share feedback with their agents to improve customer interactions.
Contact Lens also uses natural-language processing to help supervisors uncover new issues such as a price discrepancy between a web site and an email promotion on the contact detail page by visually identifying words and phrases in call transcripts that indicate reasons for customer outreach.
Supervisors can automatically monitor all their agents’ interactions for customer experience, regulatory compliance and adherence to script guidelines by defining custom categories on a new page in Amazon Connect that allows them to organise customer contacts based on words or phrases said by the customer or agent, for example a customer mentioning a competitor, membership in a customer loyalty programme, certain regulatory disclosures and so on.
The machine-learning capabilities can automatically detect and redact sensitive personally identifiable information such as names, addresses and social security numbers from call recordings and transcripts to help customers more easily protect customer data.
Later this year, Contact Lens will introduce features that provide supervisors with real-time assistance by offering a dashboard that shows the sentiment progression of live calls in a contact centre. This dashboard continuously updates as the interactions evolve and allows supervisors to look across live calls to spot opportunities to help their customers. Real-time alerting gives supervisors the ability to engage and de-escalate situations earlier.
“Amazon Connect has grown very quickly in its first few years as customers find it very attractive to use the same contact centre technology along with the high scale, strong performance, low cost and embedded AI that Amazon has used to scale in its first 25 years,” said Larry Augustin, vice president of AWS. “Contact Lens leverages various AWS capabilities – such as storage, transcription, natural language processing and search – but stitches them together for customers into an easy-to-use contact analysis tool, all usable from the Amazon Connect user interface and with no machine learning or heavy programming required.”
Contact Lens capabilities are built right into the Amazon Connect experience. Contact Lens provides metadata such as transcriptions, sentiment and categorisation tags in customers' Amazon Simple Storage Service (S3) buckets in a well-defined schema. Businesses can export this information and use additional tools such as Amazon QuickSight or Tableau to do further analysis and combine it with data from other sources.
One user is Traeger Pellet Grills, a privately held American manufacturer of wood-fired pellet grills and related accessories.
“When we decided to migrate hundreds of agents from our incumbent telephony platform, we needed something that was simple, scalable and open,” said Bryan Teggart, head of customer experience at Traeger Pellet Grills. “We wanted something future proof. Amazon Connect was the solution. We were able to replicate and enhance our existing infrastructure within days of creating our AWS accounts. For us, one of the most powerful aspects of Amazon Connect is Contact Lens. Contact Lens allows us to simplify agent workflows by systemically identifying key contact attributes, for example product model, contact reason, customer sentiment. Those attributes are invaluable to our CX leadership and product and engineering teams. Last month, we analysed more than 15,000 hours of agent-customer interactions. We used to spend most of our time trying to identify issues. Now, we spend most of our time fixing them.”