Conversational AI with Amazon Connect

The Challenge

Our client is a top five major utilities provider, who provide gas and electrical services to the domestic and industrial markets in the UK and also in many other European countries. They had a requirement to provide a virtual assistant to reduce the calls being received by their customer services team and to provide callers with improved self-serve capabilities.

The requirement to Alscient consisted of the following:

  • To provide a virtual assistant which provided the caller with a wide range of services. The call intents were to cover a broad range of areas, including making a payment, account query, a returning contact, appointment query, balance enquiry, meter reading submission, raising a complaint, direct debit query and loss of supply.
  • The solution also needed to provide both agent-assisted and IVR payment collection in a PCI compliant manner.
  • The virtual assistant also needed to offer out a route from the IVR if the caller was struggling to match to an intent or if the caller wanted to speak to an agent.
  • The solution needed to provide Identity and Verification (ID&V) depending on the nature of the enquiry being offered.
  • To provide an operational dashboard which gave an overview of calls made and their enquiry type along with transactional rates of total payments submitted.

The Solution

Alscient addressed these requirements in a phased approach. We initially captured the full set of requirements for the solution and provided a high-level design which described how the solution would work.

Following sign-off of the design, a story map was created which demonstrated the end-to-end caller journey. The solution was based on Amazon Connect, Amazon Lex and supporting services. DTMF input was received through the IVR for card payment details, with the details being encrypted within the call logs and DTMF tone masking taking place on the final call recording. A Lex bot and a series of Lambdas were used to provide the virtual assistant. Hundreds of utterances were matched to each intent and a confidence score was used to assist with intent matching. In all over 25 different intents were created with a redirection being made in the contact flow depending on if the caller had already been through ID&V or not. CloudWatch Dashboards and a series of CloudWatch alarms were used to provide operational metrics on the AWS services which were in use. DynamoDB was used in support of providing different responses based on the enquiry type. S3 was also used to store the call recordings and Polly was used to provide text-to-speech prompts within the call flow. Additionally, Lambda’s were used within the call flows to make callouts to our client’s other systems in response to the intent action.

Following completion of build, the solution was system tested and then migrated across to a UAT environment where our client performed their own testing of the solution. After completion of User Acceptance Testing (UAT) the solution was then promoted to pre-production and then following sign-off through to live.

The Result

The solution has been live since February 2023 and allows customers to self-serve with common enquiries without needing for the request to be serviced by an agent.

The solution has led to a reduction in calls needing to be handled by agents and has improved the overall service offering provided to end customers. Additional staff savings have also been reported due to the lower number of calls that need to be handled by agents.

Our client indicated that this was one of the largest IT projects they undertook last year and were very happy with the speed of delivery and innovation provided by the solution. They are now looking to further extend the solution to their other markets and make use of some of the newer conversational AI capabilities which have since been introduced to Amazon Connect and Amazon Lex.

Have a project you would like to talk with us about?

0113 8000 200 or info@alscient.com