Customer support · Production

Accenture creates a Knowledge Assist solution using generative AI services on AWS

The problem

Employees spend countless hours hunting for information, causing frustration and reduced productivity; when answers cannot be found, escalations or uninformed decisions create organizational risk. Customers find enterprise knowledge bases cumbersome and a poor experience risks defection. Unstructured content makes traditional keyword search time-consuming and ineffective.

First attempt

The existing FAQ bot had a limited scope of questions it could answer, and traditional question-answering platforms had identified gaps that required generative AI to bridge.

Workflow diagram · grounded in source
1
User submits query via web
trigger
“The user submits the request to the conversational front-end application hosted in an Amazon Simple Storage Service (Amazon S3) bucket through Amazon Route 53 and Amazon CloudFront”
2
Amazon Lex routes to orchestrator
routing
“Amazon Lex understands the intent and directs the request to the orchestrator hosted in an AWS Lambda function”
3
Conversation history retrieved
integration
“The function interacts with the application database, which is hosted in a DynamoDB-managed database. The database stores the session ID and user ID for conversation history.”
4
Kendra retrieves relevant context
ai_action
“Another request is sent to the Amazon Kendra index to get the top five relevant search results to build the relevant context. Using this context, modified prompt is constructed required for the LLM model.”
5
Vector similarity search
ai_action
“Content relevance is determined through similarity of raw content embeddings and the user query embedding by using Pinecone vector database embeddings”
6
Claude-2 generates response
ai_action
“A request is posted to the Amazon Bedrock Claude-2 model to get the response from the LLM model selected. The data is post-processed from the LLM response and a response is sent to the user.”
7
Reporting and analytics
feedback_loop
“The CloudWatch log group is configured with a subscription filter that sends logs into Amazon OpenSearch Service. Once available in OpenSearch Service, logs can be used to generate reports and dashboards using Kibana.”
Reported outcome

The Knowledge Assist solution reduces training time for new hires by over 50% and cuts escalations by up to 40%, with dramatic improvements in employee knowledge retention and productivity, and reduced need for call center agent interaction as citizens self-serve.

Reported metrics
New hire training timeover 50%
Escalationsup to 40%
Employee knowledge retention and productivitydramatic improvements
Call center agent interactionlessening the need for call center agent interaction
Reported stack
Amazon BedrockClaude-2Amazon TitanAmazon KendraAmazon LexDynamoDBPineconeAmazon CloudWatchAmazon OpenSearch ServiceKibana
Source
https://aws.amazon.com/blogs/machine-learning/accenture-creates-a-knowledge-assist-solution-using-generative-ai-services-on-aws?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Knowledge Assist solution reduces training time for new hires by over 50% and cuts escalations by up to 40%, with dramatic improvements in employee knowledge retention and productivity, and reduced need for call c…

What tools did this team use?

Amazon Bedrock, Claude-2, Amazon Titan, Amazon Kendra, Amazon Lex, DynamoDB, Pinecone, Amazon CloudWatch, Amazon OpenSearch Service, Kibana.

What results were reported?

New hire training time: over 50%; Escalations: up to 40%; Employee knowledge retention and productivity: dramatic improvements; Call center agent interaction: lessening the need for call center agent interaction (source-reported, not independently verified).

What failed first in this deployment?

The existing FAQ bot had a limited scope of questions it could answer, and traditional question-answering platforms had identified gaps that required generative AI to bridge.

How is this customer support AI workflow structured?

User submits query via web → Amazon Lex routes to orchestrator → Conversation history retrieved → Kendra retrieves relevant context → Vector similarity search → Claude-2 generates response → Reporting and analytics.

Related customer support cases