At Essent, we’re always looking for ways to optimize our processes and improve the quality of customer support. We were already using large language models, Generative AI, and conversational AI for our voice and live chat channels. 

Then we had an idea:

What if we could empower a bot with the knowledge we already have to improve our customer support process?

Introducing our award-winning agent assist bot: Bolt! (check this post on LinkedIn: Click here)

With over 1.6 million customer interactions annually, our 700 customer service agents face the daily challenge of quickly finding accurate information in a vast and ever-changing knowledge base of over 1,100 articles.

We already use Guru as our internal knowledge base. With Bolt, we built upon Guru by making its knowledge available through a conversational interface within Microsoft Teams.

Previously, agents spent a significant amount of time manually searching for answers or waiting for helpdesk assistance. This led to delays, repeat calls, and lower customer satisfaction. To tackle these inefficiencies, the Conversational team at Essent developed Bolt: an AI-driven agent-assist bot integrated into Microsoft Teams. 

Bolt leverages real-time updates from Guru, advanced AI models, and intuitive natural language processing to deliver quick, accurate answers. 

Agents interact with Bolt directly in Teams by typing natural language questions. Responses are returned instantly, drawing from the latest knowledge synced via AWS.

Whether providing detailed responses, summaries, or top relevant links, Bolt streamlines workflows and enhances agent performance.

IMPACT AND ARCHITECTURE

The results speak for themselves:

•  85% of helpdesk tasks are automated
•  1+ minute reduction in average handling time
•  20,800 more successful interactions per year
•  Faster onboarding for new agents
•  Higher job satisfaction
•  €1.8 million in annual savings

These gains are driven by a powerful backend system. 

At the heart of Bolt’s operation is an automated knowledge ingestion pipeline built on Amazon Web Services (AWS). 

This pipeline ensures Bolt always has access to the latest articles and documentation from Guru.
We run the sync once daily to capture new or updated content, removing the need for
manual updates. 

The pipeline looks like this:

1.  A Lambda function exports specific Guru collections.
2.  Once the export completes, Guru sends a webhook to an API Gateway endpoint.
3.  This triggers an AWS Step Function that orchestrates the flow.
4.  Files are stored and versioned in Amazon S3 for traceability.
5.  A transformation Lambda converts .yaml and .html files into Cognigy’s .ctxt format.
6.  These are uploaded to the Cognigy Knowledge Store, instantly updating Bolt’s brain.

This integration between AWS and Cognigy forms the backbone of Bolt’s real-time intelligence.

Tech Stack Summary:

• AWS Lambda
• AWS Step Functions
• API Gateway
• Amazon S3
• Cognigy AI Agent
• Guru Knowledge Base

BOLT'S BRAIN: COGNIGY CONFIGURATION

Cognigy is the platform that powers Bolt’s conversational capabilities. We use AI Agent and Knowledge AI to orchestrate flows, integrate with systems, and structure responses.

Bolt is configured with custom tone-of-voice instructions, knowledge source mappings, and logic. For example:

• Ask follow-up questions only when user input is unclear

• Use informal “you” when addressing agents

• Prefer bullet points and step-by-step instructions

Under the hood, Knowledge AI uses Retrieval-Augmented Generation (RAG) to deliver dynamic, context-aware answers pulled from structured content such as manuals and articles.

This reduces the maintenance overhead compared to traditional intent-based bots.
The Knowledge Store further organizes and groups knowledge sources, making content retrieval more efficient and accurate during runtime.

On top of this, we developed Flows in Cognigy, modular node structures that handle things like:

•  Error handling

•  Feedback collection

•  Analytics logging

•  Main onboarding interaction

Agents can choose how they want their answers delivered:

•  A summary with links

•  The full article

•  Only links to relevant content

  
  

WHAT'S NEXT FOR BOLT

Bolt has already made a significant impact, but we’re not stopping here. 

Upcoming features include:

• Support for complex workflows: Handling multi-step use cases more effectively
• Automated email drafting: Generating email responses to common queries to Reduce agent load and ensure consistent messaging

CONCLUSION

Bolt has become an essential part of customer service at Essent. 

With tangible results in efficiency, satisfaction, and cost savings, it showcases how applied AI can transform internal operations.

And we’re just getting started.