← Back to the library
Automated customer support ticket resolution
Problem
This support team's 85,000+ resolved tickets held the answer to most new issues that came in, but that knowledge was fragmented across Freshdesk and legacy NetSuite and Salesforce data, with no practical way for customer support agents to search across all of it. They spent significant time researching issues a colleague had likely already solved. The team needed to put that prior work in front of support agents the moment a new ticket arrived, so they could reach the root cause faster and deliver more complete resolutions.
Solution
A custom application that ingests every new Freshdesk ticket, runs a semantic search across 85,000+ historical tickets, and uses an AI model to generate a recommended next action delivered to the customer support agent as a private note in Freshdesk—including citations to the historical tickets it drew from. Early estimates show a 10% efficiency improvement, equivalent to roughly 1,000 working hours saved each year.
At a glance
Vertical
Logistics
Oil & Gas
Manufacturing
Retail and Sales
Distribution
Energy
Construction
Real Estate
Agriculture
Hospitality
Used By
Operations
Finance
Supply Chain
Building Blocks
_icon-data-ingestion.svg){kind=icon}
Data Ingestion
_icon-ai-extraction.svg){kind=icon}
AI Extraction
_icon-logic-and-validation.svg){kind=icon}
Logic & Validation
Human-in-the-Loop
_icon-custom-applications.svg){kind=icon}
Custom Applications
Building Blocks
This solution was built using these Nextworld features
Data Ingestion
Developers built the Support Tickets application as a header application with two related detail applications—one for the conversation messages on the incoming ticket, and one for the similar historical tickets returned by semantic search. The header-and-detail structure was necessary because each ticket has many messages and many similar-ticket matches associated with it.
Data Ingestion
Webhooks automatically send ticket data from Freshdesk to Nextworld whenever a ticket is created, updated, or closed.
Data Ingestion
The application uses several logic blocks across the process. The primary one writes incoming ticket data to the header application, runs the semantic search against historical tickets, scores similarity, and validates the AI's recommended next action before sending it back to Freshdesk. Supporting logic blocks generate AI summaries of closed tickets so they can be searched against future tickets, and filter out spam or out-of-office replies before they trigger the main process.
Data Ingestion
AI models and prompt definitions handle the language work: summarizing closed tickets so they can be searched, and generating the recommended next action delivered back to the support agent.
Data Ingestion
Semantic search finds the closed historical tickets most relevant to the current one. Unlike keyword search, it matches on meaning rather than exact wording. The search runs against ticket summaries stored as vector embeddings in the Nextworld database, scanning more than 85,000 tickets and returning the top 100 candidates scored as same, similar, or not similar.
