Ticket triage · Production

Adyen builds LLM-powered smart ticket routing and support agent copilot with LangChain

The problem

Growing merchant volume and transaction load put rising pressure on Adyen's support teams, and ticket hand-offs between teams were a primary driver of slower response times. The team wanted to scale support capacity through technology without growing headcount.

Workflow diagram · grounded in source
1
Merchant submits support ticket
Trigger
A support ticket enters the routing system to be directed to the right support person based on its content.
source quote
“A smart ticket routing system designed to get a ticket to the right support person as quickly as possible based on content”
2
Theme and sentiment analysis
Ai action
An internal tool analyzes the theme and sentiment of each ticket and dynamically updates its priority based on the user.
source quote
“an internal tool that analyzes the theme and sentiment of each ticket, and dynamically updates its priority based on the user”
3
Route to suited technical expert
Routing
The LLM-driven approach enables merchants to receive support from the technical experts most suited to respond quickly.
source quote
“this LLM-driven approach enables merchants to receive support from the technical experts most suited to respond quickly”
4
Document retrieval from vector database
Ai action
The system finds the most relevant and up-to-date document from a collection of public and private documents stored in a vector database.
source quote
“store them in a vector database with an embedding model that optimized for effective retrieval. The team's first milestone on its way to generating proposed ticket responses was finding the most relevant and up-to-date document from a collection of public …”
5
Copilot generates suggested response
Ai action
An LLM produces a suggested response for support agents through their proprietary copilot.
source quote
“connect to an LLM to produce a suggested response for support agents through their proprietary copilot”
6
Agent reviews and applies suggestion
Human review
Support agents work from their queues with easily-modifiable potential answers to customer inquiries at their fingertips.
source quote
“With the right set of tickets in their queues and easily-modifiable potential answers to customer inquiries at their fingertips, support agents are more efficient and more satisfied”
Reported outcome

Adyen's LLM-driven ticket routing and copilot made support agents more efficient and satisfied, with document retrieval far outperforming traditional keyword-based search and immediately establishing team trust in the new system.

Reported metrics
Support agent efficiency and satisfactionmore efficient and more satisfied
Retrieval vs keyword-based searchfar outperformed traditional keyword-based search
Time to build document collection4 months
Reported stack
LangChainLangSmithvector databaseembedding modelKubernetes
◆ Does this fit your context?

Compare to your context

Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.

Compare to your context →
~30 seconds · free
Source
https://blog.langchain.dev/llms-accelerate-adyens-support-team-through-smart-ticket-routing-and-support-agent-copilot/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Adyen's LLM-driven ticket routing and copilot made support agents more efficient and satisfied, with document retrieval far outperforming traditional keyword-based search and immediately establishing team trust in the…

What tools did this team use?

LangChain, LangSmith, vector database, embedding model, Kubernetes.

What results were reported?

Support agent efficiency and satisfaction: more efficient and more satisfied; Retrieval vs keyword-based search: far outperformed traditional keyword-based search; Time to build document collection: 4 months (source-reported, not independently verified).

How is this ticket triage AI workflow structured?

Merchant submits support ticket → Theme and sentiment analysis → Route to suited technical expert → Document retrieval from vector database → Copilot generates suggested response → Agent reviews and applies suggestion.

WHAT TO DO WITH THIS

Now compare it to your context

This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.