Customer support · Production

Elastic builds a GenAI proof of concept for automated customer support case summaries and draft replies

The problem

Elastic's support engineers spent significant time creating case summaries for escalations and transitions between engineers, while meeting SLA-driven timely response requirements was resource-intensive.

First attempt

The initial Vertex AI proof of concept produced only a 15.67% positive response rate — much lower than expected — because the LLM was trained on public data and lacked access to Elastic's product documentation and internal knowledge base articles, leaving it unable to address technical support queries accurately.

Workflow diagram · grounded in source
1
CRM button triggers workflow
trigger
“we asked our CRM team to add a custom button on all cases that would call an external endpoint”
2
New case triggers draft reply
trigger
“leveraged an existing automation for all newly created cases and called a new Google Pub/Sub queue to handle all the incoming requests separately”
3
Cloud Function retrieves case text
integration
“The function accepted the Salesforce unique case ID as input and retrieved the case details as text”
4
Vertex AI generates response
ai_action
“The retrieved text would then be automatically sent to Vertex AI combined with the following engineered prompt”
5
AI response posted to Chatter
output
“The AI-generated response was posted to the case via a Salesforce Chatter Post”
6
Engineer reviews AI draft
human_review
“automating a reply for our support engineers to review”
7
Feedback collected via Chatter
feedback_loop
“we could use standard Chatter features for "likes" to identify positive sentiment and threaded responses to capture subjective feedback”
Reported outcome

Elastic moved from proof of concept to an approved project to build a scalable Support AI Chat Assistant, with a plan to integrate Elasticsearch via a retrieval augmented generation architecture for improved response accuracy.

Reported metrics
PoC days open44
AI-generated content items940
Feedback responses collected217
Positive sentiment rate15.67%
Show all 5 reported metrics
PoC days open44
AI-generated content items940
Feedback responses collected217
Positive sentiment rate15.67%
Target accuracy benchmark>80%
Reported stack
Vertex AIGoogle Cloud PlatformSalesforce Service CloudGoogle Cloud FunctionPub/SubSalesforce ChatterElasticsearch
Source
https://www.elastic.co/blog/genai-customer-support-building-proof-of-concept
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Elastic moved from proof of concept to an approved project to build a scalable Support AI Chat Assistant, with a plan to integrate Elasticsearch via a retrieval augmented generation architecture for improved response…

What tools did this team use?

Vertex AI, Google Cloud Platform, Salesforce Service Cloud, Google Cloud Function, Pub/Sub, Salesforce Chatter, Elasticsearch.

What results were reported?

PoC days open: 44; AI-generated content items: 940; Feedback responses collected: 217; Positive sentiment rate: 15.67% (source-reported, not independently verified).

What failed first in this deployment?

The initial Vertex AI proof of concept produced only a 15.67% positive response rate — much lower than expected — because the LLM was trained on public data and lacked access to Elastic's product documentation and int…

How is this customer support AI workflow structured?

CRM button triggers workflow → New case triggers draft reply → Cloud Function retrieves case text → Vertex AI generates response → AI response posted to Chatter → Engineer reviews AI draft → Feedback collected via Chatter.

Related customer support cases