Customer support · Production

GoDaddy's Lighthouse platform uses LLMs to analyze 100,000+ daily customer service transcripts and surface issues in hours instead of weeks

The problem

GoDaddy processes up to 100,000 English-language customer service transcripts daily, but manual review by supervisors and analysts could not scale to extract company-wide intelligence or detect emerging issues quickly enough to influence daily operations.

First attempt

Manual review and sampling approaches failed to surface systemic issues, and early attempts at automated analysis using ad hoc prompts produced inconsistent results across similar conversations.

Workflow diagram · grounded in source
1
Customer interactions transcribed
trigger
“The recorded interactions get converted into transcripts - a rich source of information that holds valuable insights into customer pain points, agent effectiveness, and opportunities for improvement or growth”
2
Lexical pre-filtering via OpenSearch
routing
“Filter transcripts by geography, date, and product metadata via OpenSearch”
3
LLM prompt-driven analysis
ai_action
“Pass filtered set to a domain-specific prompt in Lighthouse”
4
Structured insight generation
output
“Generate structured insights (e.g., sentiment distribution, causal drivers)”
5
Dashboard delivery via QuickSight
integration
“QuickSight provides business users direct access to insights through familiar dashboard interfaces”
6
Continuous prompt quality evaluation
feedback_loop
“continuous prompt evaluation through regression-style benchmarking detects drift in output quality using automated tests against curated transcript sets”
Reported outcome

Lighthouse processes the full daily volume of 100,000+ transcripts in approximately 80 minutes and surfaces emerging issues in hours instead of weeks, enabling GoDaddy to catch and resolve problems before they escalate into significant service disruptions.

Reported metrics
Daily transcripts processed100,000+
Daily volume processing timeapproximately 80 minutes
Issue detection latencyhours instead of weeks
Customer call reductionreduced further customer calls
Reported stack
Lighthouselarge language models (LLMs)Claude v2GoCaaSOpenSearchAWS LambdaS3Amazon QuickSight
Source
https://www.godaddy.com/resources/news/harnessing-ai-to-navigate-millions-of-customer-conversations-at-godaddy
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Lighthouse processes the full daily volume of 100,000+ transcripts in approximately 80 minutes and surfaces emerging issues in hours instead of weeks, enabling GoDaddy to catch and resolve problems before they escalat…

What tools did this team use?

Lighthouse, large language models (LLMs), Claude v2, GoCaaS, OpenSearch, AWS Lambda, S3, Amazon QuickSight.

What results were reported?

Daily transcripts processed: 100,000+; Daily volume processing time: approximately 80 minutes; Issue detection latency: hours instead of weeks; Customer call reduction: reduced further customer calls (source-reported, not independently verified).

What failed first in this deployment?

Manual review and sampling approaches failed to surface systemic issues, and early attempts at automated analysis using ad hoc prompts produced inconsistent results across similar conversations.

How is this customer support AI workflow structured?

Customer interactions transcribed → Lexical pre-filtering via OpenSearch → LLM prompt-driven analysis → Structured insight generation → Dashboard delivery via QuickSight → Continuous prompt quality evaluation.

Related customer support cases