Customer support · Production

Coinbase builds LLM-powered Conversational Coinbase Chatbot (CBCB) to handle tens of thousands of customer support queries monthly

The problem

Coinbase's customer query volume surged to tens of thousands per month, with traffic spiking during crypto bull runs, requiring a system that could personalize responses using real-time account data while maintaining compliance.

First attempt

Standard commercial and open-source LLMs lacked the Coinbase-specific context needed to accurately address account restrictions, platform policies, and unique product features.

Workflow diagram · grounded in source
1
Customer query arrives
trigger
“customer queries often cover a variety of topics like account restrictions, platform policies, recent transactions, and unique Coinbase product features”
2
Query rephrasing
ai_action
“Rephraser: Refines customer queries to better match our knowledge base, ensuring accurate interpretation by the chatbot.”
3
Article retrieval and ranking
ai_action
“Article Retriever: Bases responses on relevant help content by dynamically retrieving information from knowledge bases using multiple semantic indices for improved accuracy. Ranks the retrieved articles for optimal relevance to the user'…”
4
Domain-specific logic and account data
ai_action
“various real-time APIs that reflect actual account status and transactions history of the customer”
5
Response styling
ai_action
“Response Styler: Guarantees that responses meet conversational standards, including appropriate tone, clarity, and style.”
6
Guardrails compliance check
validation
“Guardrails: Ensures compliance with legal, security, and privacy standards by enforcing strict input and output protocols.”
7
Personalized response delivered
output
“the system delivers personalized, accurate, and compliant responses”
Reported outcome

CBCB handles tens of thousands of customer support queries monthly, delivering faster answers without live agent wait times and freeing CX agents to focus on more complex issues.

Reported metrics
Customer support queries handled monthlytens of thousands
Customer response timefaster answers without needing to wait for a live agent
CX agent focus on complex issuesautomating routine inquiries allows our customer experience (CX) agents to focus on more complex and impactful issues
Reported stack
LLMML modelknowledge bases
Source
https://www.coinbase.com/en-ar/blog/behind-the-scenes-of-the-conversational-coinbase-chatbot
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

CBCB handles tens of thousands of customer support queries monthly, delivering faster answers without live agent wait times and freeing CX agents to focus on more complex issues.

What tools did this team use?

LLM, ML model, knowledge bases.

What results were reported?

Customer support queries handled monthly: tens of thousands; Customer response time: faster answers without needing to wait for a live agent; CX agent focus on complex issues: automating routine inquiries allows our customer experience (CX) agents to focus on more complex and impactful issues (source-reported, not independently verified).

What failed first in this deployment?

Standard commercial and open-source LLMs lacked the Coinbase-specific context needed to accurately address account restrictions, platform policies, and unique product features.

How is this customer support AI workflow structured?

Customer query arrives → Query rephrasing → Article retrieval and ranking → Domain-specific logic and account data → Response styling → Guardrails compliance check → Personalized response delivered.

Related customer support cases