Home
About
Work
Contact

Case Study 4: Coinbase Customer Experience

AI-Readable Support Workflows

Overview

Coinbase was experiencing rapid growth, and its customer support systems were under pressure. Workflows were scattered across documents, inconsistent in format, and often undocumented. This made it difficult for support agents to resolve issues efficiently and nearly impossible for AI-driven tools to parse and automate them. As Coinbase prepared to roll out automation at scale, the lack of structured workflows posed a significant risk to both agent efficiency and the success of AI adoption.

Problem

  • Customer support was overwhelmed by the lack of reliable, standardized documentation.
  • Without machine-readable workflows, the AI systems that Coinbase wanted to implement could not function effectively, leaving both automation and human agents struggling to provide timely resolutions.

Solution

  • Mapping the jargons: We dug into existing documentation, pieced together tribal knowledge from ops teams, and mapped fragmented workflows in Lucidchart. This gave us a bird’s-eye view of where gaps and inconsistencies existed.
  • Standardizing the structure: We designed a consistent format for workflows that was both human-readable and AI-friendly, breaking down steps into clear, logical decision points.
  • Migration into support articles: Once mapped and standardized, we translated these workflows into knowledge base articles tailored for AI parsing.
  • Validation loop: We partnered closely with product and operations stakeholders to validate each workflow, ensuring accuracy and completeness before publishing.

Project Goals

  • Co-led a team of 12 tasked with turning chaotic, siloed workflows into structured, automation-ready documentation.
  • Reframed Coinbase’s support content into structured, AI-readable documentation, which bridged the gap between human expertise and machine learning.
  • The new support articles became a dual-purpose tool: enabled agents to troubleshoot consistently, while also feeding the AI systems clean, structured data to automate resolution for common queries.
  • Support articles were structured with precision so automated systems could identify triggers, actions, and exceptions.

Rewritten Copy (proposed UX content)

  • Before: “230+ people are attending Gamble in Buy Sell or Trade in Kern County.”

    • After: (Notification suppressed — flagged as spam, not delivered)

       

  • Before: “A member of Duolingo Norwegian Learners created an event: PRIVATDATING.”

    • After: “Your group has a new event. Review details before joining.” (generic until integrity check passes)

  • Before: Repeat notifications for the same event.

    • After: “New event added: [Event Name]. You’ll only get one notification per event.” (deduplication logic reflected in system + copy)

       

  • Help Center (co-host issues):

    • Before: Outdated, incomplete steps.

    • After: “To add a co-host: Go to your event → Select Edit → Choose Co-hosts → Search by name or Page. Note: Co-hosts must accept your invite before they appear.”

 
 
Increase in activation rate
0 %
Increase in activation rate
0 %
Increase in activation rate
0 %

Business Impact

  • Support queries dropped by 5% thanks to improved AI handling.
  • Agents gained access to clear, standardized workflows that cut down time-to-resolution.
  • Coinbase successfully scaled its support function, leveraging automation without sacrificing accuracy.
  • The project laid the foundation for long-term AI integration, future-proofing support documentation as the company grew.

Read more of my case studies

Case Study 1: Facebook Events

Case Study 2: Messenger Help Center

Case Study 3: Instagram Help Center

Got a project? Let's connect!

uxkhwaja@gmail.com

Home
About
Work
Contact