P-018 · Case study
SHIPPEDAdvisor AI Agent
GoReach's second AI product: an agent financial advisors use across their whole client book. Launched November 2026.
Case study
The problem
GoReach's advisors had AI help on separate screens, and nothing that could work across a whole client book. Checking which clients needed attention, or what a portfolio's numbers meant, took manual digging. In a regulated business, one wrong figure in front of an advisor is a real risk.
The approach
A separate service, so it could ship and roll back without touching the customer assistant. Answers are built from the platform's own data rather than the model's memory, and each capability has to pass its own evals before it's switched on. Advisors stay in control: drafts land in their usual composer, and nothing reaches a client without them.
What worked
Launched in November 2026 with four capabilities: portfolio explanations, conversation briefings, reply drafts and questions across the book. It rolls out advisor by advisor.
What I'd do differently
Settle the boundaries with advisors first: which questions the agent answers, which it hands back to them, and how it reports data it couldn't reach. Those decisions took more of the design work than anything about the model.
More detail
Advisors ask about one client or every client at once. The agent explains a client's portfolio, briefs the advisor before a conversation, answers questions across the book, and drafts replies for the advisor to review and send. Every figure in an answer comes from the platform's own data, and when something is out of reach it says so instead of guessing. I designed it and led the build across the backend, the advisor workspace and the customer assistant.