Confessions of a Product Specialist at Third Bridge
When we first integrated Third Bridge’s expert insights into LLMs like Claude or ChatGPT, I spent hours doing something most analysts wouldn't admit to: I was failing.
I was treating the AI like a sophisticated Google search. I’d ask, “What are the risks for [Ticker]?” or “Summarize this transcript.” The results were... fine. They were accurate, grammatically correct, and entirely useless for generating alpha. They were giving me very average information, the kind of consensus view that’s already priced in before the markets even open.
As a Product Specialist, my job was to find the edge. After hundreds of iterations, I realized that the secret is in the context architecture we build for the AI.
The evolution of the prompt
The journey from an average prompt, to something that would truly generate alpha is to move from vague commands to structural frameworks. LLMs typically suffer from "the path of least resistance." If you give them a lazy prompt, they give you a lazy answer. To get the depth required for an Investment Committee (IC) or a long-term equity thesis, you have to trap the AI into being brilliant.
Role-based constraints
The breakthrough happened when I stopped asking the AI to just summarize, and started training it to interrogate and audit.
Analysts don't just want facts, they want the deviations. For example, when I prompted the AI to:
“Source the latest management commentary from [Company]'s earnings call, then benchmark it against Third Bridge expert intelligence. For each key claim, identify whether the expert consensus confirms, nuances, or contradicts management's narrative, and note how many experts independently reached that conclusion.”
The output shifted from a generic summary to a targeted risk report. The AI stopped being a librarian and started acting like an actual analyst.
Context is the new currency
Prompt engineering has evolved into context design. At Third Bridge, we sit on a goldmine of proprietary transcripts. When you combine that "human intelligence" (the raw data) with a "structural prompt" (the logic), you get something the rest of the market doesn't have: Proprietary synthesis. A well-crafted prompt turns three earnings calls and a dozen expert interviews into a single, structured view of where management's story breaks down, in the time it takes to pour a coffee.
The lesson: Don't be a generalist
If you are still using one-sentence prompts, you are leaving insights on the table. The AI models we use today are capable of incredible reasoning, but they need to be told how to think, who to be, and what to ignore.
At Third Bridge we’ve distilled these Product Specialist secrets into a series of ready-made LLM skills. We’ve done the trial and error, so you don’t have to.
Ready to upgrade your workflow?
Don't settle for average. Request a demo today to unlock the power of the Third Bridge MCP with intelligent prompting.
