LARGE LANGUAGE MODELS IN THE FUNDAMENTAL RESEARCH PROCESS
Research you can defend.
provenance is an AI research platform built for asset managers — where every statement traces back to its source, every instruction is visible, and your investment process becomes an asset you own, refine, and compound.
The problem with AI in investment research
AI can now read every broker report, every filing, every data feed your team receives — and draft research in minutes that would take an analyst days.
But a number or statement you cannot trace is a fact you cannot use. Not in front of an investment committee. Not in front of a client. Not in front of a regulator.
And the problem runs deeper than data. When an AI produces a conclusion, you need to know not just what it read, but what it was asked to do — which questions shaped the analysis, which standards were applied, which instructions governed the output. Without that, you cannot explain why the research says what it says. You can only hope it’s right.
Generic AI tools give you answers. They cannot tell you where those answers came from, or what drove them there. For an asset manager, that’s not a research tool you can rely on.
What provenance does
provenance puts a simple discipline at the centre of AI research: nothing enters an AI-generated report without a trail.
Every fact is anchored. Broker research, market data, filings — everything the system reads is catalogued against its exact source: the document, the page, the exhibit, the data point, the moment it was retrieved.
Every instruction is visible. The questions your analysts ask, the templates they use, the standards they apply — these aren’t buried in hidden prompts. They’re first-class objects in the system, versioned and auditable. When a report’s conclusion changes, you can see whether it was the data that moved or the instruction that was refined.
Every step is recorded. Research is produced not by one opaque AI, but by a chain of small, specialised steps — each one logged: what it was asked, what it read, what it concluded. The reasoning is as traceable as the data.
Your process, compounding
provenance doesn’t impose a research process. It runs yours. Your resource analysts ask different questions from your banks analysts; your initiation reports have a house structure; your team knows which red flags matter. That proprietary process is captured explicitly — your questions, your standards, your templates — and applied consistently to every name you cover. The knowledge that today lives in your best analysts’ heads becomes an asset your firm owns.
And because every report is fully traceable — both the data it drew on and the instructions that shaped it — every weakness is findable. When an output isn’t good enough, you can see exactly which question, which step, which source, or which instruction produced it, and fix that one thing. The fix is permanent; every future report embodies it. If you want research you can refine, you need research you can debug.
AI-assisted research with an evidentiary standard: every number has a data source, every statement has an instruction source, every conclusion has a trail — and your investment process becomes a compounding asset.