Company Deep-Dive Skill
The Skill that turns 'do a deep-dive on this ticker to our house standard' from a two-day analyst task into a two-hour first draft, with sources, no hallucinations and the house shape.
Deep-dives are the highest-leverage output of a research team and the slowest thing to produce; they also drift in shape as different analysts stamp their taste on them.
Every deep-dive comes out in house shape, sourced end-to-end, in a fraction of the time — so analysts spend their day on the judgement calls, not the assembly.
What this Skill needs and what it returns
- • Ticker or private company name
- • House thesis lens (long/short bias, holding period)
- • Sector context from Firm Memory
- • Company one-pager — business model, unit economics, moat, key risks
- • Financial snapshot — three-year P&L, balance sheet, cash flow, key ratios
- • Filings & transcripts summary — last four quarters, flagged deltas
- • Peer set — comps table with the firm's chosen multiples
- • Red flags — accounting, governance, litigation, insider transactions
- • House-standard memo, ready for analyst edit
- • Sector notes and past deep-dives from Firm Memory
- • House thesis frameworks
- • Preferred data vendors and their licence terms
- • Every material claim cites a source in Firm Memory or an external URL.
- • No forward-looking statement is generated without an explicit 'analyst view' block for a human to fill.
- • Model choice: private-hosted for anything touching internal notes, frontier for public-web synthesis, per the Model Router policy.
- • Source coverage: fraction of claims with a valid citation.
- • House shape: match to the firm's memo template.
- • Analyst rework: median edit distance between first draft and final.
- • Hallucination rate against a known ground-truth set of past deep-dives.
Hedge Funds
skills
Ask any head of research where the team's time goes and the honest answer is the same: assembling the deep-dive. Not the judgement — the assembly. Pulling the filings, tabulating the comps, summarising the transcripts, laying it all out in the house shape. It is high-status work that is mostly low-signal to produce and high-signal to read. The Company Deep-Dive Skill is the OS's answer to that.
This is not a chat-with-a-PDF tool. It is a versioned Skill that reads from your Firm Memory (sector notes, past deep-dives, house frameworks), calls the Model Router with policy-appropriate models, and produces a memo in the exact shape your team already reads. The analyst opens it, adds the parts only they can add — the view, the trade, the risk — and moves on to the next name.
Assembly is a Skill. Judgement is a person. Get the split right and one analyst covers ten names a week instead of two.
How we ship it
- Week 1 — capture the house shape. We take five recent deep-dives and reverse-engineer the template, the tone, the level of detail, the standard exhibits.
- Week 2 — build the retrieval scopes. Firm Memory gets pointed at the sector library, the frameworks and the vendor licences.
- Week 3 — wire the Model Router. Private-hosted for internal-notes synthesis, frontier for public-web filings and transcripts, small model for peer-set assembly.
- Week 4 — build the eval set. Ten past deep-dives become ground truth. The Skill has to pass on source coverage, house shape and rework distance.
- Week 5–6 — pilot with two analysts. Iterate on the prompts, the shape, the exhibits, until the analysts choose the Skill over their old workflow.
Ship this Skill
This Skill is one of several we build inside a KJ Capital AI-First OS. Engagements start with a two-week £15k AI Diagnostic. We audit your systems, design the OS across the six layers, and ship the first version of Skills like this one inside the following six to twelve weeks.
FAQ
Can this replace an analyst?
No. It replaces the assembly part of an analyst's day. The judgement — the view, the trade, the sizing — remains theirs. Firms that try to remove the analyst end up with memos nobody trusts.
How is this different from ChatGPT with browsing?
ChatGPT with browsing does not know your house shape, does not read your past deep-dives, does not follow your Model Router policy, and does not log anything for audit. It is a component, not a Skill.
How long does the first version take to ship?
Four to six weeks after the Diagnostic, if the firm can commit two analysts to piloting and iterating.
What data vendors do you support?
The common ones — Bloomberg exports, Refinitiv, S&P, FactSet — through their licensed APIs. We do not scrape licensed data.
Ship this Skill inside your firm
Every KJ Capital engagement begins with a two-week £15k AI Diagnostic. We design your AI-First OS end-to-end, cost the build, and name the first three Skills to ship.