Company State
Maintain a continuous model of the firm's full relationship with each company.
Company state is the firm's complete, current picture of its relationship with one company. Who has met them, what was said, what was sent, what changed since, what is still open. The score and the filters tell you which company to pay attention to. State tells you everything you already know about them the moment you do.
State is generated, not stored
You should not maintain a “state field” on a company record. The minute you try, it goes stale. State is something an LLM produces on demand by sweeping your systems in parallel and assembling the answer. That works on one condition: the underlying data has to be good, accurate, and connected. Garbage CRM, no transcripts, no wiki — no state. Everything before this module exists to make this module possible.
People, companies, deals, notes, owners, last-touched.
Who met whom, what was said, what is scheduled next.
Verbatim record of every conversation, searchable.
Memos, diligence notes, internal write-ups.
The Slack thread where the real opinion lives.
Decks, data rooms, contracts, attachments.
One company, on demand: the state skill
For the “where are we with this company” question, the right shape is a skill that runs inside ChatGPT or Claude, uses your connectors to hit all of the above systems in parallel, and returns a single briefing: relationship timeline, work product trail, open threads, and live chatter. Triggers on natural phrasing — “where are we with,” “status on,” “catch me up on,” or right before a call.
Same input every time, same shape of output every time. The partner does not have to remember which tab the deck was in or which Slack channel the last debate happened in. The skill knows where to look and what shape the answer should take.
Many companies, on demand: state across the pipeline
The single-company skill answers tell me about this one. The other half of state is tell me about a slice of the pipeline — and that is a different problem. You are no longer summarizing sources for one record; you are running a query across the whole deal database and getting back a structured, printable list.
The pattern that works: give your internal intelligence layer a tool that can run SQL-style queries against a live mirror of the CRM pipeline data, and let it return the matching companies with the fields you care about. Questions then sound like questions, not like database work:
- “Prepare a report of every company we have connected with in the LA area for our visit next week.”
- “Which companies in the pipeline raised in the last ninety days and have a partner-of-record assigned?”
- “Show me everything in the energy bucket we haven't touched in six months.”
Under the hood it is the same intelligence engine that powers the rest of the firm's retrieval — just with one extra tool for structured CRM access, and a small internal guide that teaches the model how to write the query, what columns to pull, and how to format the answer so it is actually printable and digestible by the team.
Why this is the last module
Company state is what everything else compounds into. A sharp thesis decides what enters the universe. Enrichment makes each record decision-grade. Filtering surfaces the right slice. Scoring orders attention. State is the payoff: every previous module's work, rolled up into an answer a partner can read in thirty seconds before a meeting — or sliced into a report for the team before a trip.