The CRM Claude Builds From Your Sent Folder. No Scraping.

A CRE professional ten years into a career is sitting on a relationship record no commercial database can match: tens of thousands of sent emails, years of calendar entries, call notes, and deal folders, together encoding who they know, how well, through whom, about what, and how recently. The reason this dataset went unused is that it was unstructured, and turning an inbox into a contact book used to mean an intern and a quarter. That is the constraint that moved. Extracting people, firms, roles, relationship context, and last-touch recency from correspondence is squarely the kind of work language models do well, and it runs entirely on data the professional already owns, inside accounts the firm already controls.

Contrast the plan people reach for first: scraping LinkedIn for former colleagues and industry contacts. Setting aside effectiveness, the terms are unambiguous. LinkedIn’s User Agreement obligates members not to develop, support, or use software, devices, scripts, robots, or any other means or processes to scrape the Services or otherwise copy profiles and other data, and separately not to use bots or other automated methods to access the Services, add or download contacts, or send or redirect messages. LinkedIn’s own help pages state that using prohibited software risks account restriction, and enforcement is routine. The professional consequence is asymmetric: the account being risked is the professional identity the CRM was meant to serve. The sent folder carries no such risk, and, decisively, it is a record of relationships that exist rather than profiles that do.

Structured correctly, the extract produces a person-level book: name, firm, role as described in signatures and context, the deals and properties discussed, who made the introduction, thread frequency over time, and days since last touch. It produces a firm-level view: which shops the professional actually transacts with versus merely knows. And it produces the two lists a working broker or principal will use every week: relationships going cold, high-value contacts with no touch in ninety days, and relationship paths, who in the book can introduce whom. None of this requires a single byte of external data. The calendar adds meeting cadence; the deal folder adds transaction context; the notes add what was promised. The result is not a contact list. It is the professional’s actual network, made queryable.

The honest limits belong in the design. Correspondence is one-sided: it records what the professional discussed, not what the counterparty is doing now, so titles drift and firms change. Enrichment, refreshing a title, confirming a move, is where external data legitimately enters, and the boundary is the same one this desk applies everywhere: licensed means only. That means data from providers whose commercial terms permit the use, information the contact shared directly, and public records, not automated collection from platforms whose terms prohibit it. In practice the enrichment need is smaller than expected, because the book’s value concentrates in recency and context, which the correspondence itself keeps current every time a thread advances.

The build sequence fits in a month of part-time attention. First pass: the extractor walks the sent folder and calendar, both within the firm’s own workspace under its commercial AI terms, and proposes the person and firm book with every field traced to the messages it came from. The professional reviews the proposed book once, merging duplicates and killing false positives, which is an evening of work that pays for years. From then on the system runs incrementally: new correspondence updates recency and context, a weekly digest surfaces the going-cold list and any promised follow-ups detected in threads, and the book answers questions on demand: who do we know at the firms active in this submarket, who has touched this owner, when did we last speak to the lender on that deal.

The gates match every PSV production workflow. The book is internal work product; nothing it generates contacts anyone autonomously. Outreach drafted from it, the re-engagement note, the introduction request, ships as a draft for the professional to send, in their voice, from their account. The reviewer is the professional whose relationships these are, because a wrong merge or a stale context line in a live thread costs real credibility. And the data governance is worth writing down before the build: the book derives from firm correspondence, so it lives in firm systems under the same access controls as the mail it came from, and what happens to it when someone departs is a policy decision to make on purpose, not discover in an exit dispute.