
DelphiDeal desk Weeks to days
A founder used to wait weeks for an answer, and now it comes in days.
The ask
Linkin Capital is a venture fund in sport, wellness and entertainment, run by four partners. About thirty deals a month arrive by email, WhatsApp and Slack, and deals got lost in the volume. The partners wrote every memo by hand. The expensive error is a wrong decline: the fund loses the deal and the founder relationship.
First we had to rebuild the fund's memory.
The mailboxes
The deal history lived in four partners' inboxes: more than 180,000 emails going back years, the fund's instinct for a deal buried inside them. The agent threaded that whole memory in under eight minutes.
Most deals are not deals
Most of what arrives is not a deal at all: an Affinity digest, an LP letter, a platform notice. The agent tells a real deal from the noise and almost never misses one.
One deal, many threads
Founders write to several partners at once, and two deals can share the same bankers. The agent groups the threads into one deal and never collapses two different ones into one.
No “no” without two models and a citation.
The asymmetry
Silence is not a decline, and more than half of inbound gets no reply at all, so the agent never learns from silence. A decline needs two independent AIs to land on the same reason, each quoting the line from the deck. A partner approves every decline draft.
“Nice statistics. It catches the point that we are not aligned in our answers.”
The research
A deal that advances gets market and competitor research with sources and a score on Linkin's own scorecard. Only the company name, website and sector leave for a web search; the fund's mail never does.
A panel of judges sits on every deal.
The panel
Judges with personas the partners choose score each deal. The scoring runs on two independent AIs, and a third breaks a tie.
A partner signs every answer.
In Slack
The memo, the one-pager and the committee report arrive in the partners' Slack, with follow-up questions in the thread. Linkin installed it in its workspace on 28 October 2026, two days after kickoff.
Before the founder
Every answer goes out in one format and passes a partner first. The founder call comes after the analysis, so neither side spends time on a deal that will not move.
The controls
The partners own the criteria. The agent drafts a change to how deals are judged; a partner approves it with a Slack button after seeing it replayed on past deals, and every version is kept.
From a deck landing to a founder's answer.
Every deal is weighed against the ones the partners took and the ones they passed, their own judgment on the next deal before they have finished reading it.
Book a callKeystone
The Underwriting Engine for Real Estate Investment
Next file