What Should a Fund Agent Do When It Cannot Match an Email to a Deal?

An intro email names a company your team has been watching, but the name also fits another company in the CRM. The worst response is a confident-looking new deal. A fund agent should keep the mail and say what it could not resolve. A person can then identify the right company without losing the original evidence.

Quick Answer

Do not guess the company, change a stage, or discard the message. Record the source and the reason the match failed. If a person recognizes the company, let them select the existing company or deal and rerun the same guarded action. Their choice resolves identity; it does not turn a weak signal into proof of a close.

A missing match is not missing work

The Relationship Manager in GPAgent watches configured sources for an intro the GP requested and for investment-closing evidence. It keeps an observation ledger rather than deciding from a message subject alone. If an observation has no reliable match, several plausible matches, insufficient authority, or an error, the agent can surface it in its Couldn't act on list. That is a list of specific evidence it saw, not an approval queue that pretends every email deserves a new task.

Consider two companies sharing a short name. The email might mention a founder, but a name in isolation does not identify which deal the fund meant. A team member should be able to open the observation, check the original source, and pick the right record. If neither record fits, investigate before creating one. The convenient shortcut, creating a second deal and cleaning it later, pollutes the next investment review.

The fix should use the same rules as the run

GPAgent's manual fix passes the stored observation through the same resolve, plan, and write path as a scheduled signal run. Picking a deal supplies the missing identity and records who fixed it. It does not upgrade the underlying mail's authority. For example, picking a company for a signed SAFE does not promote that notice into proof that the fund's own investment closed.

Try again is different from pick. If a required source was off or an operation failed, a rerun uses the current source policy and resolver. If you already know which deal the evidence belongs to, picking that record is the explicit human decision. Both paths leave the usual deal receipts rather than a private spreadsheet of exceptions.

The NIST AI Risk Management Framework offers a broader way to think about governing and managing AI risk. It is not a certification of this workflow. Here the practical standard is smaller: show the evidence, preserve the reason for inaction, and make a correction traceable.

See the agentic investing team guide for the Relationship Manager's scope. The point is not to eliminate uncertainty. It is to keep uncertainty from quietly turning into a fact in your deal pipeline.

FAQ

Does fixing an observation automatically create a deal?

No. The same reaction rules decide whether to create or update a deal, based on the signal, the chosen identity, and the firm's policy.

Does picking a company prove the investment closed?

No. A person's choice resolves which company the evidence refers to. It does not change a signed SAFE or a weak email into closing proof.

Can the agent look again after a source is enabled?

Yes. The manual retry uses the current policy and resolver. It may still decline to act if the evidence remains ambiguous.