For sponsorship sales & partnership teams
Three agents that do the paperwork of a sponsorship partnership.
Scout finds the brand. Closer drafts the proposal. Recap proves the deal out at renewal. Every document is built from a validated model first — every number traces to a source, every claim gets a human review before it ships.
The problem
01
Prospecting takes days, not deals
Building a credible list of sponsorship prospects means hours of research per brand, and most of it is guesswork about fit and timing.
02
Proposals get generic, or slow
A tailored proposal — with real inventory, real pricing, real comps — takes a full day to build well. So teams reuse the last one instead.
03
Recaps are where trust is won or lost
The renewal conversation lives or dies on whether the numbers in the recap actually hold up. One inflated stat costs the whole account.
The roster
Three agents, same shape: a two-command, human-in-the-loop flow that builds a validated model first — and stops for your review — before it ever writes a document.
Prospecting
The Scout
npm run scoutTurns a property profile into a ranked list of ~10–15 brand sponsorship prospects.
Every fact about a brand is researched live on the web and cited to the exact URL it came from — fit reasoning, timing, decision-maker, and a concrete first move for each one.
Proposals
The Closer
npm run closerTurns a prospect brief, your inventory, and past-deal comps into a first-draft partnership proposal.
Every dollar figure traces to the rate card or a comp — package, valuation, activation ideas, and the timing case, ready for a rep to send with light edits.
See a real proposal →Renewals
The Recap
npm run recapTurns campaign data and the original deal terms into a source-checked sponsorship recap.
Every number in the document is checked back against the model — the renewal conversation runs on numbers that actually hold up.
See a real recap →The guardrail
The reason to trust a number in one of these documents isn’t that a model produced it — it’s that the run couldn’t finish if the number were wrong. Three layers, on every run:
Layer A
Schema
A factual sentence with no source can't parse. The model itself won't accept an unsourced claim.
Layer B
Code — the gate
Every cited source must exist. Every number in the document must equal a cited, code-computed value — not a model guess. A blocking violation stops the run before a document is written.
Layer C
Cheap model — advisory
A second, cheaper model flags shaky claims, thin evidence, or boilerplate reasoning as warnings for the human reviewer. It never passes or blocks a run on its own.
In practice, per agent
- The Scout: Every note URL's domain must appear in the research log the web step actually surfaced — a fabricated citation is blocked outright.
- The Closer: The whole valuation — package, rate card, comps — survives a fresh recompute; each suggested range endpoint equals a code-computed anchor, not a model estimate.
- The Recap: Every delivered number must equal a cited factbook value within ±0.5%, and media value must survive a fresh recompute.
How it works
1
Analyze
The agent builds a structured, validated model of the output — prospects, package, or recap — then stops. Nothing gets written yet.
2
You review
A plain-language review sheet with every QC flag up top. Edit the model directly, or fix the input and re-run analyze.
3
Write
Only once the model passes QC does the agent turn it into the final document — re-checking every number and citation one more time on the way out.
Human-in-the-loop by design — the agent never sends anything on its own. It hands you a document that’s already been checked, so your edit pass is about tone and judgment, not fact-checking.
Get in touch
Tell us about your property. We’ll walk you through what a Scout, Closer, or Recap run looks like for your roster.