The loop
Three loops, not one. Loop A (GC → subs) is high volume but cold. Loop C (sub → sub) is word-of-mouth inside trade networks. Loop B, a happy sub carrying Perch to a GC who isn't on it yet, is the prize, because it manufactures net-new accounts on the valuable side.
The metric, defined
Net-new activated accounts from existing customers' bids
One number that means "the loop is working." Everything below is a dial that moves it.
The loop turns in weeks
Loop A resolves in the bid window (~2–4 weeks). Loop B is paced by how often a sub bids a new GC — bid frequency, not project length. So our real constraint isn't the calendar, it's conversion.
What to instrument
Leading indicators, because we can't wait a full bid cycle to learn whether the loop fires. Watch the flagged one hardest. It's the early signal that Loop B is real.
The bet it all rests on
Does a sub have to use Perch to respond to a bid, or can they reply by email and route around us? If they can route around us, the loop leaks at the worst possible point and K collapses no matter how many invitations go out. The single biggest lever on growth is how much of the bid response lives inside Perch.
Reputation — the conversion lever
Worst-case 1% on each loop won't grow fast enough. Verified, portable reputation is how we juice both conversion rates — and the model is Carfax: the point isn't "this sub is reputable," it's that the absence of a track record becomes the red flag. A GC can't not check.
Claim, don't build
Pre-populate a sub's profile from activity we already see, so activating = claiming something valuable they own. And train GCs to expect a Perch record — so a sub without one looks risky.
Flaunt the record
A sub with verified wins wants to show them off — so they bid through Perch and pull the new GC in to see the score. Reputation only pays off where it's visible.
Required-to-transact makes the data true. A 10× badge means something only because those ten wins happened inside Perch — un-gameable, like Carfax's authoritative-source data. The leak-proofing bet and the reputation moat are the same bet. One caution: be fair (Carfax has been sued over errors it was slow to fix) — lean positive, keep it correctable, and expect it to compound as the network grows, not on day one.
Design constraints
Subs free in the land-grab
Subsidize the side that fuels the loop. Charge the GC: they have budget and get the value.
Make sub value portable
Bid history, reusable templates, reputation they own, value that travels to every GC. That's what fires Loop B.
Never spring a tax
The spreader must feel served, never surprised.
Pressure-test it
Rough inputs, honest defaults. Move the sliders and watch the GC-to-GC coefficient: that's whether the network feeds itself.