◆ Powered by Minds by Animoca Brands Creative Minds Jam #1 · Audience Growth & Community Engagement

Every member who leaves
deserves a porch light
left on.

Porchlight is an exit-interview Mind that remembers why each member left — and autonomously wins them back the moment their stated reason is actually fixed. Quoting their own words. Honoring do-not-contact.

0.00/mo recovered · 6 members
+0.00/mo keywords miss
0.00confidence · zero keywords

The silent wound

Churn drains MRR.
And the reason vanishes.

For every membership creator — Patreon, YouTube memberships, Substack, Discord subs — cancellation is the direct revenue wound. And the loop around it is broken the same way everywhere: when a member cancels, nobody records why.

Months later, the creator fixes the very thing that drove people away — brings back a series, drops the price, cleans up the community — and the members who left for exactly that reason are never told.

The entire win-back industry is a generic "we miss you" blast that ignores why each person actually left.

maya · cancelledreason: unrecorded
"you pivoted to short clips — the long sit-down conversations i loved are gone"
dre · cancelledreason: unrecorded
"not what i signed up for since the format changed"
cai · cancelledreason: unrecorded
"long chatty sit-downs with guests…"

Six weeks later the creator brings the format back.
None of them will ever know.

How it works

One flow, devastatingly deep:
cancellation → memory → resolution → return.

Three load-bearing behaviors, all carried by a single persistent Mind — one conversation per member, kept alive across the weeks between goodbye and welcome home.

Exit interview

memory

The moment a member cancels, the Mind runs a short, warm interview and files a structured return-condition — the reason, category, and the member's verbatim quote — into per-member long-term memory. One Mind conversation per member; the memory belongs to the agent, not just a database row.

Churn intelligence

continuity

Every departure compounds into a living board — "37% of 19 departures left over the pivot away from deep-dives" — with per-reason breakdowns, open vs. won-back status, and a full audit trail of every change the Mind acted on.

Conditional win-back

autonomous follow-up

The creator announces what changed. Unprompted, the Mind judges every open reason by meaning, not keywords — and messages only the members whose reason is now genuinely resolved, quoting their own words from weeks earlier. Do-not-contact is honored even when the reason is fixed.

cancel → remember why → creator fixes it → the right members come home

The anti-clone proof · captured live

Two sentences. Zero shared words.
The Mind joins them by meaning.

This is what breaks the "churn survey + drip email" pattern-match. A keyword matcher scores this pair exactly 0. The live Porchlight Mind resolved it at 0.93.

Why Maya left · filed 6 weeks ago
"you pivoted to short clips — the long sit-down conversations i loved are gone"
short clipssit-downconversations
What the creator announced · today
"Big news — the deep-dive interviews are back, weekly."
deep-diveinterviewsweekly
mind.doesChangeResolve(maya.returnCondition, announcement)

resolves    = true
confidence  = 0.93
rationale   = "The member left because long sit-down conversations were
               replaced by short clips, and the announcement explicitly brings
               back weekly long-form deep-dive interviews, which directly
               restores the exact format they loved."

This is a real, captured LIVE call to the Porchlight Mind — not a mock, not a screenshot of a prompt. On zero shared keywords, the Mind returned resolves=true at 0.93 confidence and named the exact format restoration.

The measured edge

We didn't claim the difference.
We shipped the baseline and measured it.

A keyword matcher runs in-repo beside the Mind on every single win-back. Same seed, same announcement — the dashboard shows exactly which dollars only semantics can recover.

0.00/mo The Mind recovers

6 members won back — every one because their specific reason was genuinely fixed.

0.00/mo Keyword baseline recovers

The same announcement, matched on words alone, finds only the members who happened to use the creator's vocabulary.

+0.00/mo Only the Mind finds

Same reason, different words. Two-thirds of the recovered revenue is invisible to keywords.

▲ the +$70.00/mo gap is the semantic edge — measured, on screen

Six members came home. Four of them, keywords would never have found.

Beasemantic-only · keywords miss
Caisemantic-only · keywords miss
Dresemantic-only · keywords miss
Mayasemantic-only · keywords miss
Arjunkeyword-caught
Fenkeyword-caught

Bea said "long-form lore videos." Cai said "long chatty sit-downs with guests." Dre said "not what i signed up for since the format changed." Same wound, different words — only meaning connects them to "the deep-dive interviews are back."

Why only a Mind

The Mind isn't a feature.
It's the engine.

A persistent identity per member relationship

One Mind conversation alias per member, surviving the weeks between the exit interview and the win-back. Not a session. A relationship.

Long-term memory the agent itself carries

Return-conditions and verbatim quotes live in the Mind's own memory — semantic recall of why, in the member's exact words, weeks later.

A native outbound channel it uses on its own

The Mind reaches out from its own address — it even emailed the builder unprompted on day one. No cron. No trigger from us.

Remove the Mind and what's left is the keyword baseline — which our own dashboard proves recovers $70.00/mo less on the same seed. A stateless LLM API gives none of this without rebuilding memory, identity, and channels from scratch.

See it happen

Ninety seconds:
a cancellation becomes a homecoming.

Maya cancels → the Mind remembers why → the creator brings the format back → the Mind quotes her own words back to her, and the recovered-MRR counter ticks to $105.00.

Reproduce the full arc yourself in 60 seconds: npm install && npm run demo — headless, offline, deterministic. LIVE mode is one env var away.

The vision

Churn becomes a pipeline.

A stateless tool can say "we miss you."
Only a Mind knows this person left for that reason — and that today, it stopped being true.

Every win-back Porchlight sends carries a directly monetizable metric: recovered MRR per creator. That's the traction story for the Minds Investment Programme — instrument recovered revenue across pilot creators, take the loop from modeled rejoins to real re-subscribe webhooks, and make every membership cancellation recoverable.

What's real · what's modeled — honestly

  • Real: the live Mind, its memory, and the semantic edge — including the captured resolves=true, confidence=0.93 verdict on zero shared keywords, straight from the Builder API.
  • Real: the in-repo keyword baseline that runs beside the Mind on every win-back and measures the $70.00/mo it misses.
  • Modeled: the membership storefront is our own mock (Patreon's API is restricted) — cancellations enter through it, and the Mind-side flow is real end-to-end.
  • Modeled: "rejoins" are a demo model; production would await a real re-subscribe webhook before counting recovered MRR.

The porch light
is already on.

One deep flow, a live Mind with real memory, a measured $70.00/mo semantic edge, and a 0.93-confidence verdict no keyword tool could ever return. Built for Creative Minds Jam #1.