Early access — seeking design partners

Conversion intelligence for subscription businesses.

SubCore Agent learns which users a paywall, cap, or offer actually converts — and which it would churn — then ships a per-user policy your team reviews. It even bootstraps the randomized data you don't have yet, and keeps a holdout so every rollout sharpens the next.

subcore-agent · decision brief preview
Illustrative example
brief.mdReady for review
Propertyfree_tier.monthly_token_cap
PrimaryMaximize trial-to-paid @ 30d
SecondaryMinimize token cost (soft)
GuardrailMAU change ≥ −1.5% (hard)
See how the agent got here
per-user assignment · uplift policy108,690 users
Persuadables · treated68,475
positive uplift · per-user cap
Flat / low uplift · control26,325
held at current cap
Sleeping-dogs · protected13,890
negative uplift · always control
holdout: 10% always-on randomized · measures realized uplift
Problem

The gap isn't the model. It's the decision.

Most subscription teams already sit on the data needed to lift conversion. Some have models on top of it, some don't. Either way, the path from a signal to a paywall, cap, or offer decision someone is willing to ship is still slow, manual, and fragmented — and almost nobody has the randomized data to know who the treatment actually converts versus who it would churn.

What teams already have
  • Subscription, billing, and usage tables in a warehouse
  • Paywall, trial, and offer exposure logs
  • A dashboard or two, plus past experiments
  • Maybe a conversion or pLTV model. Maybe not.
  • A trial-to-paid, paywall, or offer question waiting on an answer
What's still missing
  • Which users the treatment actually converts (persuadables) — and which it would churn (sleeping-dogs)
  • The randomized data needed to even measure that — most teams don't have it
  • Business constraints, discount budget, and guardrails baked into the decision
  • A per-user policy — not a segment or a score — ready to ship
  • A rollout someone can sign off on, with a holdout that keeps sharpening it

A conversion rate, a propensity score, or a dashboard chart is not a decision.

SubCore Agent works on the decision layer — modelling uplift (how each user actually responds to the treatment, not just who converts), bootstrapping the randomized data when it doesn't exist, and turning it into a per-user policy your team reviews and ships behind a sign-off. Then it measures what actually worked and makes the next decision sharper, under your guardrails.

Agent loop · illustrative worked example

One example: personalising a free-tier token cap to lift trial-to-paid.

An eight-stage loop — Define through Rollout — inside your perimeter, on your warehouse credentials, with a human sign-off before anything ships and a holdout that sharpens the next decision. Numbers below are illustrative, not benchmarks.

Agent activity· Framing the decision before touching data
Illustrative
Define
thinking
Decision· Define
Illustrative example
Decision brief · locked
Illustrative
Served property
free_tier.monthly_token_cap
integer · served per user at request time · registry: subcore/properties
Primary · maximize
trial_to_paid @ 30d
Secondary · soft tiebreak
minimize token_cost
Guardrails
  • MAU change ≥ −1.5%hard
  • Infra spend ≤ $48k/mohard
  • Retention D30 stablesoft
Capabilities

Built for conversion trade-offs.

Every SubCore Agent workflow is grounded in a real per-user uplift decision — who the treatment converts, who it would churn — not a generic ranking or ML metric.

Personalized free-usage caps

Which users does a higher cap actually convert — and which does it churn by removing the reason to upgrade? SubCore treats the persuadables and protects the sleeping-dogs.

Trial optimization

Which trial — length, or trial vs. straight to the paywall — actually lifts each user's chance of converting, versus which would have converted anyway on any of them?

Offer & discount targeting

Which users need an offer to convert, and how deep — versus full-price payers you'd be discounting for nothing, or price-shoppers a discount would churn on renewal?

Why it's different

Not a dashboard. Not an A/B test. A decision that improves.

Decides per user with uplift

Not a segment, not a propensity score — SubCore models how each user actually responds to the treatment. Persuadables get treated; sleeping-dogs (users the treatment would churn) are always held at control, protected by policy.

Bootstraps the data you're missing

A built-in randomized-collection + always-on holdout substrate — a thin experimentation layer — so you don't need to stand up an experimentation platform first. When you already have one (RevenueCat, LaunchDarkly, Statsig), SubCore syncs assignments to it.

Trains in your compute, on a budget you set

Credit-bounded architecture search runs in your Modal or Databricks sandbox and registers the champion to MLflow — nothing black-box, no model weights leaving your perimeter, and the credit meter is visible before you approve the run.

Safe to ship — and sharper every rollout

A human signs off before anything goes live. Guardrails, kill-switch, and auto-rollback are on by default; an always-on randomized holdout measures realized uplift so the next decision is grounded on what actually worked.

Is this you?

We're looking for teams who feel this.

  • You're a B2C subscription or freemium business trying to lift free-to-paid or trial-to-paid conversion.
  • You still decide who sees which paywall, offer, or trial with static rules and broad segments.
  • You pick treatments from intuition, then test one hypothesis at a time and wait weeks to learn.
  • You already have the data — warehouse, experiment logs, maybe a model — but turning it into a shipped, personalized policy is slow and manual.

If that sounds familiar, we think we can help — and we're looking for a few design partners to build it with.

Early access

Join the waitlist.

SubCoreAI is in early development and looking for its first design partners. Join the waitlist if your team works on trial-to-paid conversion, paywalls, offers, pricing, onboarding, or ML-informed subscription growth.

Design partner program

Looking for 3–5 early teams

We’re selecting a small group of teams working on subscription conversion — paywalls, offers, trials, pricing, and onboarding. As a partner you get early access, hands-on onboarding, and direct input into the roadmap. In return, we ask for a 30-minute onboarding call and one real workflow to test together.