Growth engine

The growth engine answers one question the rest of CoreX could not: which channel produced the install that became a paying customer — and then puts the work that follows from the answer in front of you, ranked.

It is a workspace-level app (App Store key growth), off until you turn it on, and it reads the same shared model every other CoreX app does. A Stripe SaaS and a Shopify app get the same engine.

The four screens

Screen What it is for
Scorecard Installs, install → paid, churn and ARPU for the period, with the channels underneath them.
Work queue Everything worth doing today, ranked, each item carrying the button that resolves it.
Playbooks Recurring plays — rescue a stalled install, win back a churned customer — each measured against a holdout.
Tracked links One link per place you promote the product. Every click becomes an attributable touch.

How attribution works

Every acquisition event — a tracked-link click, a free-tool run, an outreach send, an affiliate referral, a social post — is recorded as a touch against a subject (an anonymous visitor id, an email, or a customer). Milestones (sign-up, install, activated, paid, expansion) are materialized from your existing data on a schedule.

Attribution joins the two at read time, so you can switch model without losing history:

  • Last touch (default) — all the credit to the most recent touch before the milestone.
  • First touch — all the credit to the touch that started the relationship.
  • Linear — credit split evenly across every touch in the window.

A conversion with no touch inside the lookback window is reported as Direct / untracked. That bucket is deliberate: spreading unattributable conversions across your other channels would make every channel look better than it is.

Lookback defaults to 60 days and is adjustable per report. A touch older than the lookback does not get credit.

Getting started

  1. Turn on Growth engine in the App Store.
  2. Create a tracked link for every place you promote the product, and use it instead of the raw URL. Until something is tracked, every conversion lands in Direct.
  3. Name your activation event on each app (on the app's Events page, mark the event that means a customer got value) if you have not already. Without it the engine can measure installs and payments but not the step in between, which is the one that predicts revenue.
  4. Leave it a week. The scorecard needs a period to compare against, and playbook lift needs enough decided runs on both arms before it will show you a number.

What runs on its own

Schedule Job
Hourly Materialize conversions from installs, charges and sign-ups.
Daily, early Resolve activations from your activation event.
Daily Refresh the work queue, then enrol playbook audiences and measure what is due.

Running more than one app

The By app table on the scorecard puts the same four numbers side by side for every app in this workspace, so a portfolio of a dozen apps does not need a dozen dashboards. Growth only ever reads the workspace you are in; to see another one, switch workspace from the account menu.

Benchmarks

Under Install → paid, Churn and ARPU you may see a second line: the median for workspaces like yours, and where you stand against it. "Your churn is 3.6%" is a fact nobody can act on; "median is 6.1%, you are in the top quartile" is a decision.

The comparison is drawn from every CoreX workspace with enough traffic for a rate to mean something, split by workspace type (Shopify app developer, SaaS on Stripe) with a cross-type line as the fallback while a type is still small.

What is never shared. Per-workspace numbers are computed and thrown away; only percentiles and a count of contributors are stored. A segment with fewer than five contributing workspaces publishes nothing at all — with three workspaces, a "median" is one competitor's number wearing a disguise. Nothing you can read here can be traced back to another workspace, and nothing another workspace reads can be traced back to you.

Every workspace with public pages also publishes two machine-readable indexes:

  • /p/{workspace}/llms.txt — a plain-text list of your public pages with their descriptions, the format assistants read when someone asks what your product does.
  • /p/{workspace}/sitemap.xml — the same set for crawlers.

Public pages also carry WebPage structured data server-side, so a crawler that never runs JavaScript still reads the title, description and dates.

Both indexes are built from the same rule that serves the pages: visible and published only. A draft, a hidden page and a password-protected page appear in none of them — for the locked page the title alone was most of what the password was hiding.