Your AI Growth Engineer
Your AI Growth Engineer studies your content, audience, and performance, then turns what it learns into measurable experiments for your next move.
Not a scheduler. Not a caption generator. An engine for creators that reads your channel, learns what actually grows your audience, and hands you experiments you can measure.
Growth Engineer
Demo dataAnalyzing @creator...
Patterns appear here once evidence clears the bar
Platform analytics
what happened
Growth Engine
studies your history
Evidence
n, window, confidence
Hypothesis
a testable claim
Experiment
measured vs baseline
Dashboards stop at observation. Grobunchy keeps going: it separates correlation from cause, states a hypothesis, and measures the result against your own baseline, so the next post is a decision, not a guess.
How it works
Connect your creator account.
Read-only analytics access. No posting rights, ever.
The Growth Engineer studies content and historical performance.
Baselines per format, retention curves, hooks, topics, conversion.
Patterns become measurable experiments.
Control, intervention, target metric, guardrails, window.
Every result updates your Creator Growth Model.
Supported, rejected or inconclusive: the model learns either way.
Creator Growth Model
Not a generic playbook. Eight dimensions, each filled only with facts that clear an evidence bar, each one confirmed, weakened or contradicted as new results land.
Question-led hooks hold viewers past 30s more often than statement hooks.
Workflow breakdowns are accelerating with returning viewers.
Sunday uploads no longer outperform weekday uploads.
Demo facts
Growth Model learning
Demo dataExperiments
A hypothesis is only a hunch until it is measured against your own baseline. Experiments make the Growth Engineer accountable.
Experiment
EXP-014Demo dataHypothesis Opening with a stated opinion in the first 8 seconds raises subscriber conversion versus the tutorial-style cold open.
Control baseline
4.2 subs / 1k views (n=41)
Target
>= 4.8 subs / 1k views
Current
5.1 subs / 1k views (n=6)
Confidence stays low until the sample is large enough. The card will not call a winner early.
The intervention beat the control beyond the noise. The model records it as evidence.
It did not. That is a real finding too, and the model stops recommending it.
Not enough sample or too much variance. No claim is made.
Content is assigned and the measurement window is open.
Pre-publish
Paste a draft, a script or a title. The Growth Engineer compares it with your own history, not a generic rulebook, and flags where it drifts from what has worked for you. It suggests. You decide.
Draft "Why I stopped batching content"
Demo dataRelative to your last 90 days. This is a read on risk, not a forecast of views.
Growth Missions
Set an objective. The engineer picks the experiments most likely to move it, runs them in order, and reports what it learned, including when the answer is 'not this'. Progress is measured, never promised.
Mission
Demo dataMove subs-per-1k-views on long-form uploads, without lowering 30-second retention.
Metric subscribers gained per 1,000 views, long-form
Active experiments
2
Learning progress
38%Window ends in 6 weeks
Platforms
Core intelligence never assumes one platform's analytics. Where a capability is missing, the engine says so instead of pretending.
Read-only analytics, retention, subscriber conversion.
Read-only video list and lifetime counts. Invited test accounts while TikTok reviews the app.
Beta: professional (business or creator) accounts, invited testers only while Meta reviews the app.
Capability model designed; integration not live.
Pricing indicative
The subscription funds a persistent Growth Engineer and a model that keeps learning. Prices and limits below are indicative and may change before launch.
Understand what has been working.
~$29/ month
Run the loop every week.
~$79/ month
For creators who run it like a lab.
~$149/ month
Indicative pricing. No performance fees. No tier promises followers, views or virality: only evidence.