Message-market fit
Are we saying the right thing to the right people?
Tests Hooks, formats, angles, thumbnails, UGC styles, headlines
Measures CTR, thumbstop rate, CPP, qualified traffic
A Player Labs × Ecommerce Equation
Auto Throughput
Autonomous growth experiments for ecommerce. The system runs the loop — hypothesise, create, launch, measure, decide, next experiment — so the operator focuses on strategy, not manual testing.
March 2026
·
Collaborative build proposal
A Player Labs
H
Hypothesise
System proposes the next best test based on prior evidence
→
C
Create
AI generates the asset — ad, page, offer, campaign structure
→
L
Launch
Variant goes live with guardrails and contamination checks
→
M
Measure
Ad platform + store data reconciled against a profit objective
→
D
Decide
Keep, discard, scale, or rerun — with confidence scores
→
N
Next
Decision memory feeds the next hypothesis. Loop compounds.
What Karpathy's auto-research did for papers, Auto Throughput does for ecommerce growth. The system doesn't report on experiments — it runs them.
01
The Growth Equation
Loop A — Creative
Loop C — Landing Pages
Loop B — Offers
Loop D — Media Allocation
Four independent loops, one autonomous engine
Each loop optimises a different growth lever. Auto Throughput runs the experiment cycle across all four — and the shared decision layer makes every loop smarter over time.
Page conversion fit
Does the landing experience convert traffic into buyers?
Tests Page structure, copy, CTA placement, product framing, trust signals, checkout friction
Measures LPV to ATC, LPV to purchase, CVR, AOV
Auto Throughput Engine
Experiment registry
Primary profit objective
Confidence + effect size
Decision memory
Economic fit
Does the offer make money after all costs?
Tests Bundles, discounts, pricing presentation, guarantees, urgency, free shipping thresholds
Measures AOV, MER, contribution margin per visitor
Distribution fit
Are we spending in the right places at the right scale?
Tests Audience, campaign structure, budget allocation, bid strategy, placement mix, scaling rules
Measures CAC, CPP, reach, MER
Each loop runs Hypothesise → Create → Launch → Measure → Decide → Next
independently — the engine coordinates, avoids contamination, and compounds the learnings.
02
How It Works
A fully autonomous system connected to the APIs, grinding 24/7
Not a dashboard. An engine.
Auto Throughput connects directly to Meta and the store. It doesn't wait for someone to export a CSV and upload it. It pulls live data, runs classification, makes decisions, and launches the next experiment — autonomously, continuously.
Connected to
Meta Ads API
Shopify / Store
Analytics
EE already has the logic
MOAC classifies ads into SUPERHERO / HERO / SIDEKICK / CIVILIAN / VILLAIN tiers based on CPP, ROAS, and frequency. Auto Throughput takes that classification engine and runs it — plus all four loops — continuously against live data.
The gap
Distribution fit (Loop D) is where the MOAC engine lives today. Creative, Offers, and Landing Pages don't have their own engines yet. Auto Throughput builds those three loops and connects all four through a shared decision layer.
Pre-mortem: what's actually hard
Loop A — Creative
The hard part is generating the next creative hypothesis without falling into creative fatigue. Need diversity pressure — the system can't just recycle what worked.
Loop B — Offers
A bad offer test can tank margin before the system has enough sample to catch it. Need hard spend caps and loss limits per experiment. Price anchoring bleeds outside the test — customers screenshot discounts.
Loop C — Landing Pages
Most stores don't have enough traffic for fast statistical significance on page tests. Checkout path changes are highest-leverage but highest-risk — a broken checkout loses revenue now, not in a report next week.
Loop D — Media Allocation
You're testing against Meta's own algorithm, which is already optimising. Budget shifts trigger learning phase resets. Audience overlap between tests contaminates results — the engine needs isolation protocols.
03
Under the Hood
Agentic Flywheels
Each step in each loop is handled by a specialist agent — purpose-built, with its own training data, connected directly to live infrastructure. No humans in the execution loop. Humans set the objective and guardrails; agents run the experiments.
Signal
3,000 members marking every experiment: worked, didn't work, conditional
Anonymised results become reusable intelligence across verticals
Stigmergic feedback — each member's output sharpens every other member's next test
Benchmarks by vertical, spend level, and strategy emerge from real data
Orientation
Decision memory — every result stored, queryable, reusable
Hypothesis generation — next best test surfaced from prior evidence
Confidence scoring — Bayesian sequential testing, not arbitrary timers
Cross-loop learning — a creative win informs the landing page test
Coordination
Experiment sequencing — what to test next across all four loops
Contamination avoidance — isolate tests so results stay clean
Capital allocation — budget distributed by expected value
Guardrails — spend caps, loss limits, kill switches per experiment
AI Models
Claude — reasoning, strategy, hypothesis generation, decisions
Gemini — video analysis, multimodal creative assessment
OpenAI — creative generation, copywriting, image variants
OpenRouter — model routing, cost optimisation, fallback chains
Infrastructure
PostgreSQL
Redis
VPS Compute
Object Storage
Meta Ads API
Google Ads API
Shopify
Klaviyo
GA4
Webhooks
50+
Controlled experiments per day, per ad account, across all four loops
Across 3,000 members that's 150,000+ experiments per day — over a million per week — all feeding back into the shared decision layer. The orientation layer doesn't just learn from one brand. It learns from every brand, every vertical, every spend level, simultaneously.
04
The Partnership
The collaborative build
A Player Labs
commits
Startup speed + the engine
- Agentic flywheel system — orientation, coordination, and infrastructure layers
- Dedicated team building exclusively on Auto Throughput
- Existing components (MOAC, classification, spend analysis) already integrated
- Ship in weeks, not quarters — continuous delivery, working software every sprint
A startup-speed engineering team that’s already deep in EE’s world — with working components ready to integrate.
Ecommerce Equation
unlocks
The signal layer — 3,000 members
- 3,000 ecommerce businesses generating real experiment data every day
- Growth frameworks that become the policy layer inside the engine
- Vertical exclusivity — the benchmark moat no competitor can replicate
- Domain expertise that shapes which experiments the system runs first
EE’s frameworks are the intelligence. Auto Throughput is the engine that compounds them through evidence — at a scale no human team can match.
What the partnership unlocks
Now
4 agentic systems — one autonomous agent per loop: creative, offers, pages, media
Next
The loop engine — 5th agent stitches all four together, coordinates and captures cross-loop insights
Then
Signal layer live — 3,000 members’ experiment results feeding back into orientation
Moat
Benchmark network — anonymised intelligence across verticals that compounds with every member
05
The Opportunity
First-mover advantage
The ecommerce education company whose members get autonomous growth experiments first will own the benchmark data for the entire vertical. That data compounds. The window to be that company is now.
What EE gets
- Exclusive access to Auto Throughput for the ecommerce education vertical
- Non-competitive carve-out — no other education community gets this
- Product shaped by EE’s frameworks, tested on EE’s members, built for EE’s use case
- 150,000+ experiments per day feeding benchmarks no competitor can replicate
- First product in market while others are still hiring engineers
What APL commits
- Startup speed — ship in weeks what traditional teams quote in quarters
- Dedicated team building exclusively on Auto Throughput
- Existing components already working (MOAC, classification, spend analysis) — not starting from zero
- Continuous delivery — working software every sprint, not a 12-month waterfall
- Full stack execution from AI models to infrastructure to member-facing product
This is built for EE. The frameworks, the members, the data — no one else has what it takes to make this work. Let’s build it together.
Let’s talk.
06