GrowthLab AI — Marketing Strategy Simulator
An AI-powered business simulation and marketing strategy lab that combines a weekly simulator, marketing war room, multi-agent agency, and experimentation pipeline into one SaaS-style dashboard
GrowthLab AI is a portfolio-grade growth-operator simulator where you run Lumora Skincare, a Series A DTC brand with a $250k monthly marketing budget,...
A Simulator, Not a Live Ops Platform: GrowthLab is an educational business simulator built for recruiters, hiring managers, and growth teams who want to see how marketing decisions compound over time. The default experience runs on deterministic local mock data with no database and no live ad or commerce connectors. Optional OpenAI integration lets users paste their own API key for live LLM interpretation of custom plays and multi-agent debates, with graceful fallback to the offline engine.
Why I Built It: Growth teams juggle diagnosis, debate, experimentation, and budget decisions across disconnected tools, with no safe place to practice high-stakes calls. Most AI demos are chat wrappers. They do not model CAC/LTV economics, creative fatigue, cohort decay, or experimentation guardrails. I wanted one artifact that shows product thinking, marketing domain depth, and systems design in a workflow you can actually play with.
The Problem: Growth operators need to connect what changed in the business, why it happened, what to test, and what to decide, often under pressure and with incomplete information. Dashboards show metrics; they rarely teach the decision loop. I wanted to compress diagnose, debate, experiment, and decide into a coherent product experience, not another static case study.
The Solution: GrowthLab combines four concepts into one dashboard: an AI business simulator, a marketing war room, a multi-agent marketing agency, and an autonomous experimentation lab. You advance week by week through Lumora Skincare's season, review structured root cause analysis, hear from seven specialist agents, evaluate a fully specified experiment, preview the impact of your choice, and commit a decision that feeds the next week's metrics through cross-metric cascades.
How the Simulation Works: Each week starts by applying event deltas to current metrics, for example a Meta CPM spike pushing CAC up. You review the war room diagnosis, agent debate, and proposed experiment, then choose among strong, risky, weak, wait-and-watch, or a custom free-text play with impact preview. The engine applies decision deltas, runs second-order cascades such as rising CAC eroding LTV and ROAS, clamps metrics to realistic ranges, and records the outcome. After Week 8, a final summary grades your best and worst calls and surfaces unresolved risk.
Multi-Agent Marketing Agency: Seven specialist agents, spanning CMO, Performance, Creative, Lifecycle CRM, Customer Insights, Experiment Designer, and Finance, give opinionated recommendations each week, each tagged with a stance. A consensus matrix shows where they align and disagree, making trade-offs visible before you decide. With a connected OpenAI key, debates can switch from authored demo content to live LLM-generated arguments through a thin proxy.
Test & Learn and Experimentation: The Adaptive Experiment Lab follows a validation-gated pattern inspired by my Dell capstone work: pre-run checks for sample adequacy, guardrail exposure, timeline fit, and prior confidence before traffic is committed. A test and learn pipeline supports max two concurrent tests, pre-registered runtime with no early stopping, directional interim reads at the midpoint, and win, partial, or flat outcomes that land on metrics only when a test concludes.
Architecture & Tech Stack: Built with Next.js App Router, TypeScript, Tailwind CSS, shadcn/ui, Zustand with localStorage persistence, and Recharts. The deterministic simulation engine handles health scoring, decision math, cascades, and seeded synthetic scenario generation. A metric registry acts as the single source of truth across dashboard, campaigns, war room, agents, experiments, insights, and timeline views. Deployed on Vercel with 37+ unit tests on the simulation engine.
What I Learned: The biggest takeaway was designing a growth workflow as a system, not a collection of screens. Metrics need causality, decisions need previewable consequences, experiments need guardrails, and agents need structured disagreement. The future of marketing AI is not a single chatbot answer, but operator tooling that helps teams reason through economics, risk, and trade-offs in a loop they can trust and replay.
Runs on deterministic local mock data by default. Optional OpenAI key unlocks live decision interpretation and agent debates; no real GA4, Meta, Shopify, or Klaviyo data is connected out of the box.