Applied AI Roadmap — three tracks, three tempos
Operating model
Three parallel tracks, each moving at its own tempo. These are not sequential phases — they run concurrently, coupled by dependency.
- Track A — Culture & adoption (people). Tempo: continuous, human-paced, never “done.” Paces how fast workflows can be handed to colleagues. → the organization-maturity pillar.
- Track B — Infrastructure & system (the substrate). Tempo: slow, phased. Built incrementally, in step with what adoption will actually use. → the data-maturity pillar + the knowledge backbone.
- Track C — Product & workflow (the applications). Tempo: fast. Ship → test → iterate. The compounding engine — deploy as early as each item’s substrate allows. → the Applied AI surfaces.
Tracks A and B produce the shared context/knowledge layer; Track C consumes it.
The sequencing rule
Track C never waits for Track B to be “finished.” Each product workflow ships the moment the specific substrate rung it needs exists. Order Track C by least substrate required → most, banking compounding early while A and B grind underneath.
The gating law applies per item: a workflow’s realized value is capped by the weaker of {infrastructure it needs, adoption it needs}. Matching each product item to its required rung is its place in the sequence.
Two bars on Track C: shipped (it works, value proven) and handed off (a colleague runs it without you). Ship at product-speed; hand off at culture-speed.
Phased plan
Phases are waves, not gates — at any moment all three tracks are moving. Each phase shows what goes live on each track.
Phase 0 — Foundation (now)
- A — Culture: Stand up your own bench (Obsidian + Git, Codex, Cowork, Antigravity). Draft the onboarding path. First internal session on the bench.
- B — Infra: Knowledge backbone first — vault structure, naming convention, Git repo, core MCP servers wired. Provision managed accounts + tokens (Meta System User @ Business Manager, Google MCC, TikTok Business Center). Registry seed tables.
- C — Product: Reporting Tier 1 ships — Cowork + ads MCP, no warehouse required. First compounding: marketers get fast reporting on day one.
- Success signal: someone pulls ad performance through Cowork with zero manual export.
Phase 1 — First loops
- A: Train marketers on Tier-1 reporting + ideation. Begin Own → Iterate → Share on these two.
- B: Warehouse Phase 1 — one brand end-to-end, live dashboard within ~2 weeks. Seed the creative infrastructure: starter angle library, raw-material catalog, performance ledger (Obsidian).
- C: Ideation workflow ships (scrape followed brands + Foreplay, pattern/format breakdown). Creative production begins in draft.
- Success signal: first agent-drafted brief cleared at Gate 2; one live dashboard.
Phase 2 — Scale the factory
- A: Operators own the creative + social workflows; you shift from running to supporting.
- B: Warehouse fans out to all brands; multi-currency / US region activated. Request Shopee + TikTok Shop developer IDs; ingest.
- C: Creative production matured (finished Canva assets + copy variants; performance ledger closing the loop). Social media management ships — growth loop + operational track, approval gates, Zernio publishing.
- Success signal: marketers running produced creatives in their own accounts; approved calendar publishing through Zernio.
Phase 3 — Depth & exploratory
- A: Self-sustaining Own → Iterate → Share; support channel at steady state.
- B: dbt tests, scheduled narrative reports, anomaly alerting. Warehouse hardened as MMM feature store (historical panel accruing).
- C: Reporting Tier 2 fully live — unified ads + WooCommerce + fighter.my, MER-based. Explore video generation (UGC-style, pairs with ideation) and CRM/CS automation.
- Success signal: MER-based decisions replacing last-click reads; first exploratory workflow in test.
Dependency map — what gates what
| Product item | Requires | Ships |
|---|---|---|
| Reporting Tier 1 | Cowork + ads MCP only | Phase 0 |
| Ideation | Vault + scraping + Foreplay + starter angle library | Phase 1 |
| Creative production | Raw materials + angle library + performance ledger + Canva MCP | Phase 1–2 |
| Social management | Vault governance + Zernio MCP + approval workflow | Phase 2 |
| Reporting Tier 2 | The warehouse (Phase 2 fan-out) | Phase 3 |
| Video / CRM-CS | Own tooling + data | Phase 3+ |
Assumptions made (flag to correct)
- Tier-1 reporting ships first because it needs zero warehouse and returns value immediately. If a different quick win matters more, it moves.
- Infra tempo assumes the warehouse follows your Phase 0–4 plan; adoption paces it, so dates flex.
- Social management sits in Phase 2 because it depends on both vault governance and Zernio being live; if governance is ready sooner, it pulls forward.
- The right surface (AI-as-product) stays exploratory — CS/CRM is the only near-term touch, and it’s more internal-leverage than sold product.