AdPilot product plan & PRD
AdPilot product plan & PRD
Status: v0.1 (prototype complete) · September 2026 Evidence base: docs 01–05 research. This doc is the decisions; the research docs are the reasons.
1. Product thesis
A small business owner should be able to run competent, multi-platform advertising by having a conversation — not by mastering four dashboards.
AdPilot is an AI media buyer in your pocket. You describe your business and goal in one sentence; it plans the campaign, writes the creative per platform, launches through official ad APIs, watches the numbers daily, and rebalances budget with your approval. You stay in control with caps, approval cards, and a full audit log — the agent does the work you'd otherwise pay an agency $500+/month for.
Positioning statement
For small business owners, indie developers and creators who want customers but not a marketing job, AdPilot is an AI ad agent that runs your advertising across platforms for you. Unlike dashboards (Meta/Google Ads Manager) it speaks plain language and works across platforms; unlike Advantage+/PMax it decides between platforms, not just inside one; unlike agencies it costs less than one day of ad spend.
The three whitespace opportunities (validated in research)
- Cross-platform budget orchestration for micro-budgets — no product below enterprise pools/rotates a $300–1,000/month budget across platforms.
- Chat-first agentic operator at SMB prices — conversational ad management exists only at quote-only enterprise tiers.
- Mobile-first + one-prompt end-to-end — Zeely (4.58★, 21,828 ratings) proves mobile demand, but is single-platform.
2. Personas
P1 — "Sam", e-commerce side-hustler (primary)
- Sells candles/jewelry/apparel on Shopify/Instagram. Age 24–40.
- Spends $200–800/month on ads, checks everything on her phone.
- Fired a $600/month "marketing guy" after no results. Tried Ads Manager, got lost in campaign types.
- Success: "I told it what I sell, tapped twice, ads were running. It texts me when something's off."
P2 — "Dev", indie developer (secondary)
- Solo dev with a launched app, $100–500/month UA budget, technical but not marketing-savvy.
- Hates GUI dashboards, loves the chat interface and API transparency.
- Success: CPI targets hit without babysitting; budget auto-shifts to the winning channel.
P3 — "Maya", local service provider (tertiary, v2)
- Salon/cleaners/tutor. Wants "more bookings", zero ad vocabulary.
- Needs lead-form campaigns + local radius targeting; WhatsApp-simple UX expectations.
3. Core loop (what the product does)
describe business/goal → plan (budget split + reasoning)
→ approve → launch on N platforms (official APIs)
→ monitor daily (anomaly alerts, weekly plain-language report)
→ propose rebalance → you approve → applied + logged
Supporting surfaces: Content Studio (per-platform copy + image/video briefs), Campaign detail (KPIs vs benchmarks, budget stepper, pause), Audit log, Settings (caps, connected accounts, AI engine).
4. MVP scope — what's built in this prototype
| Area | In the prototype | Notes |
|---|---|---|
| Chat copilot | ✅ intent detection (launch/optimize/report/question), plan cards, suggestion cards | on-device heuristic engine; live-LLM mode via OpenAI-compatible config |
| One-prompt planner | ✅ budget/goal/product/platform-hint parsing, category inference, budget split w/ minimums, KPI targets | deterministic + testable (22 tests) |
| Content generation | ✅ hooks/headlines/body/CTA/hashtags per platform, image briefs + ready image prompts | template library; LLM/image-API upgrade path documented |
| Launch flow | ✅ per-platform adapters (mock sandbox) with realistic ids + latency | real adapters: interface defined, sequencing in docs/04 |
| Monitoring | ✅ deterministic metrics simulation, alerts (CPA/ROAS swings), weekly report | demo sim replaces platform insights calls |
| Optimizer | ✅ efficiency-scored rebalance, ±30% caps, platform minimum floors, apply flow | the flagship "cross-platform" feature |
| Trust rails | ✅ spend caps in guardrails, sandbox-by-default, full audit log, human approval for money actions | trust checklist in docs/05 |
| Multi-platform | ✅ iOS/Android/web from one Expo codebase | EAS Build ready |
5. Roadmap
v0.1 — prototype (done, this repo)
Chat planner, content studio, sandbox launch/optimize, dashboards, audit log, backend blueprint.
v0.5 — private alpha (4–6 weeks)
- Google Ads + Microsoft Ads real integrations (GREEN verdicts: basic developer token ≈ 5 business days; Microsoft public sandbox token works day 1).
- Real metrics via platform insights endpoints; alerts wired to push notifications.
- LLM planning by default (server-side), heuristic fallback; image generation via gpt-image/Flux for creatives.
- Stripe billing + auth. TestFlight / Play internal testing.
v1.0 — public beta (Q4 2026)
- Meta Marketing API (start App Review + 500-call warm-up immediately at v0.5; ~1–3 months to full access).
- TikTok (business verification ≈ weeks–2 months) — the SMB-reach platform.
- Weekly-report push, one-tap "autopilot with caps" mode, referral loop.
- Pricing live: Free (1 sandbox workspace) / $19 Starter / $49 Growth (docs/09).
v1.x — moat building (2027)
- Creative performance feedback loop (which hooks convert per category → better generation).
- Agency seats + multi-brand workspaces; WhatsApp/Telegram copilot surface for P3.
- Category playbooks (restaurants, salons, app launches) as first-class templates.
6. Key metrics
| Stage | Metric | Target |
|---|---|---|
| Activation | % of new users who launch a campaign within first session | > 40% |
| Core value | Median days from launch to first confirmed conversion | < 10 |
| Retention | M2 retention of paying users | > 70% |
| Agent quality | % of AI actions accepted without edit (approvals) | > 60% |
| Trust | % of accounts hitting spend-cap incidents | < 1% |
| Revenue | Blended ARPU month 6 | ≥ $30 |
7. Risks & mitigations (summary — full register in docs/05)
| Risk | Mitigation |
|---|---|
| Platform API approval delays (Meta warm-up) | Launch on Google/Microsoft first; sequence per docs/04 verdicts |
| Novice burns budget, blames product | Hard caps, ±30% rebalance bound, sandbox first, approval gates |
| Platform policy change / "wrapper" risk (cf. Jasper) | Own the cross-platform orchestration + data loop, not just model calls |
| AI-content labeling rules (TikTok mandatory labels 2026) | Disclosure metadata pipeline in creative export |
| Apple/Google IAP cut on mobile subscriptions | Web-first checkout for subscriptions where permissible |