MemberIntel V1 Specification

▶ Watch the 46s summary — V1 Spec — the advisor product

Status: Draft v0.1 (2026-04-29)
Owner: Blair (CEO, MemberPress / Caseproof)
Working name: MemberIntel (internal — not committed)

---

1. Overview

MemberIntel V1 is a paid SaaS advisor for existing MemberPress site operators. It connects to the customer’s MemberPress site and Stripe account, provides a dashboard of key membership metrics, and runs an AI chat that gives MP-specific, context-grounded business and technical advice. The advice is powered by a two-tier “brain” — a curated global knowledge corpus shared across all customers, plus a per-customer brain that accumulates context, actions, and learnings about that customer’s specific site over time.

V1 ships under a Connected Free + Pro model: an always-free tier for MemberPress customers (capped chat, basic dashboard, weekly digest, brain contribution) paired with a $29/month Pro tier (unlimited Sonnet chat, live insight dashboards, advanced reports, brain editing). The free tier exists to drive rapid adoption across the MemberPress customer base — feeding the global brain’s data flywheel — without bleeding money on the expensive parts of the product. The Pro tier captures revenue at the point of felt value via natural upgrade moments (chat-cap hit, advanced-report request, etc.).

V1 deliberately drops everything that depends on Caseproof’s pre-existing proprietary data (cross-customer Stripe Connect aggregation, cross-customer telemetry benchmarks, third-party SEO data) until legal coverage and product-market fit are confirmed. It also drops AI write/agent actions and the first-time-creator flow; both are deferred to V1.5.

The strategic bet: a curated global brain plus a per-customer brain that grows with use, fed by a thin always-free tier that captures data inputs cheaply, creates a stronger moat than pre-existing proprietary data would — because it’s built into the product mechanics (network effects from use), doesn’t require legal cover for pre-existing data, and is hard to copy without time and scale.

---

2. Goals

Primary

• Help existing MemberPress operators run their business more effectively, by giving them a smart MP-aware advisor that knows their specific site.
• Build a self-improving knowledge system (per-customer brain + global brain + cross-pollination) that becomes more valuable the longer customers use it and the more customers use it.
• Maximize MemberPress customer adoption of MemberIntel quickly to feed the brain — without burning cash on free users at scale.

Secondary

• Establish a paid SaaS revenue line attached to MemberPress.
• Improve MemberPress retention by giving customers ongoing value beyond the plugin.
• Build the data foundation needed to make a strong cross-platform pitch in V2 (“we have analyzed and learned from N thousand membership sites”).
• Lay technical foundations for V1.5 (agent / write-actions, first-time creator path).

V1 Success Metrics (priority order)

1. Free tier signups — % of total MemberPress customer base who connect to MemberIntel within 6 months of launch. Target floor: 10% in 6 months.
2. Free → Pro conversion rate — % of free users who upgrade to Pro within 60 days of activation. Target floor: 5%; goal: 8–10%.
3. Pro Weekly Active Users (WAU) — % of Pro users who use MemberIntel weekly. Target: 70%+.
4. Brain growth — number of approved global-brain entries (seeded + curated + cross-pollinated) at month 6. Target: 200+.

---

3. Non-Goals (explicit V1 exclusions)

• No agent / write actions. AI does not make changes inside the customer’s MP site in V1. Advisor only — generates step-by-step instructions; customer executes manually. Agent capability is V1.5.
• No first-time-creator / greenfield onboarding wizard. V1 serves existing operators only. Greenfield is V1.5 (paired with agent mode, since setup automation requires write actions).
• No cross-customer benchmarks built from pre-existing Caseproof Stripe Connect / MP telemetry data. Avoids the legal Phase 0 dependency entirely. Cross-pollinated learnings through the global brain (Section 6) are the substitute moat.
• No third-party paid data integrations. No SEMrush, Ahrefs, etc. Public site analysis is Claude-only.
• No non-MemberPress platforms. Kajabi, Thinkific, PMP, MemberMouse: V2+.
• No fine-tuning of LLM weights. All “learning” happens through brain content updates and retrieval; no model training in V1.
• No SOC 2. GDPR + CCPA only.
• No mobile app. Responsive web only.
• No public marketplace of case studies / community-shared brain entries.
• No tiered Pro plans in V1 (single $29/mo Pro tier; tiering is a post-launch lever).

---

4. ICP & Users

V1 ICP: Existing MemberPress operators

• Small to mid-sized membership site owners running an active MP site.
• They have data (members, transactions, content) and are looking for ways to grow / improve / not break things.
• They do not have time, budget, or expertise for a fractional consultant or full analytics setup.

Free-tier user profile: existing MP customer who’s curious about MemberIntel but not yet ready to pay. Wants to see whether the product understands their site and adds value before committing. Their data accumulates in their per-customer brain and contributes (anonymized) to the global brain via cross-pollination.

Pro-tier user profile: MP operator who wants the full advisor experience — unlimited Sonnet chat, live insights, advanced reports, brain editing — and is comfortable with the $29/mo investment.

The free → Pro funnel is the central conversion motion of V1.

Acquisition channels

• In-MemberPress-admin banner / menu item promoting MemberIntel.
• Email blast to MP customer base.
• memberintel.com as the canonical destination (signed-in app + marketing pages).

Positioning

Working pitch language (refine before launch; avoid claims that imply training on individual customer data):

MemberIntel is built on insights from running one of the world’s leading WordPress membership platforms — and bakes those insights directly into your site so you get business advice and technical help grounded in your actual MemberPress data.

Honesty rules in marketing copy:

• Never imply that MP usage data (from customers who haven’t signed up for MemberIntel) is used for AI training or benchmarks. It isn’t.
• Be clear that MemberIntel customers’ data is used per the MemberIntel ToS — including accumulating in their per-customer brain and contributing anonymized, abstracted learnings to the global brain (with opt-out).

---

5. Pricing & Packaging

5.1 Two Tiers

	Free (Connected Free)	Pro ($29/mo per site)
Eligibility	MP customers who explicitly opt in (no auto-enrollment)	Anyone (MP customer or not, once non-MP support arrives)
MP data sync	Yes — basic, monthly snapshot	Yes — live, daily refresh
Stripe data sync	Yes — basic, monthly snapshot	Yes — live, daily refresh
Public site analysis	Yes — weekly cached	Yes — weekly cached + on-demand re-analysis
Dashboard	Read-only, monthly snapshot, basic KPIs only	Full dashboard with AI-curated insight cards, daily refresh
AI chat	5 messages/month, Haiku model	Unlimited, Claude Sonnet
Weekly digest email	Yes (Haiku-generated)	Yes (Sonnet-generated, longer + deeper)
Per-customer brain	Yes — accumulates passively	Yes — fully editable by customer + AI
Global brain access	Browse-only (read-only)	Browse + cited in chat responses
Brain cross-pollination contribution	Yes (default-on, opt-out available)	Yes (default-on, opt-out available)
Advanced reports (cohorts, custom ranges, comparisons)	No	Yes
Priority support	No (community / docs only)	Yes
V1.5: agent / write-actions	No	Yes (when shipped)

5.2 Trial Mechanics

• No “free trial of Pro” for free-tier users. The free tier exists permanently; users upgrade when they want more. This keeps the flywheel filling without trial-expiration churn.
• For new direct-to-Pro signups (e.g., via memberintel.com marketing without an MP install): 14-day free trial of Pro, no credit card at signup, card requested on day 14.

5.3 Conversion Mechanics (the upgrade moments)

The free tier is designed to create natural, high-intent upgrade moments rather than rely on a hard paywall:

• Chat cap hit. When a free user sends their 5th chat message in a month, the AI’s response includes “You’ve used your 5 free messages this month. Pro gets you unlimited Sonnet-quality answers — upgrade for $29/mo.” High intent.
• Insight cliffhanger. Dashboard insight cards on free tier show 1-line summary; full prose explanation + recommended action requires Pro click-through.
• Report request. Free user asks for “a report on…” → response is a one-liner + “Detailed reports are a Pro feature.”
• Live data nudge. Free dashboard shows a stale-data badge (“last sync: X days ago”) → Pro shows live (today).
• Brain edit nudge. Free user views their per-customer brain → can read; “Edit goals, add notes — Pro feature.”

5.4 Free Tier Cost Economics

Estimated per-free-user monthly cost:

Cost driver	Estimate
Public site analysis (weekly Claude fetch + extraction, cached)	~$0.20/mo
5 chat messages with Haiku	~$0.10/mo
Weekly digest with Haiku	~$0.20/mo
Data sync infra + storage + auth	~$0.50/mo
Cross-pollination compute (amortized)	~$0.10/mo
Total per free user	~$1.10/mo

Estimated per-Pro-user monthly cost:

Cost driver	Estimate
Sonnet chat (avg ~50 turns/mo at moderate length)	~$3–8/mo
Daily site analysis refresh + on-demand	~$0.50/mo
Sonnet-generated digest + insight cards	~$1.00/mo
Live data sync infra	~$1.00/mo
Storage, brain ops, support	~$1.00/mo
Total per Pro user	~$6–12/mo

Pro contribution margin: approximately $17–23/mo per Pro user (after Stripe fees).

5.5 Cost Discipline (must-build, not nice-to-have)

Given the freemium model, hard cost controls are required to prevent runaway losses:

• Per-user per-month token budget caps, hard-enforced server-side. Free: tight cap. Pro: generous but not unlimited (e.g., 20K Sonnet output tokens/day before throttling).
• Aggressive caching: site analysis cached 7 days; insight cards cached until next data refresh; digest pre-computed.
• Haiku-only for free tier chat. Server-side enforcement — never accidentally route free users to Sonnet.
• Monthly cost-per-cohort dashboards (free vs. Pro) reviewed weekly during early launch to catch outlier users.
• Abuse prevention: rate limits, captcha for high-volume requests, account flagging for anomalous usage patterns.

5.6 Long-term Pricing Levers

Not committed in V1 but planned for evaluation post-launch:

• Pro Annual ($290/yr — 2 months free) once retention data is solid.
• Pro+ tier ($79–99/mo) when V1.5 lands agent / write-actions and greenfield wizard.
• Agency tier for multi-site operators (per-seat or volume-tiered).
• Free-tier sunset clause — reserve the right to cap or remove the free tier in the future if economics deteriorate. Disclosed in ToS.

---

6. The Brain (core architecture)

The brain is the knowledge system that backs every AI response. It is two-tiered.

6.1 Global Brain (shared across all customers)

• What’s in it: seed content + curated playbooks + promoted learnings.
  • Seed content (Phase 0 / 1): existing MemberPress documentation, knowledge base articles, blog posts, support article archive — indexed.
  • Curated playbooks (ongoing): hand-written by the dedicated content lead. Target ~50 in V1, growing toward 200+ post-launch. Examples: “How to launch a coaching membership,” “How to reduce churn on annual plans,” “How to price a community-only tier,” “How to migrate from $X tier structure to $Y.”
  • Promoted learnings (ongoing): anonymized, abstracted patterns surfaced from per-customer brains via the cross-pollination process (Section 6.3).
• Format: markdown files, source-of-truth in a Git-backed repository (or equivalent versioned store). Edits through a content authoring UI for non-engineers.
• Indexing: files chunked + embedded into a vector store (e.g., pgvector or Pinecone — final choice during build). Re-indexed on edit.
• Retrieval at query time: semantic search over the global brain returns top-K relevant chunks; included in the LLM prompt context.

6.2 Per-Customer Brain (scoped to one customer)

• What’s in it: everything the system has accumulated about a specific customer.
  • site-profile.md — derived from Claude-based public site analysis (niche, audience, positioning, current content patterns, current pricing structure).
  • goals.md — the customer’s stated goals, captured during early chat sessions and surfaced for review.
  • actions-log.md — actions the customer has reported taking (or that MemberIntel sees in their MP data) and outcomes.
  • insights.md — observations the AI has noted about this specific customer’s situation.
  • feedback.md — thumbs up/down + comments from the customer on past responses.
• Format: markdown files, stored in MemberIntel’s database (per-tenant isolated), version-history tracked.
• Indexing: per-customer vector store partition. Retrieval is always scoped to the active customer’s brain — never cross-tenant.
• Update mechanism: updated automatically as the customer chats (e.g., “I tried tiered pricing last month and churn went up” gets logged), reviewed by the AI for stored insights, and refreshed when the public site is re-analyzed.
• Free-tier note: per-customer brains accumulate identically for free and Pro users. Free users can read but not edit. Pro users can read and edit.

6.3 Cross-Pollination (global brain growth from per-customer brains)

• Purpose: capture learnings that emerge from individual customers and make them available to all customers in anonymized form.
• Mechanism: scheduled job (weekly or monthly) runs Claude over per-customer brain entries with positive feedback signals. It drafts candidate global-brain entries — abstracted, anonymized, with no customer-identifying details. Drafts go into a content lead review queue.
• Approval: content lead reviews each candidate, edits as needed, approves or rejects. Approved entries enter the global brain; rejected entries are discarded.
• Privacy: no per-customer brain content ever enters the global brain in raw form. No customer is ever identifiable. Customer ToS explicitly authorizes this anonymized promotion. Customers can opt out (and lose access to the cross-pollinated portion of the global brain in exchange).
• Audit: every approved global-brain entry has a traceable origin (source customer brain entries, anonymized) for compliance.
• Signal weighting: cross-pollination favors high-signal sources. A Pro user with hundreds of chat turns and rich feedback contributes more candidate material than a free user with 5 chats/month. This is automatic via volume but also explicit in the candidate-scoring step.

6.4 Why This Is the Moat

• More customers using MemberIntel → more raw material for cross-pollination → smarter global brain for everyone.
• Longer per customer → richer per-customer brain → more personalized advice for them specifically.
• Both effects compound. Both are hard to copy without scale and time.
• Built from the product mechanics, not from pre-existing proprietary data — so it ships without legal Phase 0 gating.
• The Connected Free tier accelerates this: every free user is a data contributor.

---

7. User Stories

7.1 Free-tier activation & onboarding

• US-F1. I’m an existing MP customer. I see a banner in my MemberPress admin offering free MemberIntel. I click and arrive at memberintel.com signed in via my MP license — no signup form to fill out.
• US-F2. I confirm consent to data sync + brain contribution (default-on opt-out for cross-pollination). MemberIntel begins ingesting my MP + Stripe data and runs an initial public site analysis.
• US-F3. Within ~5 minutes I see a basic dashboard with my monthly KPIs, and I receive my first weekly digest email the following Monday.
• US-F4. I open chat and ask my first question. I get a Haiku-quality answer. I have 4 free messages remaining this month, shown clearly in the chat header.
• US-F5. I hit the 5-message cap. I see an inline upgrade prompt: “You’re loving this — Pro gives you unlimited Sonnet-quality answers, $29/mo.” One-click upgrade flow.
• US-F6. I never upgrade — that’s fine. I keep getting the digest weekly, my per-customer brain accumulates, my data flows in, the global brain benefits.

7.2 Pro-tier daily / weekly use

• US-P1. As a Pro user, my dashboard refreshes daily with KPIs and 3–5 AI insight cards based on what’s actually happened in my data. Every number has prose context next to it.
• US-P2. I open chat and ask “Why did my churn spike last week?” — the AI answers grounded in my actual data and my per-customer brain (e.g., “On April 15 you raised the price on Tier 2 — 8 members canceled within 48 hours, all on Tier 2”). Responses cite the data they reference.
• US-P3. I receive a personalized Sonnet-generated weekly digest every Monday morning summarizing what changed, what’s working, and 1–3 prioritized actions for the week.
• US-P4. I can request advanced reports: cohort analyses, custom date ranges, comparisons between tiers / time periods.
• US-P5. I can edit my per-customer brain — add goals, log actions, annotate insights — so the AI’s future answers reflect my context.
• US-P6. I thumbs-up or thumbs-down every AI response. My feedback is stored in my per-customer brain and informs future responses.

7.3 Privacy & control (all users)

• US-C1. In Settings I can see exactly what data MemberIntel has on me (MP data sync status, Stripe sync status, public site analysis snapshot, per-customer brain entries). Pro users can edit per-customer brain entries; free users have read-only access.
• US-C2. I can opt out of cross-pollination contribution. If I do, I lose access to the cross-pollinated portion of the global brain (clearly disclosed).
• US-C3. I can export all my data (CSV / JSON). I can delete my MemberIntel account, which removes my personal data and my per-customer brain.

7.4 Billing & tier movement

• US-B1. I upgrade from Free to Pro in two clicks: confirm + enter card. Card on file, monthly billing begins immediately, full feature access live within seconds.
• US-B2. I downgrade from Pro to Free in Settings. My account remains, my per-customer brain remains, my entitlements drop to free immediately. Settings clearly explain what changes (live data → monthly snapshot, Sonnet → Haiku-capped, advanced reports → locked, etc.).
• US-B3. A Pro charge fails (expired card). MemberIntel sends standard dunning email sequence; access to Pro features pauses after 3 failed attempts with a 7-day grace period; if still unresolved, account drops to Free.

7.5 Cross-cutting

• US-X1. I can chat with the AI in English, Spanish, or German — and the UI is fully translated to those languages.

---

8. Data Architecture

8.1 Sources (V1)

Source	How	Use
Customer’s MemberPress site	OAuth via license / dedicated MemberIntel API endpoint on MP. Live sync (Pro) or monthly snapshot (Free).	Per-customer dashboard, chat context, brain enrichment.
Customer’s Stripe account	Stripe Connect (if on Caseproof platform) OR OAuth into customer’s own Stripe.	Revenue / churn / dunning metrics.
Customer’s public website	Claude-based fetch + structured extraction at the customer’s site URL. Initial + weekly refresh.	Site profile in per-customer brain (niche, audience, positioning, content / pricing patterns).
Global brain	Curated by content lead.	Retrieved at query time as background context for chat and dashboard insights.
Per-customer brain	Auto-generated from chat, MP / Stripe data observations, site analysis, feedback.	Retrieved at query time as personalized context for that customer.

8.2 Data Sync Cadence (tier-gated)

• Free tier: monthly snapshot of MP and Stripe data. Public site analysis weekly (cached).
• Pro tier: daily live sync of MP and Stripe data. Public site analysis weekly + on-demand.
• Engineering implication: the sync pipeline must support both modes, with cadence determined by entitlement (Section 12).

8.3 Strict Data Boundaries

• MP customers who have never signed up for MemberIntel: their MP / Stripe data is not used for any MemberIntel feature, brain, or analysis. Period. Marketing must be careful to never imply otherwise.
• MemberIntel customers (free or Pro): their data is used per the MemberIntel ToS, which authorizes (a) per-customer brain accumulation, (b) anonymized promotion to global brain via cross-pollination (opt-out available), (c) feedback signal capture, (d) standard product operation.
• No fine-tuning of LLM weights on any customer data in V1, regardless of ToS coverage.

8.4 AI Architecture

• Pattern: retrieval-augmented LLM with structured tool calls.
• Models: Claude Sonnet 4.6 for Pro chat + Pro insight cards + cross-pollination drafting. Claude Haiku 4.5 for free-tier chat, weekly digest generation, classification, summarization.
• Model routing is tier-gated, server-side. Free users always route to Haiku for chat. Pro users route to Sonnet. Never user-configurable.
• Tool surface (V1):
  • query_customer_metrics(metric, time_range) — SQL into the per-customer warehouse.
  • analyze_site(site_url) — fetch + structured Claude analysis. On-demand (Pro) or scheduled (Free + Pro).
  • search_global_brain(query) — vector search over the global brain.
  • search_customer_brain(query) — vector search over the active customer’s brain (always tenant-scoped).
  • update_customer_brain(section, content) — write into the active customer’s brain (e.g., log a new action, update goals).
• Prompt versioning & evals: prompts are version-controlled. Offline eval suite runs against representative customer scenarios on each release. Hallucination guards: chat answers must cite the data they reference (data ID, time range).
• Continuous improvement loop: thumbs feedback → preference data → drives prompt iteration, retrieval tuning, global-brain authoring, cross-pollination. No fine-tuning in V1.

8.5 Cost Controls

• Per-user monthly token budget caps (specific numbers TBD during build, modeled in Section 5.4).
• Aggressive Haiku usage for digest generation and free-tier chat.
• Brain retrieval is always pre-filtered before going to the LLM (no full-corpus dumps).
• Public site analysis cached aggressively (weekly refresh, not per query).

---

9. User Interface

9.1 Surfaces

• memberintel.com — primary destination. Marketing site (logged-out, includes pricing page) + signed-in app.
• MemberPress admin banner / menu item — entry point for existing MP customers. Single-click signs them in via MP license.
• Weekly digest email — Monday-morning AI-written summary (Haiku for Free, Sonnet for Pro).

9.2 Key Screens

• Dashboard.
  • Free: read-only, monthly snapshot, basic KPIs (MRR, members, churn, signups, conversion). “Last sync: X days ago” badge. Pro-only insight cards shown as 1-line teasers.
  • Pro: live, daily refresh, full KPIs + 3–5 AI insight cards with prose explanations.
• Chat panel. Always available in sidebar; expandable to full-screen. Threaded history. Inline charts / tables / sources cited.
  • Free: 5 messages/month, Haiku, usage indicator in header (“3 of 5 free messages used”).
  • Pro: unlimited, Sonnet.
• Brain view. Read/write access to the per-customer brain.
  • Free: read-only.
  • Pro: read + edit (goals, notes, action log).
  • Both: global brain browse-only.
• Reports.
  • Free: not accessible; clicking shows Pro upsell.
  • Pro: cohort analyses, custom date ranges, comparison reports.
• Settings. Account, billing, data sources, privacy controls (cross-pollination opt-out, data export, account deletion), notification preferences, language selection, tier status + upgrade/downgrade button.

9.3 Tier UI Patterns

• Tier badge in app header (Free / Pro) with one-click upgrade button.
• Upgrade nudge surfaces built into chat (cap hit), dashboard cards (Pro-only insight teasers), reports section (locked), brain editor (locked).
• Usage indicators for free tier: “X of 5 free messages used this month” in chat header.
• Downgrade flow for Pro → Free. Cancellation must be self-serve. Settings explain what changes when downgrading.

9.4 Visual / Brand

• AI-native feel: clean, prose-forward, content-density similar to Linear or Perplexity. Not Salesforce-style chrome.
• Brand identity TBD; working name MemberIntel only.

---

10. Compliance & Privacy

• GDPR + CCPA compliant from launch.
• DPA available for customers on request.
• Per-tenant data isolation in the warehouse. Per-customer brains are partitioned and never cross-accessed.
• ToS / Privacy Policy specific to MemberIntel:
  • Authorize per-customer brain accumulation (free + Pro).
  • Authorize anonymized cross-pollination to global brain (with opt-out).
  • Make explicit that data is not used for fine-tuning model weights in V1.
  • Make explicit that MP-only customers (non-MemberIntel users) are excluded from all MemberIntel data uses.
  • Reserve the right to change or discontinue the free tier with notice.
• Consent flow at signup: explicit checkbox covering the data-use boundaries. Free-tier users get the same consent flow as Pro users.
• Data deletion removes free-user data identically to Pro-user data — no “we keep your data because you didn’t pay us” pattern.
• Tier-change notifications — when a user downgrades from Pro to Free, clearly explain what changes (data sync slows, chat goes to Haiku + capped, etc.).
• Audit logging for every system action (data syncs, brain updates, cross-pollination promotions, tier changes).

---

11. Internationalization

• UI translated: English (default), Spanish, German.
• AI chat: any language input → response in same language. Knowledge corpora (global + customer brain) stay in their authored language; AI translates retrieved chunks as needed.
• Currency: per-customer (drawn from Stripe).
• Date / time: localized.

---

12. Technical Stack

• Hosting: Google Cloud Platform OR Heroku (Caseproof team to choose based on existing operational expertise).
• App framework: team’s choice; recommend a single-language stack (TS or Python) to match the team’s hiring profile.
• Database (per-customer warehouse): Postgres.
• Vector store: pgvector to start (no extra ops surface), or Pinecone if scale demands.
• LLM: Anthropic API directly. No LangChain or other orchestration framework in V1; orchestration logic owned in-house and version-controlled.
• Auth: OAuth-style flow with MP license; standard email + password fallback (with MFA option).
• Billing: Stripe (Caseproof’s existing relationship). Plan upgrades / downgrades / cancellations / failed payment handling — all self-serve in MemberIntel.
• Email: existing Caseproof transactional email infrastructure.
• Entitlement layer. Server-side service that owns the tier model: who can do what, how many tokens, what cadence of sync, what AI model routes where. Every feature gates through this layer. Don’t hardcode tier checks scattered throughout the codebase. Single source of truth.
• Quota tracking. Per-user, per-month counters: chat messages used, tokens used, advanced reports requested. Reset monthly. Visible to the user and to support staff.

---

13. Implementation Phases

Assumes 4 engineers + 1 designer + 1 PM, 6 months to launch, plus content lead hired in parallel by month 1.

Phase 0 — Pre-Build (Weeks 0–4, parallel)

• Hire content lead.
• Telemetry + Stripe data audit (per-customer scope only — confirm we can pull the customer’s own data via their own consent).
• Legal: ToS / Privacy Policy / DPA updates covering free tier + Pro tier + brain mechanics.
• Customer interviews: 10–15 with existing MP operators. Validate pricing tiers, top pain points, dashboard KPI selection, brain scope, and the free / Pro feature split.
• Define entitlement matrix in detail (every feature × every tier).
• Cost modeling: stress-test free-tier unit economics at projected adoption volumes (5K, 10K, 50K free users).
• Engineering ramp / architecture decisions.

Phase 1 — Operator Core (Months 1–3)

• Auth (MP-license-based + standalone account).
• Customer-data ingestion: MP sync (live for Pro, monthly snapshot for Free) + Stripe Connect query + Stripe OAuth for non-Connect customers.
• Per-customer warehouse + initial schema.
• Global brain seed: index existing MP docs + KB + blog. Author first 20 hand-written playbooks (content lead).
• Per-customer brain scaffolding: storage, retrieval, basic update mechanisms.
• AI tool surface (query_customer_metrics, search_global_brain, search_customer_brain, update_customer_brain, analyze_site).
• Public site analysis pipeline (Claude-based, weekly refresh, cached).
• Chat advisor with citations and feedback capture.
• Tier-gated AI model routing (Haiku for Free, Sonnet for Pro).
• Entitlement service.
• Quota tracking.
• Standard dashboard with tier-gated AI insight cards.
• Free vs. Pro UI differentiation (badges, upgrade nudges, usage indicators).
• Self-serve billing (upgrade, downgrade, cancel) in Settings.
• Privacy controls + GDPR / CCPA data export + deletion.
• i18n scaffolding (English + Spanish + German).
• Milestone (Month 3): internal beta with 10–20 hand-picked MP customers across both Free and Pro tiers.

Phase 2 — Polish + Launch (Months 4–6)

• Cross-pollination job (weekly schedule + content lead review queue).
• Weekly digest email generation (Haiku for Free, Sonnet for Pro).
• Feedback-loop refinement based on internal beta findings.
• Brain growth: content lead writes additional 30+ playbooks.
• AI eval suite + prompt regression tests.
• Cost monitoring + per-customer token caps.
• Free-tier signup flow polish (no friction beyond consent).
• Upgrade-funnel telemetry (where do free users hit upgrade prompts; what’s the conversion rate at each surface?).
• Cost-monitoring dashboards (free-tier cost-per-user, Pro-tier cost-per-user, cohort tracking).
• Pricing page + comparison table at memberintel.com.
• In-MemberPress-admin banner integration.
• Public launch ($29/mo Pro + always-free tier) to MP customer base.

V1.5 (post-V1)

• AI agent / write-actions (the “AI does it for me” capability — requires MP-side connector + write APIs + confirmation UX + audit + undo). Pro-only.
• First-time creator / greenfield onboarding flow (paired with agent mode for setup automation).
• Brain depth: target 100+ playbooks.
• Possible Pro+ tier introduction at $79–99/mo.
• Cross-pollination metrics + content lead tooling.

V2+

• Cross-platform support (Kajabi, Thinkific, Paid Memberships Pro).
• Tiered Pro plans (Annual, Pro+, Agency).
• SOC 2 Type 1.
• Cross-customer benchmarks (if legal Phase 0 work is unblocked).

---

14. Risks & Open Questions

Risks (ordered by severity)

1. Differentiation from generic AI. Without proprietary data, the moat is brain quality + per-customer memory. If the brain is shallow at launch, customers will compare unfavorably to ChatGPT and churn fast. Mitigation: hire content lead Day 1; require 50+ playbooks in global brain at launch; require eval suite proving MP-specific advantage over baseline LLM.
2. Free-tier cost spiral. If adoption exceeds projections and conversion lags, free-tier costs can balloon. At 50K free users × $1.10/mo = $55K/mo before any Pro revenue covers it. Mitigation: hard token caps, conservative growth ramp, weekly cost monitoring with circuit-breaker authority to reduce free entitlements if needed.
3. Free → Pro conversion underperforms. If real conversion lands at 1–2% instead of 5–10%, the model is broken. Mitigation: design for upgrade moments at multiple surfaces (chat cap, insight cliffhangers, advanced reports, brain editing); A/B test upgrade copy aggressively from launch; have a Plan B (raise the chat cap to push more users to upgrade, or tighten the cap to make Pro feel more necessary).
4. Cannibalization of paid intent. Some users who would have paid $29/mo from day one may settle for the free tier. Mitigation: free tier is intentionally limited in the things that drive habit (daily fresh insights, unlimited chat, advanced reports).
5. Hiring risk. 4 engineers + designer + PM + content lead within month 1 is a meaningful hiring sprint. Slipping by 2 months compresses the build window.
6. Site data sync reliability. Live MP / Stripe sync requires resilient pipelines (rate limits, schema changes, customer plugin updates breaking things). Realistic effort, not exotic but not trivial.
7. Hallucination on financial data. AI confidently misreporting churn / revenue is trust-killing. Tool-call grounding + citation requirements + eval suite are non-negotiable.
8. Brain quality dilution from free-tier signal. Free users provide thinner signal (5 chats/mo) than Pro users (unlimited). Cross-pollination weights need to favor high-signal sources to avoid the global brain drifting toward shallow patterns. Mitigation: explicit signal-quality scoring in cross-pollination; content lead reviews prioritize high-engagement source data.
9. Content lead becomes a bottleneck. One person reviewing all cross-pollination candidates + writing all playbooks won’t scale beyond ~1000 active customers. Plan for adding a second by V1.5.
10. Cross-pollination privacy slip. A bug in anonymization could leak customer-identifying data into the global brain. Hard rule: every promoted entry passes a manual privacy check.
11. Support burden from free users. Free users still file tickets. Without “priority support” for Pro, it’s hard to enforce tiered SLAs without seeming punitive. Mitigation: docs-first support model for free; AI-first support with human escalation.
12. Downgrade fraud. Users sign up for Pro, extract value, downgrade to Free, repeat. Mitigation: standard subscription patterns; track upgrade / downgrade patterns for anomalous signals.
13. Free-tier sunset risk. If the model proves unprofitable at scale, removing the free tier is a brand-damaging move. Mitigation: ToS explicitly reserves the right to change terms with notice; communicate the free tier as “for now” framing in marketing.
14. Brand dilution. Calling both tiers “MemberIntel” with a free / paid badge can anchor perceived value downward and make the Pro tier feel like an upcharge. Worth explicit positioning work before launch.

Open Questions

---

15. Glossary

• MemberIntel — working name for the product specified here. Not committed.
• MP — MemberPress.
• Operator — an existing MemberPress customer with an active site.
• Free / Connected Free — the always-free tier described in Section 5.
• Pro — the $29/mo paid tier described in Section 5.
• Global brain — shared knowledge corpus used as background context for all customers.
• Per-customer brain — knowledge corpus scoped to a single customer; used as personalized context for that customer only.
• Cross-pollination — the process by which anonymized learnings from per-customer brains are promoted into the global brain after content-lead review.
• Tool calls — structured function-style calls the LLM emits to query data or update the per-customer brain.
• RAG — retrieval-augmented generation; vector-search-based context retrieval used over the markdown brain corpora at query time.
• Entitlement layer — server-side service that gates every tier-restricted feature.