HitCreate — Whitepaper v3.2

AI-Native Marketing Infrastructure — the Platform That Closes the Loop A portfolio brand of FYI United Group Pty Ltd (ACN 665 393 546)

Document version: 3.2 (DRAFT — 2026-04-22) Status: Draft, pending founder sign-off on twenty open decisions (see §13) Supersedes: v3.1 amendment + v3.0 (both archived; changelog §14) Informed by: RESEARCH-279 (/root/.openclaw/workspace/docs/RESEARCH-279-loop-closure-thesis.md) and its seven underlying research tracks.

0. Purpose & what changed

v3.0 framed HitCreate as a headless agentic platform with three Surfaces (App / Agent / Ecosystem), Agency retainer as revenue anchor, and five defensible pillars (headless multi-surface, agentic orchestration core, primitive compounding, ecosystem-native, vertical context).

v3.1 concretised the App Surface V1 — seven primitives, eight-step intake, setup-fee concierge phase, two-archetype validation-gated rollout.

v3.2 is a structural reframe, driven by RESEARCH-279 findings:

1. Loop closure is promoted from "one of five pillars" to the singular moat claim. All other pillars are enablers, not standalone moats. This is a sharpening, not a contradiction — v3.0 already listed loop closure as pillar two; v3.2 centres on it.
2. Agency retainer is de-anchored. v3.0's 60% Agency target was an economic argument (most-predictable revenue). With loop closure as the moat, self-serve App + Ecosystem (MCP/SKILL distribution) + opt-in white-label reseller carry the revenue shape. Agency becomes opportunistic (NISARD-flagship pattern), not strategic. Target mix flips to 55% self-serve / 20% white-label / 15% ecosystem / 10% Agency.
3. Vertical scope narrows from aspirational breadth to two named verticals: SEQ waste operators (via Nicole) and AU fractional-professional services (via Nisbet). Everything else is templated horizontal capability added after both verticals validate.
4. NISARD is recast as a process-validation engagement, not a vertical-depth proof. Honest framing.
5. Phase B gate is time-boxed at 90 days from NISARD completion. Calendar-agnostic "we'll know when we know" framing is removed.
6. Seven canonical market shifts replace six (email deliverability crisis promoted; v3.1 numbers on ChatGPT Ads and Gen Z corrected). Three additional shifts flagged (browser agents, LinkedIn specifically, AU Privacy Act ADM deadlines).
7. Competitive threat model rewritten. Real threats are Klaviyo (5/5, loop-closed + public-company pressure to expand horizontally) and HubSpot Breeze (4/5). Lovable, GHL, Bolt, v0 are positioning-noise, not strategic threats. Frontier labs are going up-market (enterprise + GSI), not down into SMB — the compression risk is the skill-marketplace channel, not lab direct products.
8. Technical feasibility is affirmed at V1 scale (solo + contractor friends can ship the 7-primitive loop-closure platform in 6 months). "Compete with GHL head-on" explicitly requires $2-5M AUD + 15-25 FT + 18 months — not this thesis.

Sections of v3.0 / v3.1 not changed here remain load-bearing and unchanged.

0.5 Status (April 2026)

What exists: Min Min agent deployed on Hetzner CX42 Singapore, 17+ Docker containers (self-hosted Supabase, LiteLLM, Redis, n8n, Caddy, code-server). Four-tier autonomy ladder. Three-tier memory architecture. ORBIT AI effort-tracking wired to LiteLLM + Supabase. Domain architecture secured: hitcreate.io marketing, hitcreate.dev dev/tooling, hitcreate.app user content (per RESEARCH-275).

What is in flight: NISARD Pty Ltd — external prototype engagement for a fractional CFO practice. Recast as process-validation engagement (not first commercial customer, not first vertical proof). Validates the 8-step intake, DeploymentSpec output, 7-primitive generation chain, Bookii + Claude Design + Canva orchestration.

What is not yet built: The full App Surface. The Agent Surface. The ecosystem MCP servers. The agent library beyond Brand System Agent. The loop-closure infrastructure. The PLG acquisition funnel across any surface. White-label reseller tier.

What changed since v3.1 (2026-04-22): RESEARCH-279 synthesis of 7 parallel research tracks sharpened the thesis and competitive read. Key external events since v3.1: Klaviyo Composer launch (March 2026), HubSpot Spring 2026 Spotlight "Loop Marketing" framework + outcome-based agent pricing, Shopify Agentic Storefronts defaulted-on across 5.6M merchants (24 Mar 2026), ChatGPT Ad CPMs compressed from $60 → $25-35 with minimum cut to $50K, Canva AI 2.0 scaling. The compression window narrowed from v3.0's "12-24 months" to an evidence-weighted ~18 months to Q4 2027 for prosumer end-to-end compression, with ~24-30 months for vertical + loop-closure defensibility.

1. Thesis

HitCreate is the AI-native marketing infrastructure platform that closes the loop. Every other AI marketing tool generates and walks away. HitCreate generates, deploys, measures, and compounds — within a specific customer archetype, across customers, with anonymised pattern extraction, validated by a falsifiable compounding metric.

1.1 The load-bearing claim

In an AI-era marketing stack, generation is commodity. Claude, GPT-5, Gemini, Llama all produce comparable copy, pages, sequences, visuals. What separates durable products from wrappers is whether deployed-asset performance flows back into the generator to make the next iteration measurably better. Without that loop, every tool is a prompt template over a frontier model.

Loop closure is operationally definable (§3). It is architecturally testable (§3.3). It is falsifiable at scale thresholds (§3.6). It compounds superlinearly with deployment volume within an archetype. It cannot be replicated by a frontier-lab orchestrator alone because the lab does not own the deployments.

This is the singular moat claim. Everything else in this whitepaper is how we build, market, and defend it.

1.2 The founder's earlier framing, sharpened

The founder's working thesis — "LLMs need loop closure, everyone else is a wrapper in a vertical" — is directionally correct but half-right on "wrapper." RESEARCH-279 Track 4 found ~60% of named competitors have real moats (Lovable's distribution+capital, Bolt's WebContainers, Canva's 265M MAU, Replit's infra depth). The sharpened form:

"Most AI marketing tools generate and walk away. HitCreate closes the loop. The longer you use it, the better it gets at your business."

This is forward-pointing, sellable, and testable. "They're wrappers" is a private mental model; it is not the public pitch.

1.3 The strategic bet — updated

HitCreate wagers five things:

1. Generation is commodity and will remain commodity. Closed-loop generation is not commodity because it requires deployments the frontier labs do not own.
2. Attribution at single-small-business scale is structurally broken. Cookies unreliable, MTA dead, iOS LTP stripping UTMs. The only way a loop can be closed honestly at SMB scale is through cross-customer pooling within a specific archetype.
3. Vertical context is a ~20% moat, not an 80% moat. The durable 20%: warm relationships inside the vertical + deployed-customer benchmark data. HitCreate invests in two verticals deeply (SEQ waste + AU fractional-professional) rather than spreading across five.
4. Frontier labs go up, not down. Anthropic, OpenAI, Google, Meta are building for enterprise + GSI partnerships. The SMB vertical layer is being left open deliberately. The compression risk is the skill-marketplace channel (when a "Skill: SMB marketing funnel" + MCP Apps reach 70% of HitCreate's value), not direct lab products.
5. A solo founder + rotating technical friends can ship V1. RESEARCH-279 Track 7 validates this at 6-month horizon for 7 primitives + per-customer loop closure + one archetype. It does NOT support "compete with GHL head-on" — that is a $2-5M / 15-25 FT / 18-month bet.

1.4 The structural advantage

HitCreate's advantage over incumbents (GoHighLevel, HubSpot, Jasper, Klaviyo) is zero legacy architecture. Event-stream first data model from day one. Ecosystem-standard artefacts from day one. No proprietary format to unwind, no pre-agentic UI debt, no AI feature bolted onto a legacy core.

This advantage is real but not permanent. The ~18-month window is the time to establish category position before either (a) incumbents ship coherent agentic rebuilds (Klaviyo already has — Composer, March 2026), or (b) frontier-lab skill marketplaces reach "good enough" for SMB horizontal use. Watch §11 for monitoring signals.

2. What HitCreate is, and isn't

2.1 Is

HitCreate is AI-native marketing infrastructure that sits on top of whatever the customer's business actually is. It generates, deploys, measures, and iterates across seven primitives (pages, forms, scorecards, email sequences, SMS sequences, unified inbox, CRM + pipeline). It closes the loop between deployed-asset performance and future generation, within an archetype, across anonymised customer patterns.

2.2 Isn't

HitCreate does not build:

• Full apps (Lovable, Bolt, v0, Replit, Base44 own that category)
• Dev tools (Cursor, Windsurf own that category)
• Operations platforms / industry ERPs (ServiceM8, Simpro, HubSpot CRM at enterprise scale own that)
• Video generation (Sora, Veo, Runway)
• Ad buying / ad account management (Meta Ads Manager, Google Ads, ChatGPT Ads — HitCreate stays upstream of the buy)
• Social media scheduling (Buffer, Later, Hootsuite — low-margin, crowded)
• Scale-up CRM at 100K+ contacts (HubSpot owns; HitCreate stays at pipeline-view-for-builders)
• Influencer marketplaces (Aspire, GRIN, Creator.co)
• Agency services business as anchor revenue (see §8)

2.3 The integration posture

HitCreate sits on top of the customer's business:

• Ecom on Shopify? HitCreate generates the cart-recovery + quiz + zero-party scorecard.
• SaaS on Next.js / their-own-Lovable-app? HitCreate generates waitlist → trial → activation.
• Service business on ServiceM8 / Simpro? HitCreate generates lead-gen + booking + review.

HitCreate calls Claude Design for polished visuals, Canva for templated visuals, Resend for email, Twilio for SMS, Cloudflare Pages for deploy, Stripe for billing, Bookii for scheduling. HitCreate does not rebuild any of these — it orchestrates, instruments, and closes the loop on them.

3. The loop, operationally

3.1 Definition — the four criteria

A deployment pipeline is loop-closed if and only if all four are true:

1. Deployed asset emits a performance signal. Event captured, attributed to the specific asset version.
2. Signal is normalised across runs into a metric comparable across customers/deployments (conversion rate, click rate — not raw count).
3. Signal updates generator behaviour for the next generation via context injection, performance-weighted retrieval, fine-tuning, or reward modelling.
4. Next generation is measurably better on the same metric, compounding across iterations (not a one-shot improvement).

Missing any of the four and it's not loop closure — it's a component of loop closure. Analytics dashboards miss #3. Standard A/B tools miss #3. Personalisation engines miss #4 unless the selection signal updates the generator. RAG without performance weighting misses #2/#3. Only a system with all four closes the loop.

3.2 Architecture by phase

HitCreate will not claim V3 architecture while shipping V1 implementation. Honest scale-of-claim is the credibility play: "We close the loop within-customer on email/SMS subject lines today. We are building toward cross-customer compounding; measurable improvement at customer #20 per archetype." That is defensible in a technical diligence conversation. "Our AI learns from your data" is not.

3.3 Per-primitive signal discipline

Email subject-line bandit is the one primitive where V1 loop closure is legitimately high-quality from day 1. Ship this first. Prove it inside one customer in one campaign. Every prospect pitch demonstrates real compounding. Beats hand-waving.

3.4 Attribution reality — the hardest honesty cost

Per RESEARCH-279 Track 1 §4:

• Third-party cookies unreliable (Google reversed deprecation April 2025; but Safari+Firefox ~50% of web cookieless, consent rates collapsing).
• Multi-touch attribution broken (channel tools all overcount).
• iOS 17 Link Tracking Protection strips UTM parameters.
• Multi-touch attribution (MTA) structurally unreliable.

What survives at 2026:

• First-party + server-side tracking (Meta CAPI, Google Enhanced Conversions) — partial; fails on cold traffic.
• Geo-lift incrementality — gold standard; needs national ad spend (fails at SMB scale).
• MMM — needs ≥2 years of clean channel data (fails at SMB scale).
• AI-citation share-of-voice — measurable but upstream; doesn't drive signup-level signal directly.

At single-small-business scale (300-1,000 pageviews/week, 2-5% conversion):

• Per-customer attribution is directionally informative but not clean enough for automated feedback loops.
• A 5% → 7% per-iteration improvement on 100 visits is almost always statistical artefact.
• Cross-customer aggregation within archetype is not optional. It is the only path to statistical power.

HitCreate's posture: publicly admit the attribution blind spots. Competitors either don't close the loop (Jasper, Copy.ai) or claim more than they can prove (most "AI marketing platforms"). HitCreate claiming less than competitors while proving more is the credibility differentiator.

3.5 Cross-customer pooling — privacy posture

Four patterns evaluated in RESEARCH-279 Track 1 §5.3:

• Pattern A — raw pooling under T&C cover (Klaviyo, Attentive, Shopify Magic do this). Rejected at HitCreate's current scale; legal risk not tolerable until we're big.
• Pattern B — anonymised pattern extraction. Brand names, PII stripped; only structural/stylistic patterns pooled. Recommended default.
• Pattern C — federated learning / differential privacy. V2+ (too much engineering overhead for V1).
• Pattern D — opt-in aggregate benchmarks. Customers opt-in; their anonymised metrics contribute to vertical-aggregate benchmarks. Recommended enhancement.

V1 posture: Pattern B default + Pattern D opt-in. This is the privacy moat HitCreate gets that Klaviyo/Attentive/Shopify Magic can't articulate cleanly — they are Pattern A under T&C cover. HitCreate's "anonymised + opt-in benchmarks" is cleaner to explain to a sceptical scorecard-completing prospect.

3.6 Falsification bar

The primary test: for a given primitive × vertical archetype, plot rolling-median conversion across rolling 10-deployment windows. Loop closure is confirmed if the slope is positive at p<0.10 over N≥50 deployments and a feedback-off control arm is flat or negative on the same cohort (synthetic-control via LLM-as-judge is acceptable where RCT is ethically unavailable).

At realistic V1 volumes (3-10 concierge customers by month 6):

• Email subject MAB convergence within customer: feasible today. 1 customer, 1 campaign.
• Per-customer loop closure on pages: borderline. Needs 3+ months + ≥500 visits/week.
• Cross-customer within-vertical loop closure on pages: not feasible until V2. Needs ≥20 customers × 6 months per archetype.
• Platform-wide cross-vertical: V3+ only. Needs 100+ customers, 12+ months.

HitCreate commits to publishing the falsification methodology. If the metric doesn't compound by the threshold promised, we say so publicly. That's how you beat incumbents who claim more than they can prove.

3.7 The one-paragraph pitch

"HitCreate closes the loop on the primitives where closing the loop is measurable within a single customer's normal usage: email subject lines via multi-armed bandits, SMS openers the same, copy patterns via prompt-context injection of past winners. For page and scorecard primitives where single-customer volume is insufficient, we pool anonymised patterns across customers within the same vertical archetype — brand names and PII stripped, only structural patterns. Customers opt-in to the benchmark pool; they get immediate benefit from the archetype's prior winners and contribute to the library as they deploy. The moat compounds: every new customer improves the archetype's pattern library; every returning customer benefits from the updated library. At our current pace (3-10 concierge customers per archetype in 6 months), the fractional-CFO archetype will have a statistically meaningful pattern library by month 9-12."

Truthful, technically defensible, falsifiable. This paragraph beats any "our AI learns from your data" pitch in a technical diligence conversation.

4. Competitive positioning

4.1 The threat matrix (from RESEARCH-279 Track 3 + 4)

4.2 Positioning lines — the five that cut through

• vs Lovable: "Lovable builds the app. HitCreate builds the marketing system on top of whatever your app is."
• vs GoHighLevel: "GHL is a pre-agentic platform with AI bolted on. HitCreate is agent-native from day one — primitives generated, deployment automatic, and the loop closes."
• vs Canva AI 2.0: "Canva generates content. HitCreate runs your marketing system — pages, forms, sequences, scorecards, attribution. When HitCreate needs a visual, it calls Canva."
• vs Jasper / Copy.ai: "They write copy. We deploy, measure, and iterate."
• vs ChatGPT / Claude direct: "ChatGPT is a blank page. HitCreate is the running business — already knows your brand, customer, offer, and conversion data."

4.3 The compression window — 18 months median

Evidence-weighted probability distribution (RESEARCH-279 Track 2):

• 10% — compression starts in 9-12 months
• 30% — compression starts in 12-18 months
• 35% (base case) — compression starts in 18-24 months
• 20% — 24-36 months
• 5% — 36+ months

Median: ~18 months to Q4 2027 for prosumer "tell Claude to build my funnel" end-to-end capability. ~24-30 months for the vertical + loop-closure defensibility that HitCreate is building.

4.4 Weekly signal monitoring

Three signals to watch every week:

1. Frontier-lab "deploy-to-domain" primitive. Any lab shipping native landing-page deploy to a customer's domain in a consumer tier. This is the missing link that makes SMB end-to-end work today.
2. Vertical SMB skill / pack from a frontier lab. E.g. "Skill: SMB marketing funnel" in anthropics/skills, OpenAI App Directory, or Gemini release notes. Or explicit "Claude for Small Business" / "Gemini for Small Business" tier at sub-$100/mo.
3. Incumbent SMB-agent price collapse. HubSpot Breeze, GoHighLevel, Sierra, or Canva AI 2.0 shipping agent-first tiers under $100/mo aimed at solo operators.

Min Min will automate this monitoring (assigned: open decision #20).

4.5 The Klaviyo-specific risk

Klaviyo is HitCreate's #1 strategic threat. One narrative pivot (beyond ecom into services/B2C) would compress HitCreate's TAM materially. Klaviyo has:

• The data (0.5T interactions, 193K brands).
• The loop-closed product (Composer, March 2026).
• The ecosystem posture (MCP server + ChatGPT app).
• The public-market pressure (stock down 40% from IPO, growth-narrative need).

HitCreate's temporary shelter is non-ecom vertical focus. The 18-month window is the time to build cross-customer benchmark data in waste + fractional-CFO before Klaviyo pivots horizontal. If Klaviyo announces "Klaviyo for services" or "Klaviyo for B2C subscription," HitCreate's scope and sequencing change immediately (open decision #20 owner will flag).

5. Market context — the seven canonical shifts (updated)

Updated from v3.1 §8 with RESEARCH-279 Track 5 data.

5.1 Shift 1 — Search → Answer economy

AI referral traffic growing 130-150% YoY at Q1 2026 (not the v3.1 +805% Black Friday spike). On track for 20-28% of total web traffic by end of 2026. AI Overviews trigger on ~25% of Google queries. GEO is now a commercial category with Semrush AI Visibility Toolkit, Ahrefs Brand Radar, Otterly, Scrunch, Stackmatix all shipped.

Correction to v3.1: Gen Z TikTok-over-Google preference fell from 8% to 4% in 2026 (Adobe data). "50%+ of Gen Z skip Google" is wrong. The shift is query redistribution, not abandonment.

Implication: GEO-native output is the right call for every page. The OSS GEO audit tool (v3.1 §10 Drop 1) now competes with commercial incumbents — reposition as lead-magnet for the generator, not a standalone product.

5.2 Shift 2 — ChatGPT Ads live (9 Feb 2026)

CPM compressed from $60 → $25-35 in 9 weeks. Minimum spend cut from $250K → $50K (now SMB-reachable). Criteo is the ad-tech pipe (addressable base of ~17K retailers, not direct advertisers). Conversion 1.5x other channels. Retail + grocery = 44% of spend.

Perplexity killed ads (Feb 2026). Anthropic committed Claude stays ad-free ("no ads, ever" — Super Bowl spot). AI ad monetisation is OpenAI + Google, not a four-horse race.

Implication: v3.1 §11's "wait for self-serve" window is zero. High-intent LLM traffic + HitCreate conversion infra = defensible wedge. HitCreate stays upstream of the buy but generates the landing pages + scorecards ChatGPT-ad traffic lands on.

5.3 Shift 3 — Organic social collapse (including LinkedIn specifically)

Instagram organic reach 3.5% of followers (down 30-40% YoY). LinkedIn personal-profile reach -50% YoY; company pages -60 to -66%; external links -60%. Document posts (PDF carousels) outperform video 278%. TikTok USDS deal closed Jan 2026 but regulatory uncertainty parks enterprise spend.

Implication: founder-consultancy archetype (Phase B target) cannot rely on LinkedIn company pages. Scorecard-result share image (internal PDF carousel) is a V1 feature — LinkedIn-native distribution. Email sequences more valuable than ever given organic social collapse.

5.4 Shift 4 — Agentic commerce, default-on

Shopify Agentic Storefronts defaulted-on 24 Mar 2026 across 5.6M merchants (4% OpenAI fee + 2.9% Shopify). OpenAI ACP + Google UCP as open standards. Amazon Rufus: 250M users, $10B sales lift, 60% conversion uplift per engaged session. McKinsey projects $3-5T global agentic commerce by 2030.

Implication: HitCreate explicitly excludes ecom (v3.1 §2). This shift validates that exclusion. Non-ecom agentic commerce (services, B2B) is the genuine opportunity — no "ACP for services" standard yet. HitCreate's scorecard + structured schema could position as the service-business agent-readable layer before an incumbent standardises. V1.5+ opportunity.

5.5 Shift 5 — Zero-party > first-party > third-party (reframed)

Correction: cookies didn't die (Google reversed April 2025). But Safari+Firefox ~50% of web cookieless, consent rates collapsing, AU Privacy Act 2024 automated-decision-making disclosures mandatory 10 Dec 2026, EU AI Act enforcement 2 Aug 2026.

Implication: scorecard-as-zero-party-engine is structurally correct. Scorecards that tier/price/route leads are "automated decision making" under AU Act — HitCreate ships ADM-disclosure copy + audit logs by default. Competitors (ScoreApp, Interact, Outgrow, Typeform) don't. Feature, not burden.

5.6 Shift 6 — Measurement permanently broken

46.9% of US marketers investing more in MMM. AI-citation SoV tooling commercialised (Semrush AI Visibility Toolkit, Ahrefs Brand Radar — both live Q4 2025 / Q1 2026). "AI SoV" is the new rankings report category.

Implication: HitCreate builds AI-citation SoV visibility into the GEO-native primitive (not as a standalone product). Tracks which queries HitCreate-generated pages are cited on.

5.7 Shift 7 — Email deliverability crisis (new)

Gmail/Yahoo strict DMARC enforcement live Nov 2025. 30% of bulk senders still non-compliant. 22-34% spam placement for non-compliant senders. Every HitCreate email sequence hits this.

Implication: DMARC setup, domain warm-up, suppression lists, Gmail/Yahoo postmaster monitoring are V1 infrastructure — not V2 features. Without this, HitCreate sequences go to spam and the loop breaks on the first step.

5.8 Additional material shifts (watch, don't promote to canonical)

• Browser agents (Operator, Mariner, Dia). 72% enterprise AI-agent adoption (Gartner). HitCreate pages must be agent-readable — forms parseable, CTAs structured.
• AU Privacy Act ADM + EU AI Act marketing deadlines. Covered in Shift 5.
• Voice-first interfaces. Premature; watch but don't build.
• Creator newsletter economy (Substack, Beehiiv, Kit). Adjacent to HitCreate's email primitive; partnership opportunity, not competition.

6. Vertical strategy

6.1 The honest decomposition (from RESEARCH-279 Track 6)

"Vertical priors" decomposes into five components with wildly different compounding properties:

~75% of any vertical prior is replicable. ~15% requires customer-specific data. ~10% is structurally unreplicable — specifically, warm relationships and deployed-customer benchmark data. The moat is concentrated in that 10%, not the feature set.

6.2 The two verticals — locked

Everything else is templated horizontal capability. Not dismissed, but deferred until Phase D (after two archetypes validate independently).

Verticals explicitly NOT invested in during V1:

• Workplace safety (Safetii — separate brand, different product)
• Generic field service (ServiceTitan's surface — no warm-relationship asset)
• SMB founders / SaaS (7/25 moat score — no unique angle)
• AU-generic SMB (geography is not a vertical)

6.3 NISARD — recast

NISARD is a process-validation engagement, not a vertical-depth proof. It validates:

• The 8-step intake (v3.1 §4) produces a parseable DeploymentSpec.
• The 7-primitive generation chain works end-to-end.
• The Brand System Agent + Claude Design + Canva + Bookii orchestration holds together.
• The loop-closure infrastructure (event capture, feedback store, context injection) works in production.

NISARD does not prove the "fractional-CFO vertical prior" claim. That only compounds after 20+ deployments with loop-closed performance data.

NISARD is currently a co-op / services-exchange engagement, not a paying commercial customer (RESEARCH-268 §9.1). It becomes Bucket A revenue only if/when it converts to paid. Until then, NISARD is Bucket B (product build via real use case).

6.4 Third archetype — deferred

Third archetype selection waits for Phase C completion. Candidate verticals depend on warmer-pipeline signals — likely founder-SaaS (via Indie Hackers / X) or Australian B2B services. Decision is gated on Phase B + C outcomes, not speculated now.

7. Architecture

7.1 The Platform + three Surfaces framing is retained

v3.0 §3's framing — one headless Platform, three client Surfaces (App / Agent / Ecosystem) — remains correct. The change in v3.2 is surface priority.

v3.0 priority: App → Agent → Ecosystem. v3.2 priority: Ecosystem seeding begins Phase A (OSS drops); App Surface ships Phase B; Agent Surface ships Phase D. The reason: ecosystem distribution compounds, App polish depreciates against frontier-lab generation, and Agent Surface requires sufficient App-Surface state to handle-off cleanly.

7.2 The skinny stack — locked

Per RESEARCH-279 Track 7. Build almost nothing; orchestrate everything.

Non-negotiables from CLAUDE.md:

• Rule 6 (all LLM through LiteLLM)
• Postgres only (code standards)
• Docker 127.0.0.1 port binding (Rule 1)
• NODE_OPTIONS heap cap on builds (Rule 12)
• Edit on Aorus + GitHub fence + VPS pulls (DEV-FLOW)
• Sub-agent sanity check + scoped allowlist (Rule 11)

7.3 The loop-closure data model

The spine. If this schema is right in month 1, contractors can ship surfaces. If it's wrong, no amount of surface polish saves it.

customers
  id, archetype, consent_pool_b, opt_in_pool_d

deployments
  id, customer_id, primitive_type, archetype, generation_context,
  deployed_content_hash, deployed_url, deployed_at, model_version

events
  id, deployment_id, event_type, timestamp, metadata

variants
  id, deployment_id, variant_hash, allocation_pct, alpha, beta  -- for MAB

outcomes
  id, deployment_id, variant_id, metric_name, metric_value,
  window_start, window_end, sample_size

feature_signals
  id, archetype, primitive_type, signal_key, signal_value,
  updated_at, source_deployment_count

llm_calls
  id, customer_id, deployment_id, prompt_hash, model_version,
  input_tokens, output_tokens, latency_ms, cost_usd

The rest of the schema (pages, forms, scorecards, sequences, contacts, pipeline stages) is primitive-specific and grows around the spine.

7.4 The artefact contract (kept from v3.0 §3.1)

Every primitive emits ecosystem-standard portable artefacts:

Ecosystem-native artefacts are the distribution play (§9) and the hedge against frontier-lab compression (§4.3).

8. Business model

8.1 The revenue-mix flip

v3.0 target (24-month steady state): 60% Agency / 25% self-serve / 15% Ecosystem. v3.2 target: 55% Self-serve App+Agent / 20% White-label reseller / 15% Ecosystem / 10% Agency (flagship only).

Rationale: with loop closure as the singular moat, the product is the business. Agency retainer is labour-capped, not compounding; self-serve + reseller + ecosystem are all compounding. Agency retained at 10% for flagship / process-validation engagements (NISARD-pattern) — not as revenue anchor.

This directly answers the founder's 2026-04-22 pushback: "I don't want Agency as anchor. I want the product to be the business." v3.2 reflects that.

8.2 Tier structure

Base agents included all paid tiers: Brand System Agent (foundational), Min Min orchestration (App + Agent surfaces), Landing Page, Email Sequence, Social Post, SEO/GEO Content, basic Contact.

Add-on agents ($29-99/mo at Starter/Pro; baked into Business+): Video, Ad Creative, SMS, WhatsApp, Outreach, Chatbot, Review/Reputation, Scorecard.

Credits meter: LLM tokens, image generation (Claude Design API), video generation, render minutes, voice cloning, SMS sends, email overage, data scraping, platform posting over threshold. External API costs pass through transparently.

8.3 Concierge setup fees — phase-out plan

During the concierge phase (Phase A + early Phase B), setup fees compensate Mister's time:

Phase-out trigger: 3 consecutive deploys complete the Phase-8 intake-to-live pipeline in <60 minutes with no human touch. When that holds, setup fees drop to $0 (or become optional "white-glove onboarding" at the customer's election). Open decision #7.

8.4 Why self-serve + reseller + ecosystem is a defensible anchor

• Self-serve scales with marketing spend, not headcount. Loop-closure + vertical wedge + OSS distribution drives CAC down over time.
• White-label reseller is agency-channel play without running an agency. Agencies take HitCreate, rebrand, resell. HitCreate captures per-sub-account revenue + loop-closure data from reseller clients (with consent).
• Ecosystem (Surface C) is low-marginal-cost distribution into frontier agents. Every MCP call + skill-triggered use is revenue + distribution + signal.

Agency retainer — the thing v3.0 leaned on as "defensible" — is labour-capped. It doesn't compound. It doesn't scale past headcount. Retained for flagship pattern validation only.

8.5 Failure modes to guard against

• SKU bloat — three-decision pricing (tier, add-ons, credits), never a matrix.
• Credit anxiety — upfront cost disclosure on expensive operations.
• Capability gap mid-task — Min Min surfaces gaps at planning time, not execution time.
• Tier cliffs — add-ons and top-ups let users grow within tier.
• Reseller cannibalising Agency flagship — clear scope: reseller is platform-access + branding-rights; flagship is delivery-pod.
• Setup fees never phasing out — 3-consecutive-deploys trigger must be enforced.

9. Go-to-market

9.1 Three funnels feed the product, not an anchor

v3.0 framed three funnels feeding an Agency anchor. v3.2 frames three funnels feeding product revenue directly, with Agency as flagship pattern-generator, not anchor.

• Funnel A — Self-serve App. Content + SEO/GEO + paid ads (when Phase B validates). OSS GEO audit tool as lead magnet (Phase A seeding).
• Funnel B — Agent Surface. SMS / WhatsApp / email inbound. Opens Phase D+ when App Surface is mature enough to hand off cleanly.
• Funnel C — Ecosystem. Community repo contributions (awesome-claude-design, awesome-agent-skills, awesome-claude-code-subagents). MCP server listings. SKILL.md bundles. Free tier as top-of-funnel.

9.2 Phase-gated rollout (90-day gate on Phase B)

9.3 OSS distribution drops — Phase A seeding (concrete)

Per v3.1 §10, updated for v3.2:

1. GEO audit tool (hitcreate.dev). "See what ChatGPT sees when it looks at your site, and what HitCreate would generate differently." Free, email-gated, open-sourced. Repositioned as lead-magnet for the generator, not as category-defining product — incumbents (Semrush, Ahrefs, Otterly) own the "audit" category commercially. HitCreate's angle: generation-quality preview.

2. Brand System Agent SKILL.md bundle. Published to awesome-claude-design / awesome-agent-skills / anthropics/skills. Installable into Claude Code, Cursor, Claude Desktop, ChatGPT Apps, Gemini CLI. Costs near-zero to produce after NISARD ships.

3. Bookii MCP spec. Published to awesome-agent-skills once Bookii MVP ships. Establishes HitCreate as ecosystem-native before paid ads run.

Each drop precedes paid ads. They establish ecosystem-native credibility before spend starts.

9.4 Paid ads budget

Initial test budget $500-1,500 AUD/week across creative variants, capped at quarterly FYI United discretionary spend. Funded from Mister's Dreamworld income. Kills fast if CAC is clearly off. No hope-propping. First ads deferred until Phase B gate passes (no point driving traffic to an unvalidated archetype).

9.5 Dogfooding advantage

HitCreate's Ad Creative Agent generates HitCreate's own launch campaign. Internal use first (prove it on HitCreate campaigns), customer add-on second. Credibility + structural advantage — HitCreate can test more creative variants faster and cheaper than competitors using agencies.

9.6 Brand boundary — HitCreate vs FYI Digital vs M-Webb

Retained from v3.1 §6, with v3.2 update:

Rule: if the engagement fits the 7 primitives and the HitCreate constraint (marketing infra only), it runs through HitCreate. If it needs custom apps, ops platforms, or bespoke scope, it routes to FYI Digital or M-Webb.

10. Technical feasibility & team shape

10.1 The four disciplines that make V1 achievable

Per RESEARCH-279 Track 7:

1. Buy at the edge, build the spine. PostHog, Langfuse, Resend, Twilio, Stripe, Cloudflare Pages, Supabase. Build the generation → deploy → signal capture → feedback → regeneration loop.
2. Loop-closure data model in month 1. The Deployment → Event → Variant → Outcome → FeatureSignal schema. If right, contractors ship surfaces cleanly. If wrong, no polish saves it.
3. One spine owner. Founder (or single FT engineer) owns the generation → deployment → feedback loop end-to-end. Contractors ship surfaces, not loops.
4. Contractor model breaks at 300-500 customers. First full-time hire needed at Phase B → C transition (~month 6-9 post-kickoff).

10.2 Team shape timeline

10.3 Cost model at four scale points

10.4 The "compete with GHL head-on" feasibility check

GoHighLevel scale: ~$200M+ ARR, 70K agency accounts, 600+ headcount. Lovable scale: $400M ARR, $530M raised, 100+ headcount.

Competing head-on at that scale requires $2-5M AUD funding + 15-25 FT inside 18 months + abandonment of the headless-Platform framing (both GHL and Lovable are full-stack vertically integrated).

Current capacity — 20-30 hrs/week + rotating contractor friends — cannot fund that fight. Dreamworld income + agency revenue is sufficient to run the product bet at V1 scale. Not sufficient to pick a war with 600-person competitors.

The honest posture: HitCreate doesn't need to compete with GHL head-on to succeed. GHL's moat is agency-channel distribution. HitCreate's moat is loop closure within specific archetypes. Different products, different buyers (agencies reselling vs builders direct). Take the window, build the moat, raise capital only when thesis validates. Open decision #16.

11. Risks — updated

11.1 Primary risk — Klaviyo horizontal expansion

If Klaviyo announces "Klaviyo for services" / "Klaviyo for B2C subscription" or ships a native landing-page + scorecard primitive in the next 12 months, HitCreate's TAM compresses materially. Klaviyo already has the data scale, the shipped loop-closure product (Composer), the ecosystem-native posture (MCP + ChatGPT app), and the public-market pressure to expand.

Mitigation: Own the non-ecom vertical context deeply in the 18-month window. Build cross-customer benchmark data in waste + fractional-CFO before Klaviyo pivots. Monitor Klaviyo quarterly earnings, roadmap signals, Composer GA announcements, blog posts mentioning non-ecom verticals.

11.2 Secondary — HubSpot Breeze autonomous generation agent

If HubSpot ships a Breeze agent that autonomously generates + deploys + measures + regenerates landing pages or email sequences, HitCreate's SMB segment squeezes. Spring 2026 Spotlight signalled this direction ("Loop Marketing" framework).

Mitigation: HitCreate's "owned code, own your data, deploy to your domain" posture differentiates against HubSpot's walled garden for the builder archetype. Architectural debt slows HubSpot vs Klaviyo. Monitor HubSpot Spotlights (Spring/Fall).

11.3 Frontier-lab vertical extension — softer than v3.0 implied

Frontier labs are going up (enterprise + GSI), not down (SMB direct). Lower direct risk than v3.0 framed.

Actual risk path: skill-marketplace channel. A "Skill: SMB marketing funnel" bundle in Anthropic Skills + MCP Apps + Claude Design could reach 70% of HitCreate's App Surface value at user-installable cost.

Mitigation: HitCreate ships into the skill marketplaces itself (hitcreate-funnel-builder SKILL.md, Bookii MCP, GEO audit MCP). Being in the winner's marketplace beats trying to keep users out of it.

11.4 Canva AI 2.0 agentic SMB pivot

Canva has 265M MAU and $4B ARR. If they ship "agentic marketing infra for SMBs," HitCreate has no distribution answer.

Mitigation: Loop closure is not Canva's culture (they are design-first). The Canva AI 2.0 product surface remains content-generation-centric through 2026 public roadmap. Watch for acquisition announcements; Canva has M&A firepower and could buy a loop-closure player.

11.5 Email deliverability

Gmail/Yahoo DMARC enforcement + ~34% spam placement for non-compliant senders means HitCreate's email sequences will go to spam without proper infrastructure.

Mitigation: DMARC + SPF + DKIM setup is V1 infrastructure, not V2 feature. Per-customer domain warm-up, suppression list discipline, Google Postmaster + Yahoo Sender Hub monitoring. Resend handles most of this but requires correct setup.

11.6 AU Privacy Act 2024 ADM compliance

Scorecards that tier/price/route leads are ADM. Disclosure mandatory by 10 Dec 2026.

Mitigation: ADM disclosure copy + audit logs shipped by default. Feature differentiator, not compliance burden. See §5.5.

11.7 Scope / bandwidth

20-30 hrs/week + contractors is tight for 7 primitives + loop-closure infra + two vertical beachheads + ecosystem drops + NISARD delivery.

Mitigation: Phase-gated rollout per §9.2. Contractor-absorbable booklets only in Phase A + B. First FT hire explicitly time-boxed at Phase B → C transition. Declining good-but-not-top opportunities is the operating discipline.

11.8 Agency unit economics on flagship engagements

10% of revenue mix at flagship retainer pricing requires margin to work. First 3-5 flagship engagements determine whether this is viable.

Mitigation: Treat flagships as process-validation + reference-customer generation. Accept lower margin in exchange for marketing asset. Gate new flagships on prior delivery SOPs being documented.

11.9 Credits margin compression

Frontier model pricing may not continue falling linearly. Per-credit cost watch.

Mitigation: Monitor per-customer LLM cost; pass through expensive external API costs transparently.

11.10 Waste beachhead dependency

Nicole's Kartaway restraint clause must be reviewed before Local Trader waste outreach. SEQ waste is the beachhead, not the whole model.

Mitigation: Legal review of restraint clause before Phase C Nicole engagement. Other waste geographies exist as fallbacks.

11.11 Single-operator key-person risk

FYI United, HitCreate, Platform architecture all depend on Mister.

Mitigation: Documentation discipline already in place (CLAUDE.md, MEMORY.md, DEV-FLOW, RESEARCH-* index). Second operator with Agency-flagship tier open. External advisors engaged (Matthew Nisbet, Matt Webb). First FT engineering hire at Phase B → C.

12. Ecosystem posture (retained from v3.0 §9, sharpened)

12.1 The standards HitCreate emits

• SKILL.md (Anthropic Agent Skills) — adopted by VoltAgent, Google Labs, MiniMax, DuckDB, OpenAI Codex CLI.
• DESIGN.md — 9-section brand design system, 68+ published brand systems, Claude Design consumable.
• MCP — Linux Foundation (donated by Anthropic Dec 2025). De facto across Canva, Asana, Google Workspace, Slack, GitHub, Klaviyo, HubSpot.
• Subagent specs — Claude Code / Cursor / Gemini CLI / OpenCode compatible.
• llms.txt, tools indexes, JSON-LD — LLM-discoverable content and tool standards.
• ACP-compatible product feeds + UCP-compatible (for V1.5 service-business use case, per §5.4).

12.2 Posture

Ecosystem-native. Every primitive emits a standard artefact. Every agent is installable as a subagent or skill. Every external tool accessed via its standard interface (API, MCP), not custom integrations.

Consequences:

• Free distribution. Every HitCreate artefact in a community repo is a marketing surface.
• Low switching cost in. User's existing SKILL.md bundle + MCP integrations work inside HitCreate immediately.
• Low switching cost out. Users can export and leave. Strategically correct — retention on real value compounds.

12.3 Relationship to frontier labs

Anthropic, OpenAI, Google, xAI, Meta are the orchestration layer HitCreate plugs into via Surface C. Customers, not competitors. HitCreate publishes SKILL.md, subagent, and MCP artefacts consumable in each.

HitCreate does not try to build a better orchestration agent than Claude or ChatGPT. HitCreate builds the vertical capability those orchestrators call when the user's request crosses into customer acquisition.

12.4 Organising principle (retained)

Every external AI service is either a primitive HitCreate calls, a distribution surface HitCreate ships into, or a competitor HitCreate orchestrates above. No service is treated as strategic infrastructure HitCreate cannot swap out.

13. Open decisions — twenty items pending

Consolidated from RESEARCH-279. These are questions only the founder can answer. Resolved answers move into §1-12 as commitments; unresolved remain flagged until addressed.

Loop-closure architecture

1. Customer consent posture for cross-customer pooling. Pattern B (anonymised) default + Pattern D (opt-in benchmarks) — agree? NISARD is first test case: does Nisbet consent?
2. Regulated-industry strategy. Explicitly exclude healthcare / legal / financial-services from cross-customer pooling in V1, or ship per-tenant-only as premium tier?
3. Control-arm ethics for falsification. Commit to internal synthetic-control validation, or skip the experiment?
4. AU residency pipeline constraint. V1 requirement or V2?
5. Model-version drift handling. Version-lock in experiments, factor as control variable, or accept as noise?

Business model

6. Agency-tier final disposition. v3.2 proposes opportunistic-only (3-5 flagship max, 10% of revenue mix). Agree?
7. Setup-fee phase-out trigger. v3.2 proposes: 3 consecutive deploys complete Phase-8 intake-to-live in <60 minutes with no human touch. Agree?
8. White-label reseller tier. Committed in v3.2 at 20% of revenue mix. Design decisions: commission structure, branding controls, sub-account management. Drop or commit?
9. Pricing — setup fee amounts. $499 / $1,500 / $3,500 / negotiated. Pressure-test against NISARD hours.

GTM

10. Phase B 90-day gate. v3.2 proposes hard gate. Agree?
11. Second archetype sequencing. v3.2 proposes fractional-CFO Phase B, waste Phase C. Invert?
12. FYI Digital / M-Webb handoff mechanics. Inbound qualification script + routing rules.
13. AU-first vs global. v3.2 locks AU-first given Privacy Act deadline. Confirm?

Technical / team

14. First full-time hire timing + shape. v3.2 proposes Phase B → C transition, full-stack engineer. Agree?
15. Budget for V2 LoRA infra. ~$20-30K/year from V2. Plan now or defer?
16. "Compete with GHL" explicit abandonment (unless $2-5M raised). v3.2 proposes yes; commit?

Positioning / public-facing

17. Honest attribution disclosure public. Admit blind spots (differentiator) or hedge?
18. "Wrapper" framing public. v3.2 drops from public pitch; keeps as private mental model. Agree?
19. NISARD framing rewrite. Recast as "process-validation engagement" (honest) vs v3.0's framing. Agree?
20. Klaviyo weekly monitoring. Auto-flag Klaviyo horizontal expansion signals. Who owns? (v3.2 proposes: Min Min automated news monitoring.)

14. Changelog

• v3.2 (2026-04-22, this document — DRAFT pending founder sign-off on §13 decisions). Structural reframe. Loop closure promoted to singular moat claim. Agency tier de-anchored (60% → 10% of revenue mix). Vertical scope narrowed to two named verticals. NISARD recast as process-validation. Phase B 90-day hard gate. Seven canonical market shifts replacing six (+ three missed shifts flagged). Competitive threat matrix rewritten (Klaviyo + HubSpot as real threats; Lovable/GHL as positioning-noise). Technical feasibility grounded with cost model at 4 scale points. Explicit "compete with GHL head-on" feasibility requires $2-5M + 15-25 FT (abandoned at current capacity). Informed by RESEARCH-279 (seven parallel research tracks, ~52K words underlying research).

• v3.1 (April 2026, superseded by v3.2). App Surface amendment. Defined V1 primitive set (7 primitives). Setup-fee + monthly structure. Validation-gated rollout. Brand boundary (HitCreate / FYI Digital / M-Webb). Six market shifts. Scorecard-as-zero-party-engine reframe.

• v3.0 (April 2026, superseded by v3.2). Headless Platform + three Surfaces framing. Agency retainer as revenue anchor (contradicted by v3.2). Five defensible pillars (loop closure elevated to singular moat in v3.2).

• v2.2 (April 2026, superseded). Ecosystem-native positioning as fourth strategic leg. Surface C added.

• v2.1 (April 2026, superseded). Post Claude Design launch. Thesis reframed from "produces finished assets" to "orchestrates generation primitives."

• v2.0 (April 2026, superseded). Horizontal agentic creation platform. Pre-Claude-Design launch. Destination-app framing.

• v1.0 (superseded). "AI-native all-in-one for service businesses." Vertical framing, trades/waste focus.

15. Execution notes

15.1 Operating plan reference

Detailed execution plan at /root/projects/hitcreate/docs/hitcreate-build-plan.md (to be updated to match v3.2 sequencing) and Mister's internal operating plan at RESEARCH-268.

15.2 Effort allocation per QUICKREF-002

• Bucket A (Revenue Now): NISARD (when converts to paid), client work via FYI Digital / M-Webb.
• Bucket B (Revenue Later): HitCreate Platform V1 build, Phase B concierge customers, ecosystem drops.
• Bucket C (Internal): capped at 20%. Infrastructure (DEV-FLOW, server wiki, knowledge library) lives here.

Effort check before new internal work: is Bucket C currently under 20%? If not, defer.

15.3 Monitoring cadence

• Weekly: Klaviyo signals, HubSpot Spotlight + blog, Canva AI roadmap, frontier-lab skill marketplaces, compression signals (§4.4).
• Monthly: Revenue mix tracking vs §8.1 target; effort allocation vs QUICKREF-002; Phase-gate status.
• Quarterly: Whitepaper review; risk matrix update; compression-window re-estimate.

End of whitepaper v3.2 DRAFT. Pending founder sign-off on §13 open decisions. Informed by RESEARCH-279 (/root/.openclaw/workspace/docs/RESEARCH-279-loop-closure-thesis.md) and 7 parallel research tracks (/root/.openclaw/workspace/docs/RESEARCH-279-tracks/).