Warisan Internal — Phased Launch Sequence

Sequencing Logic

Three axes decide what gets built when.

1. Cost to build & run — static < server-light < server-heavy < money-movement
2. Revenue unlocked — does it create a reason to pay today?
3. Regulatory risk — does it touch money, personal data, or legal liability?

The sequence below is ordered so you climb the cost curve only after revenue justifies it.

Phase 0 — Validate · Week 0–4

"Does anyone actually want this?"

Build: nothing new. Ship: MVP landing + free calculator + roadmap.

Cost~RM 20/mo

RevenueRM 0 (intentional)

RegulatoryNone

AINone — calculator is deterministic

What you're testing: will people use the free calculator? Do they ask "what do I do next?" Do they say "can you just do the paperwork for me?"

The manual hack: when someone asks for help, do it by hand over WhatsApp and charge RM99–149. No software. You are the backend. This validates willingness-to-pay before writing a line of backend code.

Gate to Phase 1

≥10 Resolution Packs sold manually. People pay for the paperwork. Proceed.

Phase 1 — Document Packs · Month 1–2

"Sell the paperwork — server-light, high margin"

Build: document generation + a simple one-time payment button.

Cost~RM 100–250/mo

RevenueRM149 × packs · ~95% margin

RegulatoryLow

Break-even~2 packs/month

Why this is first: highest margin, lowest regulatory risk, monetises the traffic Phase 0 already proved. One-time payments to your own account are clean — no e-money issues.

Tech: a Cloudflare Worker that fills PDF templates. Billplz/FPX Checkout for the RM149.

Gate to Phase 1.5

Consistent pack sales + users asking "can you explain this result to me?" Proceed to AI integration.

Phase 1.5 — AI Explanation Layer · Month 2

"First AI integration — Category B only, no personal data"

Deliberately small and scoped. Does not add user-facing features — makes the existing calculator meaningfully better by explaining results in plain BM.

Cost~RM 0–50/mo

RevenueIndirect — conversion lift

RegulatoryLow

AIGroq → Anthropic Claude Sonnet 4.6

Three narrow features only:

Post-calculation explainer — fires an AI call returning a plain BM paragraph explaining why each heir received their share. Input: heir categories and fractions only.
Next-steps suggester — based on track and estate size band, generates a prioritised 3-step action list.
Warisan Score improvement tips — given a score band and missing items, suggest the two most impactful next steps.

Data boundary — enforced in Cloudflare Worker before every AI call:

// This function runs before every outbound AI request
// It is not optional and has no bypass
function sanitiseForAI(payload) {
  return {
    track: payload.track,
    estateRange: bucketValue(payload.estateValue),
    heirCategories: payload.heirCategories,
    distributionResult: payload.fractions,
    scoreRange: scoreToRange(payload.score),
    missingScoreItems: payload.missingItems,
    language: payload.language,
    // Everything else stripped
  };
}

What does NOT change at this phase: Faraid calculation remains deterministic code. Warisan Score calculation remains deterministic code. No document parsing — no IC or property data goes near AI.

Gate to Phase 2

AI explanation feature shows measurable lift in Resolve pack conversion. Proceed.

Phase 2 — Rental Tracking · Month 2–4

"The sticky core — cheap to run, builds the data moat"

Build: property manager + real database + accounts. No money movement yet.

Cost~RM 150–350/mo

RevenueFamily RM19/mo subscriptions

RegulatoryPDPA

Break-even~10 subscribers

Why this is second: the retention engine and data moat. Cheap to run — tracking is just database reads/writes. Every month of data deepens lock-in.

Critical scope discipline: users log payments manually, generate tenancy agreements, see disbursement splits. No money movement, no automated notifications yet.

Gate to Phase 3

Subscribers retain past month 2. Users ask "can it remind my tenant automatically?" Proceed.

Phase 3 — Notifications · Month 4–5

"Automate the nagging — the first variable cost"

Cost~RM 300–900/mo

RevenueUpsell driver

RegulatoryWhatsApp BA

Margin protection: email reminders only on free tier (near-free). WhatsApp reminders only on Family tier — the costly channel becomes the paid upsell.

Gate to Phase 4

MRR comfortably exceeds infra + messaging cost. Legal structure for money-movement confirmed with fintech lawyer. Proceed.

Phase 4 — Payment Collection · Month 5–7

"Move the money — last, most regulated"

CostGateway 1.5–2.9%

RevenueTransaction fee + Estate tier

RegulatoryHIGH — confirm with lawyer first

Why this is last: most regulated. Routing tenant money may trigger Bank Negara e-money rules. Structure so funds flow directly tenant→landlord — Warisan never holds the money. Confirm this exact structure with a lawyer before writing code.

Gate to Phase 5

Payment volume flowing, reconciliation working, no regulatory flags. Proceed.

Phase 5 — Sovereign AI + Institutional · Month 7+

"Self-hosted model, institutional API, document parsing begins"

Triggered by one of three events: monthly Anthropic bill exceeds RM180, a signed institutional contract requiring Malaysian data residency, or a FIKRALab cohort milestone.

CostRM 180–350/mo

RevenueAPI licences RM500–5k/mo

AIQwen3 8B self-hosted, LoRA fine-tuned

Self-hosted Qwen3 8B replaces Anthropic for Category B. Anthropic kept as hot fallback. Category C (document parsing) begins — Malaysian data centre only, data never leaves Malaysian soil.

Phase 6 — Full Sovereign Stack · 2027–2028

"Malaysian data centre, ISO 27001, EPF and Amanah Raya"

CostRM 1,500–3,000/mo

RevenueInstitutional contracts RM500k+

RegulatoryISO 27001 in scope

Not driven by a product decision — driven by an institutional contract requirement. Do not invest in Malaysian data centre infrastructure until a contract requires it. The contract funds the infrastructure.

Summary

The whole picture at a glance.

Phase	What	Monthly Cost	Revenue	AI	Reg Risk
0	Validate	~RM 20	Manual RM149	None	None
1	Document packs	~RM 100–250	RM149 one-time	None	Low
1.5	AI explanation layer	~RM 0–50	Conversion lift	Groq → Anthropic	Low
2	Rental tracking	~RM 150–350	RM19/mo	Category B continues	PDPA
3	Notifications	~RM 300–900	Upsell driver	None new	Medium
4	Payment collection	Gateway %	Txn fee	None	HIGH
5	Sovereign AI + API	~RM 180–350	API licences	Qwen3 8B self-hosted	Medium
6	Full sovereign stack	RM 1,500–3,000	Institutional	All categories	High

Rules to Remember

Three rules to tape to your wall.

Never build the next phase until the current one pays for it. Revenue funds growth, not optimism.
The Faraid engine is always deterministic code. AI explains. Code calculates. This rule does not bend for convenience, speed, or investor pressure.
The expensive, regulated payment piece is LAST — and only after a lawyer confirms you never hold the money.