Walkthrough: Launch a Multi-Tenant B2B SaaS Platform (Architecture, Compliance, GTM)
This walkthrough designs a greenfield multi-tenant B2B SaaS platform end-to-end: target customer is mid-market enterprise (500-5,000 employees), product is a vertical workflow application with AI assistance, pricing 50K-500K annual contract values (ACV). Architecture covers the front-end, API tier, data layer, AI/ML layer, identity, billing, observability, CI/CD, security perimeter, and the compliance pathway through SOC 2, GDPR, HIPAA, and (optional) ISO 27001 + FedRAMP Moderate. Launch CAPEX $200-400K (mostly engineering hours capitalized + initial vendor commits); first-year OPEX scales with ARR growth curve.
Reference launches and architectural patterns: Notion, Linear, Vercel, Retool, Ramp, Mercury, Brex, Snowflake, Databricks, Figma, Cribl, Sigma Computing, Vanta, Drata, Modal Labs. Public engineering writeups: Notion’s PostgreSQL sharding (2023), Figma’s multiplayer architecture (2019, 2024), Linear’s sync engine, Vercel’s edge runtime, Stripe’s reliability engineering, Cloudflare Workers + Durable Objects model.
1. Product and platform spec
| Parameter | Target | Notes |
|---|---|---|
| Tenancy model | Pool with per-tenant isolation tiers | Shared DB / pool-namespace / silo by plan |
| Target tenants Year 1 | 80-200 paying tenants | $4-12M ARR target Year 1 exit |
| Users per tenant | 25-2,500 | Power-law, 80% under 200 |
| Regions | US-East-1 (primary), EU-West-1 (Year 1), AP-Southeast-1 (Year 2) | Customer-data-residency requirement |
| Availability SLA | 99.9% uptime (43.8 min/month downtime budget) | 99.95% target for Enterprise tier |
| RTO / RPO | RTO 4 hr / RPO 15 min | Multi-AZ; cross-region DR Year 2 |
| Latency target | p50 <150 ms, p95 <500 ms server-side | LCP <2.5s, INP <200ms client |
| Compliance | SOC 2 Type II in Year 1, GDPR ready Day 1, HIPAA Year 2 | ISO 27001 + FedRAMP Year 3 optional |
| Auth | SAML, OIDC, SCIM provisioning, MFA enforced | Plus social/email for SMB self-serve |
| API | REST + Webhooks; GraphQL for internal; OpenAPI 3.1 spec | Public + private surface |
| AI capability | Embedded LLM features (search, drafting, summarization, agents) | RAG over tenant data + tool-use |
| Launch CAPEX | $200-400K | Engineers + vendor initial commits |
2. Tenancy and isolation model
Three tiers of tenant isolation, allocated by plan + risk:
- Pool tier (default) — shared DB cluster, tenant_id discriminator on every row, row-level security (Postgres RLS) enforced. Cheapest; 90-95% of tenants. Cross-tenant leakage prevented by RLS policies tested in CI with adversarial test suite. See sql-nosql-design.
- Namespace tier — dedicated schema per tenant in shared cluster; same compute pool but separate logical DB. Used for Enterprise tier + customers with custom retention requirements.
- Silo tier — fully dedicated AWS account or Kubernetes namespace, optionally separate VPC. Reserved for HIPAA/regulated workloads + top-100-ACV accounts. Provisioned via Terraform module from
tenant-silo.
All tiers use the same application image — isolation lives at infrastructure, not code. Tenant routing happens at the edge (Cloudflare Worker or AWS Lambda@Edge) based on host header and JWT claims.
References: AWS SaaS Lens (Tod Golding’s “SaaS Architecture Fundamentals” 2020-2024), the canonical Pool/Silo/Bridge model.
3. Frontend tier
3.1 Web app
- Framework: Next.js 15 (App Router, React 19) deployed on Vercel (Pro plan, ~$2,400/mo for the team + bandwidth) or Cloudflare Pages.
- Why Next.js: SSR + RSC + Server Actions cover the marketing site, app shell, and most app routes; co-located code is easier to staff than a Next-marketing + Vite-app split.
- Why Vercel over self-hosted: 20K/mo. Vercel KV (Upstash Redis under the hood) + Edge Config for feature flags + Edge Middleware for auth-gate.
- Component library: shadcn/ui + Radix UI primitives, Tailwind CSS 4. No Material UI / Chakra (theming + bundle size).
- State: React Query (TanStack Query 5) for server state; Zustand for client; URL state via Next router and
nuqs. No Redux. - Realtime: Liveblocks ($500-2,000/mo) or Pusher Channels for presence + cursors; Server-Sent Events for one-way streams (LLM tokens). WebSockets only where required.
3.2 Marketing site
Separate sub-app — Astro 5 + MDX for content, deployed on Cloudflare Pages. The marketing site and the app share only the design tokens + Tailwind config, not React components. Marketing emails via Resend (~$20-200/mo) with React Email templates.
3.3 Mobile
Mobile shell via React Native + Expo (EAS Build 20M.
4. API tier
4.1 Service runtime
- Primary language: Go 1.23+ for the API services. Rationale: low memory footprint per replica (60-120 MB vs 300-600 MB for equivalent Node), trivial concurrency primitives, strong typing, fast cold-start. See concurrency-primitives.
- Framework: stdlib
net/http+ chi router; no Gin/Fiber (avoid framework lock-in). - Schema-first: OpenAPI 3.1 spec is source of truth; generates Go server stubs (via
oapi-codegen) and TypeScript client (viaopenapi-typescript-codegen). PR-blocking spec lint via Spectral. - GraphQL (internal only): for the dashboard’s complex aggregation queries. gqlgen on Go side, Apollo Client + codegen on TS side. Public API stays REST + Webhooks — easier for customer integration teams.
4.2 Cluster
- Kubernetes on AWS EKS (managed control plane $73/mo per cluster + node costs). EKS Auto Mode (GA Dec 2024) eliminates 90% of node-group management; cluster-autoscaler + Karpenter for spot node pools.
- Node selection: c7g.large/xlarge Graviton instances (ARM64) for the API tier — ~20% cheaper per vCPU than x86, ~40% better perf/W. m7i for any x86-only dependencies (ML inference with non-ARM CUDA paths in tooling). See realtime-embedded (ARM Cortex-A context).
- Service mesh: Linkerd (lighter than Istio; mTLS by default, automatic retries, no Envoy sidecar cold-start tax). Avoid Istio at Year 1 — too much complexity per engineer. See containers-service-mesh.
- Why EKS over GKE / Fly.io / Render: most enterprise customers’ compliance teams have audited AWS; GCP audit packet is harder to procure. ECS/Fargate-only was considered (simpler ops, no nodes) but rejected for service-mesh + per-pod GPU + custom scheduling needs.
4.3 Service decomposition
Resist microservice fragmentation. Year 1 service count target: 6-10 services, not 50.
| Service | Responsibility | Language |
|---|---|---|
gateway | TLS termination, JWT validation, rate limit, request routing | Go |
api-core | Primary CRUD + business logic | Go |
api-billing | Stripe webhook handling, invoice gen, usage metering | Go |
api-ai | LLM proxy, prompt assembly, RAG retrieval, tool-call dispatch | Python/FastAPI (LangChain/LangGraph ecosystem) |
api-search | Elasticsearch/OpenSearch query layer | Go |
workers | Async background jobs (export, email, webhook delivery, ETL) | Go |
realtime | WebSocket fan-out, presence | Go (centrifugo or hand-rolled) |
web | Next.js SSR (on Vercel, not in EKS) | TypeScript |
See microservices-patterns for the decomposition reasoning.
5. Data layer
5.1 OLTP primary database
- PostgreSQL 16 on AWS Aurora PostgreSQL — managed, ~30% cheaper than DIY RDS, native read-replica scaling, zero-downtime patching, point-in-time restore.
- Cluster: 1 writer + 2 readers, r7g.2xlarge (8 vCPU, 64 GB RAM) Year 1. Scale to db.r7g.8xlarge or sharded with Citus by ~$10M ARR.
- pgvector extension for embeddings (HNSW indexes, m=16, ef_construction=64 baseline). See rag-embeddings-vector-search.
- RLS policies on every tenant-scoped table; CI test asserts no query bypasses RLS without explicit
bypass_rlsrole. - Backup: continuous via Aurora to S3; weekly logical pg_dump archived to glacier; quarterly DR-restore drill.
See database-internals.
5.2 Cache + queue
- Redis (ElastiCache Serverless) for cache + ephemeral state + rate-limit counters. Serverless tier auto-scales; pay per RPU.
- SQS (queues) + EventBridge (cross-service events) for async — preferred over Kafka at Year 1 (Kafka adds 2 ops engineers worth of complexity). Move to MSK (managed Kafka) at ~$10M ARR when fan-out + replay matter.
- S3 for blob storage; lifecycle rules to Glacier after 90 days for tenant exports.
5.3 Analytics + warehouse
- Snowflake for warehouse (~$20-150K/yr Year 1). Fivetran or Airbyte to land Postgres CDC + Stripe + Salesforce + HubSpot data into raw schemas; dbt Cloud for transforms. See sql-nosql-design.
- Why Snowflake over Databricks / BigQuery: most B2B SaaS customer-success teams know SQL not Spark; usage-based pricing fits volatile B2B workload better than BigQuery’s slot model at small scale. Revisit at >$50M ARR.
- Tenant analytics surfaced in-app: pre-aggregated rollups in Postgres (materialized views, refreshed via pg_cron) or ClickHouse Cloud for high-cardinality dashboards.
5.4 Search
- OpenSearch (AWS managed) for product search + log search. ~$800-3,500/mo. Elastic abandoned in 2021-2024 due to license drama; the OpenSearch fork stabilized.
- Tantivy / Meilisearch / Typesense considered for typeahead — Typesense Cloud easier and ~$50-500/mo.
6. AI/ML layer
6.1 Model strategy
- Anthropic Claude (Sonnet 4.7, Opus 4.7) as primary frontier model via Anthropic API + AWS Bedrock failover. OpenAI GPT-4.x as secondary (avoid hard lock-in to one provider).
- Open-source weights (Llama 4, Qwen 3, Mistral Large) self-hosted on Modal or RunPod for cost-sensitive bulk tasks (classification, summarization at >1M req/day where API spend would exceed $50K/mo).
- Embeddings: OpenAI text-embedding-3-large (3,072 dim) or Voyage AI voyage-3-large; stored in pgvector. See rag-embeddings-vector-search.
- Speech: Deepgram or AssemblyAI for transcription; ElevenLabs or Cartesia for TTS — only if voice is a feature.
6.2 Application architecture
- Prompt assembly in
api-aiservice: system prompt + tenant-config-injected context + RAG-retrieved chunks + tool-definitions. - Tool-use / agents: tools defined in OpenAPI, executed in
api-corewith full tenant-RBAC enforcement; LLM call returns a tool-call,api-aidispatches to internal endpoints, observes result, recursive call. - Retrieval pipeline: ingest → chunk (semantic chunking, 500-1500 tokens with overlap) → embed → upsert to pgvector. Query-time: hybrid BM25 + vector with reciprocal-rank fusion, then rerank with Cohere Rerank 3 or Voyage rerank-2 (top 50 → top 10).
- Evaluation: Braintrust, Langfuse, or self-hosted Phoenix (Arize) for trace + eval datasets. Manual + LLM-judge eval per release on a frozen 200-prompt golden set. See prompt-engineering-agent-systems.
6.3 Safety + cost
- Prompt-injection defense: structured tool-input parsing, no untrusted-data execution paths, output filtering for secret-token shapes.
- Cost guardrails: per-tenant LLM spend cap (set in billing engine), Bedrock cross-account guardrail policies, model fallback ladder (Opus → Sonnet → Haiku → cached completion).
- PII handling: pre-redaction via Microsoft Presidio + custom regex pipeline; tenant-config toggle for “send raw to LLM”.
7. Identity, authentication, authorization
7.1 Auth provider
- Auth0 (1,500-7,000/mo at scale) OR Clerk (125/mo + per SSO connection) for SAML/OIDC/SCIM.
- For B2B SaaS targeting mid-market+, WorkOS is the strongest fit: pricing scales linearly with enterprise SSO connections, SCIM is first-class, audit-log export is built in. Auth0 is heavier and harder to leave once embedded.
- Self-hosted alternatives (Keycloak, Ory Kratos, SuperTokens): rejected for Year 1 — auth bugs are existential, and an in-house team takes 3-6 engineer-months to ramp securely.
- See auth-authz.
7.2 Session + token model
- Short-lived JWT access tokens (15 min) signed RS256 with rotating keys (JWKS published at
/.well-known/jwks.json). - Refresh tokens stored in HTTP-only Secure SameSite=Lax cookies; sliding-expiration 30 days.
- API keys per tenant for programmatic access; hashed at rest with HMAC-SHA-256, prefix shown in UI for identification (“sk_live_…“).
7.3 Authorization
- Role-based + attribute-based hybrid via OpenFGA (Auth0’s fork of Google Zanzibar) or Permify. Resource model: tenant → workspace → object → role. Permissions checked at API boundary AND row-level (via Postgres RLS).
- Tenant admins manage their own RBAC via in-app UI; system admin scopes are immutable in code.
8. Billing and revenue ops
8.1 Billing platform
- Stripe Billing as billing engine. ~2.9% + 5; invoicing for net-30 + enterprise terms.
- Subscription management directly in Stripe via the JSON API + Stripe Customer Portal embedded.
- Metronome ($1.5-5K/mo + take-rate) for usage-based metering if pricing includes per-event consumption (LLM tokens, API calls). Otherwise direct usage records in Stripe.
- Maxio (Chargify) or Ordway considered for complex CPQ; rejected Year 1 (over-engineered).
8.2 Revenue stack
- Salesforce + HubSpot (depending on GTM maturity). Most pre-Series-B SaaS run HubSpot ($1,200-15,000/mo); Series-B+ migrate to Salesforce.
- Outreach / Salesloft for sales engagement; Apollo.io for prospecting.
- CPQ: Stripe Quote → Stripe Invoice; no Salesforce CPQ Year 1.
- Revenue recognition: Stripe Revenue Recognition module (ASC 606 compliant out-of-box) feeds NetSuite or QuickBooks. See accounting-foundations.
8.3 Payment risk
- Radar (Stripe’s fraud engine) for card transactions.
- Sift or Forter Year 2 if self-serve SMB grows and chargeback rate >0.5%.
9. Observability
9.1 Stack choice — Datadog vs OTel-native
Two paths:
| Path | Cost Year 1 | Pros | Cons |
|---|---|---|---|
| Datadog all-in | $50-180K/yr | Fastest setup, integrated APM+RUM+logs+infra | Lock-in; egress-pricing surprise at scale; agent CPU cost |
| OTel + Grafana Cloud or self-hosted | $20-80K/yr | Standards-based (OTLP), portable, cheaper at scale | More integration work; SRE time |
Recommendation Year 1: Datadog (20K/mo. Keep APM on Datadog longer. See observability-stack.
9.2 Instrumentation
- OpenTelemetry SDK in every service (Go + Python + TS), OTLP exporter routed to Datadog OTel collector. Vendor-neutral instrumentation now de-facto standard; portable.
- Structured logs: zerolog (Go), structlog (Python), pino (Node); JSON to stdout, scraped by Vector or Datadog agent.
- Metrics: Prometheus-style histograms; per-tenant labels gated by cardinality budget (no
user_idlabels — explodes cardinality). - RUM (real user monitoring): Datadog RUM or Sentry — for frontend Core Web Vitals + JS error tracking.
9.3 SLOs
- API gateway p95 latency < 500 ms / 99.5% error-free over 28 days.
- LLM endpoint p95 < 8 s, error rate < 2%.
- Background job latency p99 < 60 s for user-facing tasks; < 15 min for daily batch.
- Burn-rate alerts at 2% and 5% budget consumption per 1 hr and 6 hr windows.
10. CI/CD and developer experience
10.1 Pipeline
- GitHub Enterprise Cloud ($21/user/mo) for source + actions. Branch protection requires 2 reviews + CI green + signed commits.
- GitHub Actions for CI: build + lint + unit test + integration test (testcontainers-go, dockertest, ephemeral Postgres) + container build + sign with cosign + push to ECR.
- ArgoCD on EKS for GitOps deploys:
main→ staging auto, manual approval → prod. ApplicationSets for per-tenant silo deploys. - Trunk-based development with short-lived feature branches. Atlas (Ariga) or pgroll for online Postgres schema migrations.
10.2 Environments
- dev — local Docker Compose + ngrok webhook tunneling
- preview — ephemeral per-PR namespace in staging EKS, with seeded test tenant; teardown after 7 days
- staging — full-fidelity replica of prod, real auth (separate tenant), sanitized PII-free data refresh nightly
- prod — multi-AZ, multi-region eventually
10.3 Feature flags
- LaunchDarkly ($600-3,000/mo) or Statsig (lower price + experimentation built in) or self-hosted Unleash.
- Every new feature ships behind a flag; gradual rollout 1% → 10% → 50% → 100% with kill-switch.
10.4 Secrets
- HashiCorp Vault (HCP managed, $0.30-1.50/secret-hr) or AWS Secrets Manager for runtime secrets.
- Doppler or Infisical for developer-day-1 secret syncing.
- No secrets in
.envfiles in git; pre-commit hook scans withggshieldortrufflehog.
11. Security perimeter
11.1 Network
- Cloudflare in front of everything: DNS + DDoS + WAF + Bot Management. ~$200-2,000/mo Pro/Business; Enterprise plan negotiable.
- AWS WAF as second-line on ALB; CloudFront in some scenarios for static caching.
- VPC architecture: 3-AZ in us-east-1 + eu-west-1. Private subnets for compute; NAT gateways with regional egress IP allowlists; transit gateway for cross-region. PrivateLink endpoints for AWS services (no NAT data charges).
11.2 Content security
- CSP header strict — no
unsafe-inline, nonce-based script tags via Next middleware. - HSTS preload-listed (max-age 1 year + includeSubDomains + preload).
- CORS locked to
app.{tenant}.{rootdomain}patterns; no*. - CSP report-uri to Sentry or csp-collector.
11.3 Application
- OWASP Top 10 baseline; SAST via Semgrep + GitHub Advanced Security; SCA via Dependabot + Snyk or Mend.
- SBOM generated on every container build via Syft; signed with cosign; pushed to OCI registry.
- Container scan with Trivy in CI + Amazon ECR scan on push.
- Secret scanning: GitGuardian for org-wide pre-commit + post-push.
11.4 Pen testing + bounty
- Annual third-party pen test (HackerOne pentest service, Bishop Fox, NCC Group) ~$30-80K.
- Public bug bounty (HackerOne or Bugcrowd) ~$30K-150K/yr in bounties at Year 2-3.
12. Compliance pathway
12.1 SOC 2 Type II
- Vanta or Drata ($15-40K/yr) for evidence collection + control monitoring + auditor liaison.
- Audit firm: Prescient Assurance, A-LIGN, or Schellman (35-80K Type II).
- Timeline: Type I in 3-6 months (point-in-time); Type II in 9-12 months (continuous evidence over 6-month observation window).
- Trust Services Criteria: Security required; Availability + Confidentiality strongly recommended for SaaS; Processing Integrity + Privacy if relevant.
12.2 GDPR
- DPA (Data Processing Agreement) template ready Day 1, signed with each EU customer.
- SCCs (Standard Contractual Clauses, EU 2021/914) for transfers; or EU-US Data Privacy Framework (DPF) registration if processing EU PII in US.
- DSR (Data Subject Rights) tooling: export, delete, rectify endpoints exposed in API + admin UI. Most SaaS use Transcend or DataGrail ($15-50K/yr) for the request-handling workflow.
- EU data residency: separate eu-west-1 deployment with EU-only customer routing; no replication to US for those tenants.
12.3 HIPAA (Year 2)
- BAA (Business Associate Agreement) signed with AWS, Stripe, Datadog (Sensitive Data Scanner add-on), Cloudflare Enterprise. Vercel signs BAA on Enterprise tier only.
- Anthropic + OpenAI: BAA available; per-customer enablement; logs disabled.
- Encryption-at-rest mandatory (Aurora KMS, S3 SSE-KMS), encryption-in-transit mandatory (TLS 1.2+; HSTS).
- HIPAA-eligible tenants placed in silo tier (Section 2) with dedicated AWS account + restricted IAM.
12.4 ISO 27001 + FedRAMP (Year 3)
- ISO 27001 expands SOC 2 evidence base; cert via BSI or Schellman; ~$60-120K external cost.
- FedRAMP Moderate: $1.5-5M sponsor-led process via a 3PAO (Coalfire, Schellman, Kratos); 18-30 month timeline; only worth it for govt-adjacent verticals.
12.5 Sub-processor disclosure
Public subprocessors.html page listing every sub-processor (AWS, Stripe, Anthropic, OpenAI, Datadog, Vercel, etc.) — required by GDPR Art 28; expected by every Enterprise procurement team. Updated 30 days before any addition.
13. Launch CAPEX + Year-1 OPEX
13.1 Pre-launch CAPEX (engineering hours capitalized + setup)
| Item | Cost |
|---|---|
| Engineering team (6 FTE × 8 mo to GA × $14-18K loaded mo) | $750,000 |
| Design + frontend (2 FTE × 6 mo × $13-16K) | $180,000 |
| Security + compliance bootstrap (consulting + Vanta + first Type I audit) | $85,000 |
| Logo / brand / marketing site / launch content | $40,000 |
| Initial vendor commits + non-refundable retainers (Snowflake credit, AWS) | $60,000 |
| Legal (ToS, DPA, MSA, privacy policy, IP filings) | $35,000 |
| Office tooling (Linear, Notion, Figma, GitHub, 1Password) Year 1 | $24,000 |
| Pre-launch capitalized total | ~$1,174,000 |
The “245K. Most SaaS founders quote the lower bound; the fully-loaded number (including burn-equivalent engineering) is materially higher.
13.2 Year-1 OPEX (200 tenants exit)
| Line | Annual cost |
|---|---|
| Infrastructure (AWS EKS, Aurora, ElastiCache, S3, transit) | $180,000 |
| Vercel + Cloudflare + edge | $48,000 |
| LLM API spend (Anthropic + OpenAI + Voyage) | $260,000 |
| Snowflake + Fivetran + dbt Cloud | $95,000 |
| Datadog (APM + Logs + RUM) | $120,000 |
| Auth0/WorkOS + Stripe (no separate cost line beyond rev-share) | $42,000 |
| Vanta + audit firm Type II | $58,000 |
| Email/SMS (Resend, Twilio, Postmark transactional) | $14,000 |
| SaaS tools (Linear, Notion, Figma, GitHub, 1Password, Slack) for 30 staff | $95,000 |
| Security tooling (Snyk, Semgrep Pro, GitGuardian, HackerOne) | $48,000 |
| Insurance (cyber + E&O + D&O) | $85,000 |
| Legal + accounting + R&D tax credit prep | $120,000 |
| Engineering headcount Year 1 (15 FTE blended $230K loaded) | $3,450,000 |
| GTM headcount (8 FTE blended $190K loaded) | $1,520,000 |
| Operations + finance + people (5 FTE blended $170K) | $850,000 |
| Marketing programs + paid acquisition | $400,000 |
| Conferences + travel | $80,000 |
| Total Year-1 OPEX | ~$7,465,000 |
At 1-3M MRR), most B2B SaaS burns $3-5M Year 1 net of revenue.
14. Risk register
- AWS region outage — multi-AZ helps, multi-region required for 99.95% SLA. us-east-1 has had material outages (Dec 2021, Jun 2023, Dec 2024).
- Vendor concentration: Vercel + Anthropic + AWS + Stripe — failure of any one impacts product. Pre-negotiated fallback contracts.
- LLM API rate limits + pricing changes — Anthropic and OpenAI both raised + cut prices in 2024-2025; build a routing layer (LiteLLM, OpenRouter, or in-house) from Day 1.
- GPU supply — if self-hosting models, H100/H200 capacity is gated by allocation queues; Modal, Together, RunPod, Lambda Labs provide elasticity but with markup.
- Compliance creep: HIPAA + FedRAMP each adds ~$500K-2M/yr loaded engineering + audit cost. Don’t accept these in MSA term sheets without pricing it.
- Data residency politics: EU-US DPF could be invalidated again (Schrems III pending CJEU 2026); plan EU silo from Day 1, not retrofit.
- AI hallucination liability: explicit ToS limitation, customer-side review workflows for any external-facing generated content, internal eval gates on accuracy-critical surfaces.
- Founder/team key-person: maintain CISO + DPO designated roles even when CEO wears the hat early; both become audit-required at SOC 2 Type II and GDPR scale.
15. GTM and launch sequencing
15.1 Pre-launch (months -6 to 0)
- Design partner cohort: 8-15 customers at $0 or 50% discount in exchange for product feedback + case studies. NDA + LOI signed.
- Closed beta — feature-flag-gated, invite-only via Stripe Checkout link in waitlist confirmation.
- SOC 2 Type I audit started month -3, completed month 0.
- Documentation site (Mintlify, Docusaurus, ReadMe) — API reference, getting-started, integration guides, security overview.
- Trust Center (Vanta Trust Report, SafeBase, Drata Trust Center) — public surface for security posture.
15.2 Launch (month 0)
- Product Hunt + Hacker News + LinkedIn coordinated post.
- Founder podcast / interview circuit (Lenny’s Newsletter, Acquired, 20VC).
- 3-5 customer case studies published Day 1.
- Pricing page live with self-serve start (no credit-card-free signups — drives 10× signup volume but tanks conversion + creates abuse vector).
15.3 Post-launch (months 1-12)
- ICP refinement via cohort analysis (which tenants convert + expand?).
- Outbound SDR program once $1M ARR.
- Partner ecosystem: integration directory with HubSpot, Salesforce, Slack, Microsoft Teams, etc.
- Series A fundraise once metrics support: 35M @ 250M @ 200M @ $13B.
16. Adjacent
- design-data-center-cooling-system — the physical layer the SaaS depends on
- distributed-systems-fundamentals — partitioning, consistency, failure modes
- database-internals — Postgres internals + RLS
- rag-embeddings-vector-search — the AI/ML retrieval layer
- auth-authz — identity + RBAC patterns
- observability-stack — Datadog vs OTel decision
- accounting-foundations — ASC 606 revenue recognition for subscription SaaS