2nd/10_Wiki/Topics/AI_and_ML/SaaS.md

---
id: wiki-2026-0508-saas
title: SaaS (Software as a Service)
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Software as a Service, Cloud Software, Multi-tenant SaaS]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [saas, multi-tenancy, subscription, ai-native, vertical-saas]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: TypeScript
  framework: Next.js / Stripe / Postgres RLS
---

# SaaS (Software as a Service)

## 매 한 줄
> **"매 multi-tenant, subscription-based, browser-delivered software — 매 install 의 X, 매 always-latest"**. 매 1999 Salesforce ("End of Software") 로 출발 → 매 2010s SaaS 1.0 (horizontal CRM/HR) → 매 2020s vertical SaaS (Toast, Procore, Veeva) → 매 2026 AI-native SaaS (Glean, Harvey, Cursor) 가 매 outcome-based pricing 으로 매 seat-based 모델 을 흔드는 시기.

## 매 핵심

### 매 Pillars
- **Multi-tenancy**: tenant-isolated data on shared infra (RLS, schema-per-tenant, DB-per-tenant).
- **Subscription**: MRR/ARR, Stripe billing, dunning, proration.
- **Self-serve onboarding**: PLG funnel, time-to-value 분 단위.
- **Continuous delivery**: weekly/daily ship, no version skew.
- **Observability**: per-tenant SLO, usage analytics, churn signals.

### 매 SaaS metrics
- **ARR/MRR**, **NRR** (Net Revenue Retention) — best signal of product fit.
- **CAC payback** — months to recoup acquisition.
- **Magic Number** = (ΔARR × 4) / S&M spend.
- **Gross margin** — typical SaaS 70-85%, AI-SaaS 50-70% (inference cost).
- **LTV/CAC** ≥ 3, payback ≤ 12 months.

### 매 Pricing models
- **Per-seat**: classic Salesforce/Slack — saturating in AI era.
- **Usage-based**: Snowflake, Twilio, OpenAI — aligns to value.
- **Outcome/agent-based**: 2026 AI-native — pay per resolved ticket, per qualified lead.
- **Hybrid**: platform fee + usage overage.

### 매 응용
1. Horizontal SaaS (CRM, HRIS, ITSM, comms).
2. Vertical SaaS (legal, dental, construction, hospitality).
3. Developer SaaS (GitHub, Vercel, Sentry).
4. AI-native SaaS (Cursor, Glean, Harvey, Decagon).
5. Embedded SaaS (in-app commerce, fintech).

## 💻 패턴

### Multi-tenant Postgres RLS
```sql
ALTER TABLE documents ENABLE ROW LEVEL SECURITY;
CREATE POLICY tenant_isolation ON documents
USING (tenant_id = current_setting('app.tenant_id', true)::uuid);

-- App connection middleware
SET LOCAL app.tenant_id = '7f3a...';
```

### Stripe subscription with proration
```typescript
import Stripe from 'stripe'
const stripe = new Stripe(process.env.STRIPE_SECRET!)

await stripe.subscriptions.update(subId, {
  items: [{ id: itemId, price: 'price_pro_monthly', quantity: 25 }],
  proration_behavior: 'create_prorations',
  billing_cycle_anchor: 'unchanged',
})
```

### Usage-based metering (Stripe meters, 2026)
```typescript
await stripe.billing.meterEvents.create({
  event_name: 'api_calls',
  payload: {
    stripe_customer_id: customer.id,
    value: '1',
  },
})
// Subscribe customer to metered price; Stripe aggregates and bills monthly
```

### Tenant context middleware (Next.js)
```typescript
// app/middleware.ts
import { NextResponse } from 'next/server'

export async function middleware(req: Request) {
  const session = await getSession(req)
  const tenantId = session?.tenantId
  if (!tenantId) return NextResponse.redirect('/login')
  const res = NextResponse.next()
  res.headers.set('x-tenant-id', tenantId)
  return res
}
```

### Per-tenant rate limit (Upstash Redis)
```typescript
import { Ratelimit } from '@upstash/ratelimit'
import { Redis } from '@upstash/redis'

const limiters = new Map<string, Ratelimit>()
function tenantLimiter(plan: 'free' | 'pro' | 'enterprise') {
  const limits = { free: 100, pro: 1000, enterprise: 10000 }
  return new Ratelimit({
    redis: Redis.fromEnv(),
    limiter: Ratelimit.slidingWindow(limits[plan], '1 m'),
  })
}
```

### PLG signup with magic-link (no password)
```typescript
// Send magic link, no password friction
const token = jwt.sign({ email, tenantId: nanoid() }, SECRET, { expiresIn: '15m' })
await sendEmail(email, `https://app.example.com/auth?t=${token}`)
// On click: provision trial tenant, redirect to onboarding wizard
```

### AI-native SaaS — usage-aware inference cost
```typescript
async function chat(tenantId: string, msgs: Message[]) {
  const tier = await getTier(tenantId)
  const model = tier === 'enterprise' ? 'claude-opus-4-7' : 'claude-haiku-4-5'
  const resp = await anthropic.messages.create({
    model, max_tokens: 1024, messages: msgs,
    metadata: { user_id: tenantId },
  })
  await meterUsage(tenantId, {
    input_tokens: resp.usage.input_tokens,
    output_tokens: resp.usage.output_tokens,
    model,
  })
  return resp
}
```

### Outcome pricing instrumentation
```typescript
// Charge only when AI agent successfully resolved
async function recordResolution(tenantId: string, ticketId: string, resolved: boolean) {
  if (resolved) {
    await stripe.billing.meterEvents.create({
      event_name: 'resolved_ticket',
      payload: { stripe_customer_id: tenantId, value: '1' },
    })
  }
}
```

### Tenant-isolated S3 (per-prefix IAM)
```python
# Generate scoped STS token per tenant request
sts = boto3.client("sts")
policy = {"Version": "2012-10-17", "Statement": [{
    "Effect": "Allow",
    "Action": ["s3:GetObject", "s3:PutObject"],
    "Resource": [f"arn:aws:s3:::tenants-bucket/{tenant_id}/*"],
}]}
creds = sts.assume_role(
    RoleArn=ROLE, RoleSessionName=f"tenant-{tenant_id}",
    Policy=json.dumps(policy), DurationSeconds=900,
)
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Early-stage horizontal | Shared schema + RLS, Stripe per-seat |
| Vertical w/ compliance | Schema-per-tenant or DB-per-tenant |
| AI-native, variable cost | Usage-based + cap to protect margin |
| Enterprise w/ SOC2/HIPAA | Single-tenant deploy option, BYOC |
| PLG self-serve | Magic-link + provisioned trial in <60s |

**기본값**: shared-DB RLS + hybrid pricing (platform fee + usage) + Stripe meters + PLG onboarding.

## 🔗 Graph

## 🤖 LLM 활용
**언제**: in-product copilot, customer support deflection, churn prediction from usage signals, content/email generation, dynamic onboarding.
**언제 X**: pricing/billing computation — must be deterministic for audit and revenue recognition.

## ❌ 안티패턴
- **No tenant isolation**: WHERE tenant_id checked only in app layer → IDOR breach.
- **Per-seat pricing for AI**: high-usage user breaks margin; need usage cap or tier.
- **Free tier without limits**: abuse → infra cost spirals.
- **Single-region SaaS for global**: latency + data residency violations (GDPR).
- **No self-serve cancel**: regulatory risk (FTC click-to-cancel 2024) + churn spikes.
- **Version skew**: different customers on different versions → support combinatorics explode.

## 🧪 검증 / 중복
- Verified (a16z SaaS metrics, OpenView PLG benchmarks, Stripe Billing docs, AWS SaaS Lens).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — multi-tenancy, pricing models, AI-native SaaS 2026 |