Where specific costs land for cloud-native, ML/data-intensive B2B SaaS businesses. Use this as a reference when building your financial model.
Cost Waterfall Summary
For a business that acquires data, runs ML models in the cloud, and delivers derived insights via UI/API:
COGS (15-35% of revenue)
├── Customer-facing compute (variable)
├── Customer-specific storage (variable)
├── Customer-specific data (variable)
├── Data subscriptions for delivery (fixed)
├── Payment processing (transactional)
└── Customer support/success (semi-variable)
Variable Costs (10-25% of revenue)
├── Sales commissions
├── Implementation/onboarding
├── Client-specific ML training
└── Variable marketing
OPEX (40-70% of revenue)
├── R&D: Engineering, product, ML research
├── S&M: Base salaries, marketing programs
└── G&A: Executive, finance, HR, legal
Cloud Compute
| Use Case | Classification | Rationale |
|---|
| Production inference (serving client requests) | COGS - Variable - Direct | Directly serves customers, scales with usage |
| Production API hosting | COGS - Variable - Direct | Required to deliver service |
| Reserved production instances | COGS - Fixed - Indirect | Base capacity for delivery |
| Client-specific model training | Variable Cost - Variable - Direct | One-time acquisition cost |
| General ML training/research | OPEX (R&D) - Fixed - Indirect | Product development |
| Dev/test/staging environments | OPEX (R&D) - Fixed - Indirect | Engineering infrastructure |
Cloud Storage
| Use Case | Classification | Rationale |
|---|
| Client data & outputs | COGS - Variable - Direct | Scales with clients, directly attributable |
| Client model artifacts | COGS - Variable - Direct | Required to serve specific client |
| Training data archives (all customers) | COGS - Fixed - Indirect | Required for delivery, shared |
| Dev/test data | OPEX (R&D) - Fixed - Indirect | Engineering resources |
| Logs & analytics | OPEX (R&D/G&A) - Fixed - Indirect | General operations |
Data Acquisition
| Use Case | Classification | Rationale |
|---|
| Multi-year data subscription (all customers) | COGS - Fixed - Indirect | Required to deliver product, shared across customers |
| Per-client commercial data purchase | COGS - Variable - Direct | Purchased specifically for client |
| Free/open data (Sentinel, Landsat) | COGS - Variable - Direct (processing cost) | Processing cost scales with clients |
| Data for product R&D only | OPEX (R&D) - Fixed - Indirect | Product improvement, not delivery |
Key distinction: A multi-year satellite data subscription is COGS (required for delivery) but fixed (doesn't scale with customers). This creates favorable operating leverage — the cost is spread across more customers as you grow.
ML Model Work
| Use Case | Classification | Rationale |
|---|
| Client-specific model training | Variable Cost - Variable - Direct | Onboarding cost, one-time per client |
| Client-specific model retraining | COGS - Variable - Direct | Ongoing delivery cost |
| General model R&D | OPEX (R&D) - Fixed - Indirect | Product development |
| Model serving/inference | COGS - Variable - Direct | Direct delivery cost |
| Pipeline maintenance | OPEX (R&D) - Fixed - Indirect | Platform upkeep |
People Costs
| Role | Classification | P&L Line |
|---|
| Customer Success Manager | COGS - Semi-variable - Direct | Cost of Revenue |
| Support Engineer | COGS - Semi-variable - Direct | Cost of Revenue |
| Implementation Engineer | Variable Cost - Variable - Direct | S&M / Acquisition |
| Sales AE (base salary) | OPEX (S&M) - Fixed - Indirect | S&M |
| Sales AE (commission) | Variable Cost - Variable - Direct | S&M / Acquisition |
| ML Engineer (product) | OPEX (R&D) - Fixed - Indirect | R&D |
| ML Engineer (client work) | COGS or Variable - Direct | Depends on ongoing vs one-time |
| Product Manager | OPEX (R&D) - Fixed - Indirect | R&D |
| Software Engineer | OPEX (R&D) - Fixed - Indirect | R&D |
| Executive | OPEX (G&A) - Fixed - Indirect | G&A |
| Finance/HR/Legal | OPEX (G&A) - Fixed - Indirect | G&A |
Third-Party Services
| Service | Classification | Rationale |
|---|
| Payment processing (Stripe) | COGS - Transactional - Direct | 2.9% + $0.30 per transaction |
| Auth provider (production) | COGS - Semi-variable - Indirect | Required for delivery |
| Analytics (customer-facing) | COGS - Variable - Direct | Part of delivered product |
| CRM (Salesforce) | OPEX (S&M) - Fixed - Indirect | Sales infrastructure |
| Dev tools (GitHub, CI/CD) | OPEX (R&D) - Fixed - Indirect | Engineering tools |
| General analytics (internal) | OPEX (G&A) - Fixed - Indirect | Business intelligence |
Typical Ranges by Category
COGS Components
| Category | Typical % of Revenue | Notes |
|---|
| Customer-facing compute | 5-12% | Higher for ML-heavy products |
| Customer storage | 2-5% | Scales with data volume |
| Customer-specific data | 0-15% | Depends on data strategy |
| Fixed data subscriptions | 5-20% | Higher early, improves with scale |
| Payment processing | 2-3% | Industry standard |
| Customer support/success | 2-5% | Scales with customer count |
| Total COGS | 15-35% | |
| Gross Margin | 65-85% | Lower end for data-heavy |
Variable Costs Components
| Category | Typical % of Revenue | Notes |
|---|
| Sales commissions | 5-10% | 8-12% of ACV typical |
| Implementation | 5-15% | One-time, front-loaded |
| Variable marketing | 2-5% | Performance/paid channels |
| Total Variable | 10-25% | |
OPEX Components
| Category | Typical % of Revenue | Notes |
|---|
| R&D | 15-25% | Engineering, product, ML |
| S&M (fixed) | 15-30% | Base salaries, programs |
| G&A | 8-15% | Executive, finance, HR, legal |
| Total OPEX | 40-70% | Higher for growth-stage |
Operating Leverage from Fixed COGS
When COGS includes significant fixed costs (like data subscriptions), margins improve with scale:
| Scenario | 10 Customers | 50 Customers | 100 Customers |
|---|
| Revenue | $500K | $2.5M | $5M |
| Variable COGS (10%) | $50K | $250K | $500K |
| Fixed COGS (data: $100K) | $100K | $100K | $100K |
| Total COGS | $150K | $350K | $600K |
| Gross Margin | 70% | 86% | 88% |
This is why correctly classifying data subscriptions as fixed COGS (not variable) matters — it reveals the true operating leverage in the business.
Quick Reference
| Cost | Behavior | Function | Traceability |
|---|
| Multi-year satellite data | Fixed | COGS | Indirect |
| Per-client imagery | Variable | COGS | Direct |
| Production GPU inference | Variable | COGS | Direct |
| ML training (client model) | Variable | Variable Cost | Direct |
| ML training (general R&D) | Fixed | OPEX (R&D) | Indirect |
| Client data storage | Variable | COGS | Direct |
| General data archives | Fixed | COGS | Indirect |
| Payment processing | Transactional | COGS | Direct |
| Sales commission | Variable | Variable Cost | Direct |
| Implementation labor | Variable | Variable Cost | Direct |
| Engineering salaries | Fixed | OPEX (R&D) | Indirect |
| CSM salary | Semi-variable | COGS | Semi-direct |
| Pipeline maintenance | Fixed | OPEX (R&D) | Indirect |
| Reserved production infra | Fixed | COGS | Indirect |