The CFO's AI Question
Every CFO at a service organization using AI faces the same question:
"Is AI spend COGS or R&D?"
This isn't an academic exercise. The answer directly impacts:
- Gross margin — The headline metric for business quality
- Company valuation — Especially for SaaS and services firms
- Investor perception — Whether AI is an investment or a cost drain
- Financial reporting — Compliance with GAAP/IFRS standards
Get it wrong, and you either overstate margins (misleading investors) or understate them (undervaluing your business). Neither is acceptable.
Understanding AI Costs on the P&L
Above the Line vs. Below the Line
The income statement has a critical dividing line: gross profit.
Simplified Income Statement
Above the line (COGS): Costs directly tied to delivering what you sold. For a services firm: consultant salaries, subcontractor fees, and — increasingly — AI inference costs used in delivery.
Below the line (R&D, SG&A): Costs to build products, run the business, or enable sales. Not directly tied to delivering a specific client engagement.
Why Gross Margin Matters
Gross margin = Gross Profit ÷ Revenue
For the example above: $4M ÷ $10M = 40% gross margin.
This metric tells investors:
- Unit economics — Does each dollar of revenue generate profit?
- Scalability — Can the business grow without proportional cost growth?
- Business model quality — Is this a high-margin SaaS or low-margin labor shop?
Valuation Impact
A 10-point difference in gross margin can mean a 2-3x difference in valuation multiple. Classifying AI costs incorrectly isn't just an accounting issue — it's a valuation issue.
AI Cost Classification Framework
Here's a practical framework for classifying AI costs:
| AI Usage | Classification | Rationale |
|---|---|---|
| AI inference for client deliverables | COGS | Direct cost of delivery |
| AI used in client-facing services | COGS | Enables revenue generation |
| Training custom models | R&D | Creating an asset (may capitalize) |
| Building AI product features | R&D | Product development |
| Internal productivity tools (Copilot, etc.) | SG&A | Operational efficiency |
| Sales/marketing AI (lead scoring, etc.) | SG&A | Sales enablement cost |
The Key Test
Ask yourself:
"Would I incur this AI cost if I didn't have this specific client or project?"
Yes → Below the line (R&D or SG&A)
No → Above the line (COGS)
Real Examples
✓ COGS Example
A consulting firm uses Claude to analyze client data and generate insights for a deliverable. The tokens consumed are directly tied to the engagement → COGS.
✓ R&D Example
A software company fine-tunes an LLM to improve their product's capabilities. The training costs are building an asset → R&D (potentially capitalize).
✓ SG&A Example
A company pays for GitHub Copilot licenses for all engineers. This improves productivity but isn't tied to specific clients → SG&A.
Gross Margin Impact Analysis
Let's see how AI affects margins in practice:
Scenario 1: AI Improves Margin
| Metric | Traditional | AI-Augmented |
|---|---|---|
| Project Revenue | $100,000 | $100,000 |
| Labor Cost | $65,000 | $45,000 |
| AI Cost | $0 | $5,000 |
| Total COGS | $65,000 | $50,000 |
| Gross Margin | 35% | 50% |
Result: AI reduces labor cost by more than it costs, improving margin by 15 points.
Scenario 2: AI Erodes Margin
| Metric | Traditional | AI-Heavy |
|---|---|---|
| Monthly Revenue | $10,000 | $10,000 |
| Platform Cost | $500 | $500 |
| AI Cost (variable) | $0 | $4,500 |
| Total COGS | $500 | $5,000 |
| Gross Margin | 95% | 50% |
Result: AI-heavy offering with fixed pricing and variable AI costs destroys margin.
The "Unlimited AI" Trap
Many companies offer "unlimited AI features" as part of flat-rate subscriptions. This creates a dangerous economic trap:
The Unit Economics Problem
- You charge fixed monthly fee: $500/month
- Customer uses AI heavily: $400/month in tokens
- Your gross margin on this customer: 20%
- You need 80%+ gross margin for SaaS economics to work
Power User Cost Concentration
AI usage follows a power law distribution. A small number of users drive the majority of costs:
- Top 10% of users often generate 60-80% of AI costs
- Bottom 50% of users may generate less than 5% of costs
- Your average customer is profitable; your power users are destroying margin
The Fix
- Usage-based pricing — AI costs scale with value delivered
- Tiered limits — Cap AI usage per pricing tier
- Cost visibility — Know which customers drive AI costs
Accounting Treatment Guidance
GAAP/IFRS Considerations
The accounting treatment of AI costs depends on their nature:
| Cost Type | Treatment | Standard |
|---|---|---|
| AI inference (client delivery) | Expense as incurred (COGS) | ASC 606 / IFRS 15 |
| Model training (internal use) | May capitalize if criteria met | ASC 350-40 |
| Model training (for sale) | Capitalize development costs | ASC 985-20 / IAS 38 |
| AI SaaS subscriptions | Expense as incurred | By function (R&D/SG&A) |
Capitalization of AI Development
For internal-use AI (ASC 350-40), capitalization is allowed when:
- Preliminary project stage is complete
- Management commits to funding the project
- It's probable the project will be completed and used
- The AI will provide expected functionality
Auditor Alert
AI cost capitalization is an emerging area. Expect auditor scrutiny on capitalization decisions. Document your rationale clearly.
Building AI Cost Visibility for Finance
Finance teams need specific data to properly account for AI costs:
What Finance Needs
Total AI spend by month
Aggregate view for trend analysis
AI spend by cost center
For proper P&L allocation
AI spend by client/project
For COGS attribution
AI spend by use case
For COGS vs. R&D vs. SG&A split
Unit economics (cost per output)
For pricing decisions
Attribution Requirements
To properly allocate AI costs to COGS, you need:
- Client/project tagging — Every AI call tied to a billing entity
- Use case classification — Is this delivery, R&D, or operations?
- Cost aggregation — Roll up from API calls to monthly totals
Deep dive: AI Cost Allocation — The Complete Guide covers the technical implementation of AI cost tracking.
Board and Investor Communication
When presenting AI economics to the board:
Metrics That Matter
AI Cost as % of COGS
Shows AI's role in delivery economics
AI ROI by Use Case
Where AI creates vs. destroys value
Gross Margin Trend
With and without AI costs separated
AI Efficiency Ratio
Output value per dollar of AI spend
How to Frame the Story
- Investment narrative: "We're investing $X in AI to improve delivery efficiency by Y%"
- Margin story: "AI is improving gross margin by X points by reducing labor intensity"
- Risk management: "We have visibility and controls on AI costs to prevent margin erosion"
AI Cost Accounting Readiness Checklist
AI costs are classified correctly (COGS vs. R&D vs. SG&A)
Critical for accurate reporting
Client-level AI cost attribution is in place
Critical for accurate reporting
Monthly AI spend is tracked and reported
Critical for accurate reporting
Gross margin impact of AI is understood
AI capitalization policy is documented (if applicable)
Board/investor reporting includes AI metrics
Pricing reflects AI cost variability
Power user cost concentration is monitored
Summary
AI costs belong in COGS when they're directly tied to client delivery. They belong in R&D when building AI capabilities. They belong in SG&Awhen improving internal operations.
Getting this classification right matters because gross margin drives valuation. A 10-point margin difference can mean a 2-3x valuation difference.
Build the visibility to track AI costs at the client and use-case level. Without it, you're guessing at your true economics.
Related Resources
AI Cost Allocation Guide
The complete guide to tracking AI costs in service delivery
What Is Service Economics?
The discipline of understanding what services actually cost
Blended Delivery Economics
How DigitalCore tracks human + AI costs
Service Reality Check
Assess your service economics maturity in 5 minutes