The AI Startup CFO Playbook: Token COGS, R&D Capitalization, and the New SaaS Metrics
Written by Harry Prabandham
Curated by Rubric Financial
1 / 6
Why AI Companies Need a Different Finance Setup
- SaaS gross margins are 70-85% because the marginal cost to serve a customer is near zero. AI companies have REAL COGS: inference compute scales linearly with usage.
- Implication: 'flat-tier pricing' breaks at AI companies. A $99/mo tier that includes unlimited usage becomes loss-making when one customer hits 100x normal usage.
- Margins range wildly: pure inference resellers (OpenAI wrappers) often run 20-40% gross margin. Companies with proprietary models can hit 60-75%. Edge inference + smaller models can approach SaaS-like 80%+.
- Most early-stage AI startup financial models DON'T model token costs properly; they apply a flat 'COGS %' assumption that breaks at scale. Get this right early.
Related Resources
What Belongs in SaaS COGS and How to Compute Gross Margin
A practical guide to defining SaaS cost of revenue and calculating gross margin correctly, including AI inference costs and benchmarks.
CFO & StrategyB2B Surcharging: Passing Credit-Card Fees to Customers
How SaaS companies can recover credit-card interchange costs on subscriptions, and the legal and commercial tradeoffs of doing so.
CFO & StrategyBeating Funded Competitors by Getting Smaller, Not Bigger
Why cutting features and narrowing your market wins against overfunded competitors trying to serve everyone, and how focus becomes your unfair advantage.
About the author
Harry PrabandhamFounder & CEO
Founder and CEO of StartupCFO. MBA from Wharton, MS in Computer Science, and decades of experience building and advising venture-backed startups.
More articles by Harry →Related tools and reading
Startup Metric Tree
Every KPI by industry, from north star and runway down to the drivers, with formulas and benchmarks.
GuideWhat Belongs in SaaS COGS and How to Compute Gross Margin
A practical guide to defining SaaS cost of revenue and calculating gross margin correctly, including AI inference costs and benchmarks.
InsightWhat Is a Metric Tree? How One Number Breaks Down Into the Levers That Move It
A KPI list can tell you a number missed. A metric tree tells you why: one outcome at the top, broken into the metrics that produce it, down to the levers you can pull.
GlossaryARR (Annual Recurring Revenue)
Annualized value of your subscription revenue at a point in time.
InsightStartup CFO Digest: Week 41, October 2026
This week's funding and valuation data shows active capital deployment tempered by selectivity, while foundational questions about moats, liquidity timelines, and board governance demand founder attention. AI valuations remain frothy, but the underlying shifts toward workflow integration and long-term liquidity structure signal where returns will actually come from.
GlossaryMRR (Monthly Recurring Revenue)
Monthly equivalent of ARR, useful for month-over-month tracking.
Want this run on your actual numbers?
A fractional CFO can turn what you just read into a board pack, a forecast, and a spending plan built from your own ledger.
Want the full sample as a PDF?
No spam, ever. If the download doesn't start, email us.
Or talk it through:
