How is unit economics evolving?
№ 017 · Gross margin trend
Definition
MetricGross margin, subscription revenue and blended, over time
Unit% of revenue
Line of gross margin % over time. Watch for slow erosion from cost-of-revenue creep. Build by taking revenue minus cost of revenue (hosting, support, payment processing) divided by revenue per period. Example: 78% Q1, 76% Q2, 74% Q3 (drifting down 4 points in 9 months means investigate).
Benchmarks
| Bottom 30% | Median | Top 30% |
|---|---|---|
| Under 70% subscription margin, or blended margin decaying quarter over quarter | ~79% subscription gross margin (stable for years); public SaaS blended ~76% | 82%+ subscription gross margin |
Private B2B SaaS subscription gross margin, 2022-2025; blended margin runs lower where professional services are material; AI-native products run structurally lower on inference costs · Subscription gross margin has been one of the most stable SaaS benchmarks (79% median in both 2022 and 2023 per Benchmarkit). The new dispersion is AI: Bessemer reports AI application gross margins of 60-65% and top AI leaders as low as ~25% (per Accel data), so an AI-heavy roadmap can legitimately compress margins; separate AI COGS in the trend to see it.
| Category | Bottom 30% | Median | Top 30% |
|---|---|---|---|
| Traditional SaaS vs AI-native | traditional under 70%; AI leaders reported as low as ~25% | traditional: ~79%; AI application: 60-65% | traditional subscription: 82%+; AI application: 65%+ |
When it looks bad
A slow downward glide of 1-2 points per quarter with no single step change, the signature of infrastructure or inference costs growing faster than revenue rather than a one-off event.
Subscription margin 80%, 78.5%, 77%, 75% over four quarters while revenue grows 5% per quarter. At that slope the company loses 8 points in two years and the LTV math underneath every CAC decision is quietly wrong.
What to do about it
- Split COGS into hosting, third-party inference or data, support, and services delivery, and trend each as % of revenue; a blended line cannot tell an AWS commitment problem from an LLM-cost problem, and the fixes are different.
- Reprice where AI features raised COGS: High Alpha and OpenView data shows most companies see AI monetization headroom; move AI-heavy capabilities into a metered or premium tier rather than absorbing inference in a flat price.
- Negotiate committed-use discounts and re-architect the top 2 cost drivers once a year; at $5M+ ARR, 3-5 points of hosting margin from reserved capacity and egress hygiene is common and flows straight to every unit-economic metric.
- Keep services gross margin honest and separate: bundling low-margin implementation into subscription flatters neither line and corrupts the payback and LTV inputs downstream.
Sources
- Benchmarkit Software subscription gross margin 79% median in 2023, unchanged from 2022; among the most stable SaaS metrics benchmarkit.ai ↗
- Benchmarkit Tracks total revenue gross margin, subscription gross margin, and services gross margin separately by segment benchmarkit.ai ↗
- SaaSletter (reporting Accel Globalscape and Bessemer data) Public SaaS gross margin 76% (Accel); AI application leaders at ~25% simple average, vs 60-65% AI gross margins in Bessemer reporting saasletter.com ↗
- High Alpha Benchmarks software gross margin (subscription revenue less COGS over subscription revenue) by quartile as of Q2 2025 2994607.fs1.hubspotusercontent-na1.net ↗
- High Alpha and OpenView AI compute costs pressuring gross margins; companies advised to reprice to offset rising COGS, with 63%+ seeing AI monetization opportunity highalpha.com ↗