Benchmarks · AI Startup (usage-based)
What investors check first in an AI startup
Healthy ranges for an AI startup at an early stage: gross margin 55%–85%, monthly accounts churn up to 3%, year 1 inference cost per unit vs price 0%–45%, year 2 usage growth per account / month 0%–4%.
How an AI startup makes money
Accounts x platform fee + usage units x price.
That one line is the revenue engine. Every number in it should be something you can measure today and defend with your own data. Investors rebuild it in their heads in the first five minutes, so your model should too.
The benchmark ranges
| Metric | Healthy (green) | Watch (amber) |
|---|---|---|
| Gross margin | 55%–85% | 40%–90% |
| Monthly accounts churn | up to 3% | up to 5% |
| Year 1 inference cost per unit vs price | 0%–45% | 0%–65% |
| Year 2 usage growth per account / month | 0%–4% | 0%–8% |
Year 1 inference cost per unit vs price. If compute eats more than ~half of usage revenue, margins depend on falling model prices.
Year 2 usage growth per account / month. Sustained high usage growth needs proof from your existing accounts.
Ranges are rules of thumb for early-stage companies, used by the Model Health Score in the AI Startup (usage-based) edition. They are not market data; check current sources and talk to founders who raised recently.
The drivers in the model
These are the inputs the AI Startup (usage-based) edition asks for, with the values used for the sample company, Promptly AI Inc. (AI automation for customer-support teams, priced per seat plus usage). Where a value changes over five years, the first and last are shown.
| Driver | Sample | Why it matters |
|---|---|---|
| Accounts at start (month 0) | 25 | Active accounts you already have before month 1. |
| Organic / referral new accounts in month 1 | 2 | Word of mouth, SEO, referrals, partnerships. |
| Organic new-account growth per month | 6% → 2% | Compounds monthly. Above 10% a month needs strong evidence. |
| Paid marketing spend | $10,000 → $85,000 | Ads, sponsorships, paid campaigns (people costs are in Headcount). |
| Paid CAC (cost to win one account via paid channels) | $3,500 → $4,700 | Paid new customers = paid spend / paid CAC. CAC usually rises as you scale. |
| Monthly churn | 3.5% → 2% | Share of accounts lost each month. |
| Platform fee per account per month, Year 1 | $300 | Fixed subscription part of the price. |
| Annual platform fee increase | 5% | |
| Usage per account per month at start | 200 | Use your billable unit: thousand requests, tokens, minutes... |
| Usage growth per account per month | 2.5% → 0.6% | Accounts usually use more as they automate more workflows. |
| Price per usage unit | 4 → 2.2 | Prices for AI usage often fall over time. |
| Inference / model cost per usage unit | 1.8 → 0.8 | Model API or GPU cost. Falling costs are the main margin lever. |
What the sample model looks like over five years
| Promptly AI Inc. | Y1 | Y2 | Y3 | Y4 | Y5 |
|---|---|---|---|---|---|
| Revenue | $729k | $1.94M | $4M | $6.89M | $10.4M |
| Gross margin % | 56% | 58% | 61% | 63% | 66% |
| EBITDA | -$563k | -$1.10M | -$991k | -$465k | $997k |
| Cash at year end | $1.72M | $7.65M | $6.73M | $6.26M | $7.21M |
| Paying accounts (year end) | 73 | 173 | 337 | 558 | 824 |
| ARR (year end) | $1.19M | $2.87M | $5.48M | $8.84M | $12.7M |
| Headcount (year end) | 9 | 16 | 22 | 28 | 30 |
| LTV : CAC | 4.1x | 4.0x | 4.9x | 5.9x | 6.8x |
| CAC payback (months) | 7.0 | 8.4 | 8.1 | 7.7 | 7.4 |
Fictional sample company, Base case. Figures are illustrative and come from the SAMPLE file of the AI Startup (usage-based) edition.
AI Startup (usage-based) terms to know
- Inference cost
- Cost of running the AI model for each request (API fees or GPU time).
- Usage-based pricing
- Customers pay per unit consumed (requests, tokens, minutes).
- Gross margin (AI)
- Often lower than classic SaaS because compute scales with usage.