The framework
Token is the new labour
Technology companies want to understand how AI improves their products and internal productivity. Benchmarks are scarce, and changing token costs, use cases and adoption make a single spending target hard to define.
We can still measure AI investment and compare it with business performance. Revenue growth, net revenue retention, gross revenue retention and free cash flow help measure value creation. The question is how AI contributes to them.
This framework starts with two questions:
- Is the company leveraging AI?
- Is the use of AI driving business benefits?
First, measure the investment. Then assess its contribution to business outcomes. The four ratios start with token and labour costs, alongside revenue and headcount. They can be calculated from your ledger and compared with business performance over time.
Three AI-cost buckets
Delivery token
Customer-serving tokens and inference used to deliver your product or service, including self-hosted GPU spend.
Efficiency token
Token consumed by your own teams to get work done. Its cost is an investment in internal productivity.
Training and fine-tuning
One-off R&D spend on model training and fine-tuning during the quarter.
Total AI cost is the sum of these three buckets. Exclude AI-adjacent tooling, including Pinecone, Weaviate and LangSmith, from all AI costs.
Revenue means GAAP revenue recognized in the reporting period. For a quarterly submission, use revenue realized in that quarter. Exclude annualized run-rate, bookings and contracted revenue not yet recognized.
The four ratios
These are cost-based measures of AI investment. Read them alongside business outcomes to assess productivity. A higher ratio alone does not show better performance.
AI intensity
Formula: DELIVERY TOKEN ÷ REVENUE
How much of your revenue goes to AI that serves your customers?
Customer-serving tokens and inference, including self-hosted GPU spend, as a share of GAAP revenue recognized in the same period.
Compare with: growth rate and churn rate.
AI burn
Formula: (DELIVERY + EFFICIENCY + TRAINING) ÷ REVENUE
How much of your revenue goes to AI in total?
Delivery, efficiency, and training and fine-tuning costs as a share of GAAP revenue. Excludes AI-adjacent tooling.
Compare with: revenue growth and free cash flow.
AI labour
Formula: EFFICIENCY TOKEN ÷ (EFFICIENCY TOKEN + LABOUR)
Of what you spend on your team and the AI they use, how much goes to AI?
Efficiency token cost as a share of combined efficiency token and labour costs.
Compare with: each function's own output metric.
AI maxing
Formula: ANNUALISED EFFICIENCY TOKEN COST ÷ HEADCOUNT
How much do you spend on AI per employee, per year?
Annualised internal AI spend per period-end full-time equivalent (FTE), shown in USD. Quarterly spend is multiplied by 4.
Compare with: peer teams and individuals in the same function.
Start with company-wide revenue, AI costs, labour costs and headcount. Department breakdowns and performance metrics are optional.