AI startups increasingly sell contracts that mix subscriptions, committed spend and usage-based pricing. That makes “ARR” less straightforward than it is in a simple seat-based SaaS business. A customer may sign a $120,000 annual commitment but consume only $40,000 of service in the first six months. Another customer may have no large contract but generate rapidly growing monthly usage. Both relationships matter, but they tell different stories about revenue quality.

Start by defining contracted ARR precisely

Contracted ARR should reflect recurring value that the customer is obligated to pay under the current agreement. It should not automatically include optional usage, unsigned expansion or hoped-for renewals.

If a contract has minimum spend plus variable overage, separate the committed portion from the forecast overage. This keeps the metric reproducible across customers.

Usage revenue reveals realized demand

Actual consumption shows whether customers are using the product strongly enough to support the contract. A large commitment with very low usage can signal implementation delays, poor activation or a contract that may be difficult to renew.

High usage above the commitment can signal expansion opportunity, but it can also create margin pressure if serving cost grows faster than price.

Track committed, consumed and billed separately

These three numbers can diverge. A customer can be committed to $10,000 per month, consume $6,000 of service and still be billed $10,000 under a minimum agreement.

Management reporting should show the differences so revenue strength is not confused with product adoption.

Renewal risk often appears in the consumption gap

If a customer consistently uses far less than its contracted allowance, renewal negotiations may focus on downsizing. Usage-to-commitment ratio can therefore be an early health signal.

Compare the ratio by cohort and contract type rather than using one threshold for every product.

Over-consumption creates a different risk

Customers that exceed commitments may look attractive, but only if the overage economics are healthy. Heavy video, voice or agent usage can produce large compute bills.

Expansion analysis should include contribution margin, not only additional revenue.

Prepaid credits complicate the picture

A customer may purchase credits upfront and consume them slowly. Cash arrives early, but revenue and serving cost may occur later.

Outstanding credit balances should be shown beside contracted ARR so the company understands future delivery obligations.

Bookings are not ARR

A multi-year deal with a large total contract value can produce impressive bookings without creating equivalent annual recurring revenue. Report TCV, annualized commitment and recognized revenue separately.

This reduces the risk of mixing sales momentum with current economic scale.

Usage trends help forecast expansion

Customers approaching their commitment ceiling can be strong candidates for plan upgrades. Track usage velocity and projected exhaustion rather than waiting until the account suddenly incurs large overage.

Customer success can use the same signal to start a commercial conversation early.

Low usage may still be strategic

Some enterprise contracts ramp slowly because deployment, security review or internal adoption takes months. A temporary consumption gap is not automatically churn risk.

The account plan should distinguish expected ramp from unexpected inactivity.

Connect revenue quality to infrastructure planning

Contracted ARR helps forecast demand only when it is combined with expected usage patterns. Infrastructure teams need estimated tokens, images, video minutes or agent runs, not just dollars.

Finance and engineering should therefore share one consumption forecast.

Board reporting should show both sides

A useful board view can include contracted ARR, recognized recurring revenue, usage-to-commitment ratio, expansion usage, gross margin and major renewal dates.

This tells a much richer story than one ARR headline.

Use cohorts to see whether contracts mature well

Compare customers by first contract month and follow usage, expansion and renewal over time. Healthy cohorts should generally show product adoption catching up with contracted value.

Our article on AI subscription cohort economics provides a framework for combining revenue and serving cost.

Add a renewal forecast before the contract is near expiry

Six months before renewal, estimate whether each account is likely to renew at the same commitment, expand or contract. Base the forecast on product usage, active seats, support activity and executive sponsorship rather than sales optimism alone.

This lets finance distinguish healthy contracted ARR from revenue that is technically committed today but already showing signs of future compression.

The gap is a management signal, not an accounting trick

Contracted ARR tells you what customers have promised. Usage tells you what they are actually doing. AI startups need both because future renewal, margin and infrastructure demand depend on whether those two numbers converge in a healthy way.