TL;DR: Usage-based pricing charges customers based on consumption rather than fixed monthly fees. This creates revenue that fluctuates with customer behavior instead of growing predictably like subscription revenue. Modeling it requires understanding customer usage patterns, growth curves, and seasonality rather than just logo counts and pricing tiers. Build your forecast from customer-level usage projections aggregated up, not top-down trend lines. The volatility is real but manageable if you track leading indicators and segment customers by usage maturity. Companies that master usage-based modeling can forecast revenue within 10% accuracy despite apparent unpredictability.
Most SaaS revenue forecasting is straightforward. You know how many customers you have, what they’re paying monthly, and what churn looks like. New customers add revenue, churn removes revenue, expansion adds more. The math is clean.
Usage-based pricing breaks this model. A customer who consumed $5,000 in services this month might consume $8,000 next month or $3,000. Their “subscription value” isn’t fixed. Revenue per customer varies based on their behavior, seasonality, and growth.
This makes forecasting harder but not impossible. You just need different inputs:
Traditional SaaS forecast: Customer count × average MRR × retention rate
Usage-based SaaS forecast: Customer count × average usage per customer × price per unit × usage growth rate × retention rate
The core difference is you’re forecasting customer behavior (usage) not just customer count. This requires understanding what drives usage in your specific business.
Start by defining your usage metric clearly:
API calls: Customers pay per thousand API calls
Compute hours: Customers pay per hour of processing time
Data processed: Customers pay per GB of data analyzed
Transactions: Customers pay per transaction processed
Active users: Customers pay based on monthly active users
Storage: Customers pay per GB of storage used
Your pricing is usually tiered (volume discounts at higher usage levels):
$0.10 per unit for 0-10,000 units
$0.08 per unit for 10,001-100,000 units
$0.06 per unit for 100,000+ units
Calculate revenue per customer by tracking their usage and applying your pricing:
Customer A: 5,000 units × $0.10 = $500
Customer B: 50,000 units = (10,000 × $0.10) + (40,000 × $0.08) = $4,200
Customer C: 200,000 units = (10,000 × $0.10) + (90,000 × $0.08) + (100,000 × $0.06) = $14,200
Understanding your usage distribution across customers is critical. Many usage-based SaaS companies have extreme concentration where top 10% of customers generate 60-70% of revenue.
Customers don’t consume at constant rates. They follow predictable usage growth curves:
Month 1-3 (Onboarding): Low usage as they’re implementing and testing. Might be 20-40% of steady-state usage.
Month 4-6 (Ramp): Usage accelerates as they roll out to more users or use cases. Often grows 30-50% monthly during this phase.
Month 7-12 (Growth): Continued expansion but at slower pace. Maybe 10-20% monthly growth.
Month 12+ (Maturity): Usage stabilizes or grows with customer’s underlying business growth. Typically 5-10% quarterly growth.
These curves differ by customer segment:
SMB customers might plateau quickly at $500-1,000 monthly usage.
Mid-market customers might plateau at $3,000-8,000 monthly.
Enterprise customers might never plateau, continuing to grow as they expand usage across divisions.
Build usage curves from historical data. Take customers who signed 18 months ago and track their usage month by month. This reveals the typical growth pattern you can apply to forecasting new customers.
We worked with a usage-based API company and found that average customer usage followed this pattern: $200 in month 1, $450 in month 3, $800 in month 6, $1,100 in month 9, $1,250 in month 12, then 3% monthly growth afterward. Knowing this pattern let us forecast new customer contribution accurately.
Model usage-based revenue by cohort just like subscription revenue, but add usage layer:
January 2024 cohort: 20 customers signed
– Month 1 revenue: 20 customers × $200 average usage = $4,000
– Month 3 revenue: 19 customers × $450 average usage = $8,550 (one churned)
– Month 6 revenue: 18 customers × $800 average usage = $14,400
– Month 12 revenue: 17 customers × $1,250 average usage = $21,250
Track each cohort through their lifecycle applying your usage curves and churn rates. Sum across all cohorts to get total revenue.
This approach accounts for:
– New customers starting at low usage and ramping
– Existing customers maturing and growing usage
– Churned customers removing revenue
– Usage volatility within expected ranges
The forecast is more complex than subscription models but much more accurate than top-down trend extrapolation.
Not all customers follow the same usage curve. Segment by meaningful attributes:
By use case: Customers using you for primary workflow have different usage than customers using you for secondary use case.
By customer growth stage: Fast-growing startup customers increase usage rapidly. Mature enterprise customers grow usage slowly.
By pricing tier: Customers on volume pricing use more but pay less per unit. Customers on basic pricing use less but pay more per unit.
By industry: Some industries have seasonal usage (retail spikes in Q4, tax software spikes in Q1).
Build separate usage curves for each segment. Forecast each segment independently then sum them up.
We helped a data processing company segment customers into three buckets:
– High frequency (daily usage): $3,000 average monthly, 5% churn, 8% monthly usage growth
– Medium frequency (weekly usage): $800 average monthly, 8% churn, 4% monthly usage growth
– Low frequency (occasional usage): $150 average monthly, 15% churn, 1% monthly usage growth
Modeling these separately produced revenue forecast that was 12% more accurate than treating all customers as identical.
Usage-based revenue is more volatile month-to-month than subscriptions. Customers increase or decrease consumption based on their business needs.
Expect 15-25% variance in monthly revenue from individual customers. This is normal, not a problem.
Reduce forecast volatility by:
Aggregating across many customers: 500 customers each with 20% variance produce more predictable total than 50 customers with same variance.
Looking at trends over quarters not months: Monthly volatility averages out over quarters. Focus quarterly forecasts on rolling 90-day averages not individual months.
Identifying seasonal patterns: Many usage businesses have quarterly or annual seasonality. Model it explicitly based on historical data.
Leading indicators: Track metrics that predict usage before it happens. If customers are uploading more data, usage will spike next month. If customer active users are declining, usage will drop.
Minimum commitments: Some usage-based companies add minimum monthly commitments ($500/month minimum) which creates a revenue floor and reduces volatility.
We worked with a customer whose monthly revenue varied from $180K to $280K seemingly randomly. After analyzing, we found strong quarterly seasonality (Q4 and Q1 were high, Q2 and Q3 lower) and week-of-month effects (end-of-month spikes). Modeling these patterns reduced forecast error from 25% to 8%.
Usage-based revenue recognition is simpler than subscription in one way: you recognize revenue as usage happens, not ratably over time.
Customer uses $5,000 worth of services in January, you recognize $5,000 revenue in January. No deferred revenue because you’re recognizing as you deliver.
But two complications:
Billing timing: Usage typically gets measured throughout the month and billed in arrears. Usage from January 1-31 gets billed February 5 after you’ve tallied consumption. You recognize revenue in January (when delivered) but don’t collect cash until February.
Minimum commitments: If customers commit to $10,000 quarterly minimum, you might recognize some revenue ratably. Customer uses $8,000 in Q1 but committed to $10,000. You recognize the full $10,000 (either ratable over quarter or on final day when commitment obligation is clear).
Most usage-based companies find revenue recognition simpler than subscription businesses but cash flow management more complex due to arrears billing.
Here’s the model structure:
Customer acquisition forecast: How many customers will you add each month?
Initial usage profile: What’s the average usage for new customers in first 3-6 months?
Usage growth curves: How does usage grow over customer lifecycle?
Churn rates: What percentage of customers churn monthly?
Pricing tiers: What do you charge per unit at different volume levels?
Apply these inputs month by month:
Month 1: Start with existing customers and their current usage. Add new customers at initial usage levels. Apply usage growth to existing customers based on tenure. Subtract churned customers.
Month 2: Take month 1 ending customers. Add new customers. Apply usage growth curves. Subtract churn.
Continue forward.
The model should show:
– Customer count by cohort
– Average usage per customer by cohort
– Revenue per customer by cohort
– Total revenue
– Revenue by customer segment
We build models that project 24-36 months forward with monthly granularity. The outer months get less reliable but even directional guidance is useful for planning.
Usage-based revenue is more predictable when you track leading indicators:
User onboarding: New users being added to customer accounts predicts usage growth 30-60 days later.
Data ingestion: Customers uploading more data will consume more processing, predicting revenue in 15-30 days.
API authentication requests: Customers setting up new integrations predicts usage growth in 60-90 days.
Active users: Growing active user count typically predicts growing usage with 30-day lag.
Customer hiring: Enterprise customers announcing team expansion predicts usage increase as they scale.
Track these monthly for all customers. When leading indicators trend up for a customer segment, raise your usage growth assumptions for that segment in your forecast.
We helped a compute-platform company build a model where 75% of revenue variance was explained by user onboarding 45 days prior. This let them forecast revenue with 11% accuracy 2 months in advance.
Some situations make usage-based forecasting particularly challenging:
Early-stage with limited history: You need 12-18 months of customer data to build reliable usage curves. Before that, you’re guessing.
Seasonally-sensitive businesses: Usage patterns that change dramatically by quarter or month require longer history to model accurately.
High customer concentration: If top 10 customers are 70% of revenue, their individual behavior dominates. You can’t rely on averages.
Emerging use cases: When customers start using your product in new ways, historical patterns stop predicting future usage.
Major product changes: If you significantly change pricing or add features that change consumption patterns, historical curves become less predictive.
In these situations, build multiple scenarios (base, upside, downside) and update forecasts monthly as you get new data. Accept higher uncertainty and plan accordingly.
—
Q: Is usage-based revenue more volatile than subscription revenue?
Yes, individual customer revenue is 2-3x more volatile month-to-month with usage-based pricing. But this volatility decreases as you aggregate across customers and over time. 10 usage-based customers have unpredictable monthly revenue. 200 customers have pretty predictable quarterly revenue. The volatility also works both directions: upside surprises are as common as downside surprises. Model conservatively and you’ll often beat the forecast, which is better than modeling aggressively and missing.
Q: Should we add minimum commitments to reduce revenue volatility?
It depends on your goals. Minimum commitments create a revenue floor which helps with forecasting and cash flow. But they can reduce customer acquisition (creates barrier to entry) and reduce expansion (customer already met minimum so why increase usage?). We typically recommend minimum commitments for enterprise customers where contracts are negotiated anyway, but pure usage-based pricing for self-serve or SMB customers where you want low friction. You can also use hybrid: first 6 months pure usage to prove value, then minimum commitment at renewal.
Q: How do we forecast usage-based revenue when we don’t have much historical data yet?
Start with cohort analysis even if you only have 3-6 months of data. Track how early cohorts are behaving and build initial usage curves from that limited data. Acknowledge the uncertainty by building multiple scenarios (conservative, moderate, aggressive usage growth). Update your forecast monthly as you get more data. Look for analogous businesses (competitors, similar products) and use their publicly disclosed consumption patterns as benchmarks. The key is updating quickly as new data arrives rather than committing to a forecast built on insufficient data.