Forecasting for Usage-Based AI Companies

Usage-based pricing didn’t appear as a trend. It showed up as a response to how these businesses actually operate.

As AI-driven products moved from experimentation to production, the subscription models that once worked for SaaS began to fracture. Revenue became tied to tokens, minutes, inference calls, credits drawn down, and compute consumed. Growth accelerated, but predictability suffered.

For AI companies operating on consumption-based models, the challenge isn’t just pricing. It’s forecasting revenue, margin, and growth when customer behavior, not contract value, drives outcomes.

From working with companies navigating this shift, one thing becomes clear quickly: the teams that forecast well aren’t chasing AI narratives. They’re focused on where usage risk actually shows up and they talk about it plainly.

The Real Problem With Forecasting Usage-Based AI Revenue

Most AI companies don’t struggle because they lack data. They struggle because their forecasting models were built for a world where revenue was static.

Usage-based AI breaks those assumptions:

• Revenue scales with customer behavior, not just bookings
• Margin fluctuates with infrastructure usage
• Commitments and overages introduce timing risk
• Product adoption directly impacts financial outcomes

Yet many teams still forecast using static Salesforce fields, spreadsheets, or manually adjusted rollups. The result is forecasts that look confident but fail under real-world usage volatility.

Where Usage Risk Actually Lives

One of the biggest mistakes we see is targeting or forecasting “AI companies” as a category. The better lens is understanding where usage risk shows up inside the business.

For example:

• Inference-heavy platforms face unpredictable cost and revenue swings as usage spikes
• API-first AI products struggle to forecast commit drawdown and expansion timing
• Embedded AI features introduce adoption-driven revenue variability that finance teams can’t see early enough

Forecasting accuracy improves dramatically when teams model these realities instead of flattening them into average growth rates.

Billing Maturity Tells You More Than ARR

Usage-based AI companies evolve quickly, and their billing maturity often outpaces their forecasting infrastructure.

We commonly see:

• Early-stage teams using flat commits and spreadsheets to manage usage
• Mid-stage teams introducing hybrid pricing with manual Salesforce adjustments
• Enterprise-scale companies managing prepaid credits, multi-year commitments, and renewal risk across systems

Each stage introduces more complexity, not less. Without a forecasting model designed for consumption, confidence erodes just as the stakes increase.

Forecasting Revenue Without Understanding Cost Is a Trap

For AI companies, revenue and cost are inseparable.

Cloud spend often scales directly with customer usage. When forecasts ignore usage patterns, they also ignore margin exposure. This leads to surprises that show up too late: at close, at renewal, or in front of the board.

Usage-aware forecasting changes the conversation from “Did we hit the number?” to “Which customers are driving profitable growth, and which ones introduce risk?”

Internal Friction Is the Strongest Signal

Some of the clearest indicators that forecasting is broken aren’t found in dashboards. They show up in process pain:

• Finance rebuilding forecasts every month
• Sales committing usage without visibility into consumption
• RevOps stitching product data into Salesforce manually
• Leadership questioning forecast confidence instead of debating strategy

These aren’t operational inconveniences. They are signals that usage has outgrown the forecasting model.

Salesforce Is Still the System of Record, Just Not the System of Truth

Most usage-based AI companies run on Salesforce. The problem is that usage data lives elsewhere.

Product data sits in Snowflake or BigQuery. Forecasting happens in spreadsheets or BI tools. Salesforce becomes a static snapshot of a dynamic business.

The opportunity isn’t to replace Salesforce. It’s to bring usage-driven forecasting into it, so teams can operate from a shared, trusted view of the future.

What Better Forecasting Looks Like for Usage-Based AI

The goal isn’t perfect prediction. It’s informed confidence.

That means:

• Forecasts that respond to real usage signals
• Visibility into both revenue and margin impact
• Alignment between Sales, Finance, and Product
• Scenarios that reflect how customers actually consume

Usage-based pricing isn’t going away. Companies that continue to treat forecasting as a static reporting exercise will keep feeling the strain. Those that model around consumption tend to make decisions with more confidence.

This is the problem space revVana focuses on: helping companies forecast what actually drives their business.

When usage drives revenue, forecasting has to follow usage.