Financial models for AI and SaaS companies

Software revenue is a cohort problem, not a growth-rate problem. A model that applies an annual growth percentage to last year's revenue cannot tell you whether growth is coming from new logos, expansion within existing accounts, or a price increase that will not repeat. Investors ask that question in the first meeting, and a model that cannot answer it makes the rest of the conversation harder.

Every SaaS engagement starts from a monthly cohort build: customers acquired in each month, their starting contract value, their expansion curve, and their survival curve. Net revenue retention then falls out of the model as a calculated result rather than being typed in as an assumption — which is also how a diligence team will try to recompute it.

Gross margin after inference is the 2026 question

AI-native companies carry a cost structure that classic SaaS did not. GPU time, model API calls, vector storage and retrieval, and the human review layer that most products still need are all variable with usage — and they scale with the customers who use the product most, who are frequently the customers on the flattest pricing.

Modelling inference as a flat percentage of revenue hides exactly the risk investors are now pricing. The right structure costs it per active user or per unit of consumption, against the token or compute assumptions your engineering team actually observes, then shows gross margin improving as model costs fall and caching improves — with that improvement stated as an assumption someone owns.

Efficiency metrics carry more weight than growth

Burn multiple — net burn divided by net new ARR — has become the fastest summary of whether a company deserves the next round. Below 1.5x reads as efficient; above 3x invites a difficult conversation regardless of the growth rate above it. The model calculates it monthly and by cohort so you know which quarter moved it.

CAC payback is built by channel and segment, using fully loaded acquisition cost — sales salaries and commission, marketing spend, and the portion of customer success genuinely spent on onboarding. A blended payback number across self-serve and enterprise motions is a number nobody can act on.

Churn is modelled as logo churn and dollar churn separately, by segment. A blended figure that looks healthy because enterprise dollars mask SMB attrition is one of the most common findings in a model audit of a SaaS workbook.

The headcount plan behind the opex

Software companies are mostly people. Opex is built from a role-level hiring plan — title, start month, fully loaded cost, and the ramp period before a quota-carrying hire produces anything. Sales capacity then drives the new-logo line, which closes the loop: you cannot grow bookings in the model without hiring the people to sell them.

What this supports

The same core model serves the raise, the board pack and the valuation. For a priced round it feeds a DCF and comparable set; for the process itself it extends into a fundraising model with cap table, dilution and an investor returns view.