Pricing Strategy · 7 min read

How to Assess AI Pricing Strategy Before the Term Sheet

Why AI pricing is the most misunderstood part of any AI investment — and the framework for evaluating whether pricing power is real or narrative.

By Sasan Ghorbani · Independent AI Advisor · April 22, 2026

AI pricing is the part of an investment thesis that gets the least scrutiny and causes the most problems. Investors who spend weeks on market size and team dynamics often spend an hour on pricing — and walk away with a gross margin number that tells them very little about whether the business can actually sustain it.

This is the framework I use to assess AI pricing architecture in due diligence. It is designed to answer one question: does this company have real pricing power, or is it relying on the temporary novelty of its AI layer?

Why AI pricing is different

Traditional SaaS pricing is relatively stable. Once a company establishes a packaging model and a price point with strong NRR, the primary risk is competitive pressure and churn. The cost structure is relatively fixed.

AI pricing introduces a variable that traditional SaaS does not have: the underlying cost of intelligence. AI infrastructure costs — model API costs, compute, inference — are significant, highly variable, and falling rapidly. Companies that have not rebuilt their pricing model around current and projected infrastructure costs are carrying hidden margin risk.

The five questions that matter

1. What is the actual cost-to-serve at current scale?

Gross margin is the starting point, not the answer. The question is what drives gross margin — and whether it reflects the true cost of delivering the AI product. Many AI companies undercount infrastructure costs in their gross margin calculation, particularly compute costs bundled into engineering headcount rather than cost of revenue.

Ask for a cost-to-serve breakdown: what does it cost, in AI infrastructure terms, to serve one customer for one month? Then ask what that number looks like at 5x current usage. Companies that cannot answer this clearly have not stress-tested their unit economics.

2. Is pricing based on value delivered or cost incurred?

The strongest AI pricing architectures are built on the value delivered to the customer — time saved, decisions improved, revenue generated — rather than on the cost of the underlying AI layer. Value-based pricing is defensible against commoditisation because it does not move when model costs fall.

Cost-plus pricing is fragile. When model costs fall 60% over 18 months, competitors who pass those savings on to customers force a repricing event that erodes margins across the market.

3. What happens to the pricing model when the AI layer commoditises?

This is the stress test question. Ask the company directly: if your primary model API cost dropped by 50% tomorrow, what would change in your pricing? If the answer is 'nothing, because customers pay for the outcome, not the infrastructure,' that is a strong signal. If the answer is unclear or defensive, the pricing architecture has not been stress-tested.

4. Is there a proprietary cost advantage that is widening over time?

The most durable AI pricing positions come from companies building a cost advantage that compounds — through proprietary fine-tuning, through data network effects that reduce inference costs, or through workflow integrations so deep that the switching cost justifies a price premium. Ask what the company is doing today that will make their cost structure meaningfully better in 24 months.

5. What does net revenue retention tell you about pricing power?

NRR is the best proxy for real pricing power in the due diligence data room. Companies with NRR above 115% are expanding revenue from existing customers. Companies with NRR below 100% are losing ground on existing accounts, which is difficult to sustain regardless of new logo growth.

Look at NRR by cohort, not in aggregate. A single large customer expanding rapidly can mask a general pattern of flat or declining usage across the customer base.

The packaging question

Separate from unit economics, the packaging architecture matters. How the product is tiered — by seats, by usage, by outcomes, by feature set — determines how revenue scales with customer growth and how exposed the company is to usage-based margin variability.

Usage-based pricing is commercially intuitive for AI products but creates a specific risk: customers who use the product heavily generate more revenue, but also more cost. Companies that have not modelled the relationship between usage volume and margin at different customer sizes may discover that their largest customers are their least profitable.

What good AI pricing looks like

The strongest AI pricing architectures share three characteristics: they are built on value delivered rather than cost incurred, they create switching costs that protect against commoditisation pressure, and they scale with customer success rather than with infrastructure cost. Companies that can demonstrate all three — with NRR and cohort data to support the claim — have a genuine pricing asset. Companies that cannot are carrying more risk than their gross margin suggests.