AI Pricing Models: A Worked Revenue Model and Free Template (2026)

Most AI founders build their first model from a SaaS template. It breaks in two places. Revenue no longer follows seats; it follows usage. And every conversation your AI handles costs you money in model fees. Below, we build the model of a fictional AI support-agent startup, step by step. We compare AI pricing models, build revenue from usage and put inference costs where investors expect them. Then we show gross margin by year and how much the company needs to raise. The Excel file is free.

An AI startup financial model is a forecast that builds revenue from customers, usage and price. It counts the AI's running costs, like model fees for every conversation, as cost of revenue. It then shows gross margin year by year and how much cash the company needs before it pays for itself.

Which AI pricing model should you use: per seat, usage or outcome?

AI pricing models charge for one of four things: a seat, each use, each result, or a flat fee plus results. If your AI does the work itself, avoid pure seat pricing. The better the AI gets, the fewer seats your customer needs. Usage and outcome pricing grow with the value you deliver.

• Per seat — a monthly fee for each person on the customer's team. Say $600 per support agent.
• Per use — a price for each conversation, document or API call. Say $0.45 per conversation.
• Per outcome — a price only when the AI gets a result. Say $0.85 per resolved conversation.
• Hybrid — a flat monthly fee plus a usage or outcome price. Say $800 a month plus $0.60 per resolution.

Here's what three AI support agents charged in October 2026. Look at the unit each one sells.

Most AI sellers have landed on a mix. More than half now charge a flat fee plus usage, in a 2026 survey by Salesforce Ventures. Outcome pricing is spreading too. ICONIQ found it at 23% of AI builders in 2026.

Same customer, four prices

Here's why the unit matters. Take one customer of our fictional startup, with 6,000 support conversations a month. We set each price so this customer pays about $2,600 a month today, while the AI resolves half the conversations. Then the AI gets better and resolves 70%.

Seat revenue falls by 40%, because the customer needs fewer human agents. Per-conversation revenue stays flat. Per-resolution and hybrid revenue grow with the AI's success. Meanwhile the AI bill doesn't move. So with seat-based pricing, every product win shrinks your revenue.

Run the whole company on each price, and the gap grows over five years:

Per resolution wins on paper. But then your revenue rides entirely on how often the AI succeeds, and buyers find those bills hard to budget. The hybrid gives up some upside for a floor: the platform fee. That's why we use it as the base case. To show your pricing in the deck, see our guide to the business model slide.

AI startup financial model template (free Excel download)

Our free template is the model behind this post. It has three sheets: inputs, the model and a checklist. The model runs 20 quarters, with a summary by year and the funding need. Change any blue input, or switch the pricing model, and every number updates. We've built 500+ financial models; this one keeps what an AI company needs.

Here are the main inputs in the example. Swap in your own:

How do you build revenue from customers and usage?

Multiply customers by usage by price. Customers come from sales capacity: deals per salesperson, minus the ones who leave. Usage grows as customers roll the AI out to more teams. Then apply the price to the unit you sell: seats, conversations or resolutions. Revenue follows usage, so model it per customer, not per user.

Say a new customer sends 6,000 support chats a month. In year one, the AI resolves half of them. On the hybrid price, the customer pays the $800 fee plus 3,000 resolutions at $0.60. That's $2,600 a month. Next quarter its volume grows, and so does the bill.

New customers come from people, not a growth rate. In the example, the founders close 3 deals a quarter. Each salesperson adds 5 more, from the quarter after they join. Our guides to the hiring plan and the customer acquisition model show how to build both.

Usage growth matters as much as new customers. Here, each customer's volume grows 5% a quarter as they add channels and teams. That growth more than makes up for the customers who leave. It's what net revenue retention measures, and investors check it early.

Where do inference costs go, and what gross margin should you expect?

Inference costs, the model fees you pay each time the AI answers, go in cost of revenue. So do search, tool calls, hosting and the team that onboards customers. That's why AI gross margins start below software margins. Investors accept a low start if you can show why the margin rises every year.

Start with what one conversation costs. At every step, the AI rereads help articles, order data and the chat so far. So the tokens it reads add up fast. Count them, add the tokens it writes, and price both at your model's rates:

That cost won't stay put. The price of a given level of AI has fallen fast. Epoch AI measured drops of 9x to 900x a year, using data to early 2025. Your bill falls more slowly, because you'll keep moving to better models. So the template assumes 15% a year.

Two levers cut it further. Route easy questions to a small model: Anthropic's Haiku 5.5 reads tokens at a twentieth of Sonnet 5.5's price. And cache the help articles the AI rereads in every chat. Watch for agent loops, though. One builder in ICONIQ's 2026 report budgeted ten cents a run and saw it pass $1.50 as agents retried.

Here's what the template counts as cost of revenue:

• AI cost — model fees, search and tool calls for every conversation.
• Hosting — cloud, monitoring and data, $100 per customer a month.
• Onboarding team — the people who set up and support customers.

Compute you use to build and test the product stays in R&D, below gross profit. Our guide to the cost of revenue covers the other lines.

Year one looks weak, at 25%. Each conversation still costs $0.20, and the onboarding team is big next to early revenue. By year five, the AI cost per conversation has halved, while prices have dropped by about a fifth. The team also grows more slowly than revenue. So the margin climbs to 67%.

Our startup starts well below the average and ends above it. If yours does too, be ready to explain both ends.

We saw this line go wrong in a real raise. An AI adtech startup brought us its model for a $6M seed round. Its LLM, storage and cloud costs were missing, so its gross margin looked better than it really was.

How much funding does an AI startup need?

Add up the cash you burn until the company pays for itself, then add a buffer. That's your funding need. In our example, the startup burns about $10.8M before its first profitable quarter, early in year five. With a $1.5M buffer, it needs about $12M: a $4M seed now and an $8M Series A.

Timing matters as much as size. Without the Series A, the seed runs out at the end of year two. So the plan closes the A in year two's third quarter, after cash dips to about $1.3M. The model shows that date long before an investor asks.

Then test the bad cases. Change one input at a time and watch the cash burned before profit.

The AI cost per conversation is the input to watch. Double it, and the company burns twice the cash and still isn't profitable by year five. A weaker AI hurts almost as much, because hybrid revenue depends on resolutions. Our capital efficiency guide shows how investors judge that burn. And our dilution benchmarks show what each round costs you in ownership.

Is your AI financial model ready for investors?

Check five things. Revenue comes from customers, usage and price. AI costs for serving customers sit in cost of revenue. Gross margin rises every year, for reasons you can explain. You've run the bad cases, like a doubled AI cost. And year five is big enough to matter to a venture fund.

Want a second pair of eyes? We build and review AI companies' models as part of our financial modeling services. For more worked models, see our financial model examples. Raising soon? Read how to raise money for an AI startup and see the VCs investing in AI.