A founder came to us last spring with a pitch deck that looked ready for a Series A. The slides were clean, the market size slide had a big number on it, and the traction chart went up and to the right. When we asked for the underlying model, he sent over a spreadsheet with one tab. It had last month’s actuals and a growth rate typed into a formula. No cash view, no headcount plan, nothing that showed what happened if a big customer churned in month four. He wasn’t being careless.
He’d just never been shown what a real model looks like, because most founders haven’t.
Five models used to cover it. In 2026, most companies need a sixth, because a growing share of them have AI costs sitting inside their COGS that nobody is tracking properly.
1. The revenue model
This is the one everyone builds first, and it’s usually the weakest. A revenue model isn’t a growth rate typed into a cell. It’s a build-up of new customers, expansion revenue, and churn, broken out by month, ideally by cohort. If your revenue model can’t tell you how much of next quarter’s number comes from customers you already have versus deals you haven’t closed yet, it isn’t a model. It’s a guess with a chart attached.
2. The cash flow and burn model
Revenue and cash are not the same thing, and startups learn this the hard way when a big invoice goes out net-60 and payroll is due on the 1st. A cash flow model tracks what’s actually coming in and going out of the bank account, month by month, and it’s the thing that tells you your real runway instead of the runway you’d like to believe. One client had 14 months of runway on paper and 9 in reality, because the model didn’t account for a lump payment to a contractor that was already committed.
3. The headcount plan
Headcount is usually the single biggest expense line, and it’s the one founders plan the loosest. A real headcount plan ties every hire to a role, a start date, a fully loaded cost including benefits and taxes, and a reason the hire needs to happen when it does. When we build these with clients, the plan often shrinks. Not because the team was wrong about needing the people, but because seeing the true cost laid out changes the order they hire in.
4. The unit economics model
CAC, LTV, and payback period get thrown around in every pitch deck, but the model behind them is where the truth lives. A unit economics model shows what it actually costs to acquire a customer, what that customer is worth over time, and how long it takes to earn back the acquisition cost. We had a client convinced their paid channel was working because CAC looked fine in aggregate. Once we split it by channel, one source was quietly underwater and dragging the blended number down with it.
5. The scenario model
Every model above should exist in three versions: a base case, a case where things go better than expected, and a case where they don’t. A scenario model isn’t about predicting the future. It’s about knowing, before a board meeting or a renewal season, what your runway looks like if a key assumption breaks. Founders who build this early tend to make calmer decisions when something actually does go wrong.
6. The AI and compute cost model
This is the one most 2026 models are missing entirely. If your product uses an LLM API, runs inference on GPUs, or bakes any usage-based AI cost into the product, that cost behaves nothing like traditional hosting. One SaaS company we worked with launched an AI copilot feature in Q1. Adoption was strong. Their model API bill went from about $6,000 a month to $41,000 in a single quarter, and it showed up buried under a general “infrastructure” line where nobody caught it until the gross margin numbers came in soft. An AI cost model tracks usage per customer, per feature, and per plan tier, so cost and revenue move together in the forecast instead of surprising you three months apart.
None of these six models are complicated on their own. What’s hard is building all of them so they talk to each other, so a change in the headcount plan flows into the cash model, and a spike in AI usage flows into gross margin instead of getting discovered at quarter close. That’s the part most founders don’t have time to do themselves, and it’s the part that matters most before a raise, a board meeting, or a renewal.