Pricing, Margins and the Business Model

Pricing deserves attention before launch, because it is the clearest test of whether you have built something people value, and it sets how much room you have to grow. This chapter argues for charging from day one, for a model that moves with your costs and your customer's value, and for knowing your numbers well enough to tell whether the company is on a path to survive.

Charge From Day One

A customer who pays, even a small amount, gives you a signal that no number of free sign-ups can replace. Sign-ups show what people will try; payment shows what they intend to keep. Paying customers also complain more specifically, which is what you need early on.

Founders delay charging because they fear rejection or want the product to feel finished first. Both instincts cost time. If nobody will pay for the rough version, polish rarely changes that. Usually the problem is not painful enough, or you are talking to the wrong buyer.

Your first price need not be permanent. Treat it as a hypothesis: write down why you chose it, and plan to revisit it once you have a few dozen paying customers.

Choosing a Pricing Model

Most software is priced in one of four ways. Choose the one where what the customer pays rises with the value they get, and ideally with what it costs you to serve them.

Model	How it works	Fits when
Per seat	A fixed price per user per month	Value grows with the number of people using it, and serving each user costs little
Usage-based	Pay per request, document, minute or unit processed	Customers vary widely in volume, or each unit of work has a real cost to you
Outcome-based	Pay per result: a resolved ticket, a booked meeting, a recovered invoice	The outcome is measurable, attributable to you, and both sides agree on what counts
Hybrid	A base subscription plus usage or overage charges	You want predictable revenue but also need to protect against heavy users

Per-seat pricing is simple to understand and easy to forecast. It breaks down when one user can generate enormous load, or when your product replaces work people used to do, because success then shrinks the seat count.

Usage-based pricing tracks cost and value closely but makes bills harder for customers to predict. Buyers at larger companies often need a budget figure before they can sign, so pure usage pricing can slow sales even when it is fair.

Outcome-based pricing is the most persuasive when it works, because the customer pays only for results. It also requires you to measure the result reliably and to settle disputes about what counted. Intercom's Fin support agent is a public example: it is priced at $0.99 per outcome, and Intercom spells out what counts as one, such as a customer confirming the problem is fixed or not asking for further help. Start with this model only if the outcome is already tracked in a system both sides trust, and write the definition down precisely.

Pricing AI products

AI products change the calculation because every request has a real compute cost. A traditional software feature costs almost nothing extra each time someone clicks it. A feature that calls a large model costs money on every call, and one heavy user can cost more to serve than they pay.

Many AI products therefore end up on a hybrid model: a subscription with a sensible allowance, plus limits or usage charges above it. ICONIQ's 2025 survey of 300 software executives found the shift was toward exactly this mix of base subscription and usage charges, and that 37% planned to change their pricing within a year. AI pricing is still settling, so expect to revisit yours.

Whatever you choose, know your cost per request, your cost per active user and what your heaviest users cost. Offer an unlimited plan only after you understand that distribution.

Gross Margin and Inference Cost

Gross margin is revenue minus the direct cost of delivering the product, divided by revenue. Direct costs include hosting, model inference, third-party APIs your product depends on, payment fees and, often, the support needed to keep customers running.

Classic software businesses earned high valuations partly because serving one more customer cost very little, which kept gross margins high. AI products do not get that assumption for free. Inference is a cost of goods sold, and it scales with usage.

The gap has been visible for years. In 2020, Andreessen Horowitz reported that the AI companies it spoke with often had gross margins of 50–60%, against a 60–80%+ benchmark for comparable SaaS businesses, largely because of cloud and inference costs. Bessemer's 2025 report found the fastest-growing AI startups it tracked averaged only about 25% gross margins, sometimes negative, while a steadier group averaged about 60%. Fast growth can hide thin margins for a while without fixing them.

The cost side is moving in your favour. Andreessen Horowitz estimated in 2024 that for a model of equivalent performance, inference cost was falling about tenfold per year. The trend may not hold, but it means a margin problem today can shrink if you keep re-checking your model choices.

Practical ways to improve AI margins:

• Use the smallest model that does the job well, and reserve larger models for the steps that need them.
• Cache results that repeat, and stop resending the same context on every call.
• Cap or meter expensive actions instead of hiding them inside a flat price.
• Re-check costs regularly, since model prices and capabilities change.

Investors and acquirers watch gross margin because it shows how much of each new dollar of revenue is left to pay for growth and overhead. A business with thin margins has to grow much larger to cover the same fixed costs.

Unit Economics in Plain Terms

Unit economics answer one question: does each customer bring in more than they cost to win and serve? Four ideas cover most of it.

• Customer acquisition cost (CAC): total sales and marketing spend in a period, divided by the new customers won in that period.
• Payback period: how many months of gross profit from a customer it takes to earn back their CAC.
• Churn and retention: churn is the share of customers (or revenue) you lose each period; retention is the share you keep.
• Lifetime value (LTV): the gross profit you expect from a customer over the whole time they stay.

Here is a worked example. These are hypothetical numbers chosen to show the arithmetic, not benchmarks.

• A customer pays $100 a month.
• Your gross margin is 70%, so each customer produces $70 a month in gross profit.
• You spend $7,000 on sales and marketing and win 10 customers, so CAC is $700.
• Payback period is $700 / $70 = 10 months.
• Monthly churn is 4%, so the average customer stays roughly 1 / 0.04 = 25 months.
• LTV is about $70 × 25 = $1,750, against a CAC of $700.

Read that example as a set of levers. If churn rises to 8%, average lifetime halves and LTV falls to about $875. If inference costs push gross margin down to 50%, payback stretches to 14 months. Small changes compound, which is why retention usually deserves the closest watch. Your early data will be thin, so treat these figures as directions rather than precise measurements.

Raising Prices

Many founders underprice. They anchor on their own costs or on what they would pay themselves, when the reference point should be the value the customer receives. If almost nobody pushes back on your price, it may be too low.

Raise prices for new customers first, and watch whether conversion falls meaningfully. For existing customers, give clear notice, explain what has improved, and consider keeping them on the old price for a period. People tend to accept price changes more readily when they hear about them early and plainly. Handled badly, an increase can cost far more than it earns: after Netflix raised some plan prices by as much as 60% in 2011 and announced plans to split its DVD and streaming services, it lost about 800,000 U.S. subscribers in a single quarter.

A price increase is also a chance to restructure: add a higher tier for customers who want more, separate features that cost you more to deliver, or move heavy users onto a plan that reflects their usage.

Default Alive or Dead

In his 2015 essay "Default Alive or Default Dead?", Paul Graham asks founders a simple question. Assuming your expenses stay where they are and revenue keeps growing at the rate it has over the last several months, will you reach profitability with the money you have left? If yes, you are default alive. If no, you are default dead, and you need to change something instead of assuming the next round will arrive.

Answering it starts with runway. Runway is cash in the bank divided by monthly net burn, which is what you spend each month minus what you collect. As a hypothetical example, $600,000 in the bank with $50,000 of net burn gives you 12 months.

In practice that figure flatters you. Fundraising takes months, new hires add cost before they add revenue, and expenses tend to creep up. Recalculate runway monthly, model a version where growth slows, and decide in advance what you will cut if the answer comes back default dead.