AI Business Models

Last updated October 25, 2026

How to build an AI business

Explore the models and how they work. Figure out which models match your business, so you can design your own.

Why AI needs its own business models

AI changes the economics of software businesses. Traditionally, serving one more user or customer cost you almost nothing, so free tiers were cheap and pricing per seat made sense. With AI, everything takes compute. Compute costs money. The result? A free user can be expensive, a heavy user can be unprofitable, and what the customer values is often a finished piece of work rather than access to a tool.

That's why AI pricing is moving away from seats and toward usage, outputs and outcomes, and why the metrics that matter change too. If you sell usage credits (as a proxy of tokens), you care about how fast a new user gets a first quality output. If you charge based on outcomes, you care about how often your system succeeds. Defining your business model tells you how you can make money and which numbers to watch.

How to use this guide

Find your models. Answer the three questions below and note every model that fits. Most of you will match more than one, and will merge business models together.
Read the diagrams. Each page shows how customers arrive, pay, expand and leave, and the loops that drive growth and cost.
Pick your metrics. Each page has a table of the metrics to track, what each one means and why it matters.
Draw your own model. Every business works differently and is at a different stage. Create your own business model systems diagram and figure out what to focus on.

Three questions that place your business

Each model answers three questions differently: who pays, what they pay for, and who does the work. Find the ones that match your answers.

Model	Who pays	What they pay for	Who does the work
01 Model and inference APIs	Companies building AI into their own products	Usage: tokens	The customer, with your model
02 Consumer AI subscription	Consumers	Access, in tiers set by how much they use	The customer, with your tool
03 Subscription for usage credits	Individuals and teams	Usage: a monthly pool of credits	The customer, with your tool
04 Pay per job	Anyone with a job to get done	Output: one finished job, whether or not it works out	Your software
05 AI add-on to existing SaaS	Customers of a product they already use	Access: an add-on, a higher tier or a credit meter	The customer, with your tool
06 Outcome-priced AI agents	Businesses	Outcome: a successful result. Failed attempts are free	Your software
07 Ad-supported AI	Advertisers. Users pay nothing	Reaching your users	The user, with your tool
08 AI-native services	Businesses that would otherwise hire a firm	Output, priced like a firm: per job, commission or basis points	You: AI does most of it, people check it

Companies often match multiple rows. Crosby, an AI-native law firm, charges about $400 per contract. The price makes it pay per job (model 4). The attorneys who review every document make it an AI-native service (model 8). The first describes how the customer pays. The second describes who does the work and stands behind it.

Case Study

Candor: pay per job with subscriptions mixed in.

Candor fits pay per job on all three questions above, but it's more than that. See what the subscriptions change in its diagram and scorecard.

Explore

Selling seats to businesses? A new AI product priced per seat isn't a separate model here. Its loops are classic SaaS loops, which Lean Analytics covers. What AI changes is the cost: the seat price is fixed, but the compute each seat uses isn't. Model 5 shows how seat expansion works, and model 3 shows how credits can cap heavy use.

Choosing between models

Here's how to figure out which model fits your business. Each model's page adds what to test first and which metric to watch.

Model	Choose it when	Avoid it when
01 Model and inference APIs	You sell a capability other companies build into their own products, so your usage grows as theirs does.	Your volume can't outgrow prices that fall about 10x a year, or a few customers carry most of your revenue.
02 Consumer AI subscription	Individuals use the product often, and paying raises a usage limit rather than unlocking a better model.	People use your product in rare bursts, with nothing bringing them back frequently enough. Or the free tier costs more to serve than paying subscribers are worth.
03 Subscription for usage credits	Usage varies a lot between customers, and they want a predictable monthly price more than a price per result.	Customers can't tell what a credit buys, or a few heavy users would burn your margin before you can set limits.
04 Pay per job	Customers already buy this work one job at a time, and you can count each job.	You can't deliver every job well enough to earn the next one. The customer pays either way, so one bad output can lose them.
05 AI add-on to existing SaaS	You already have paying customers, and AI clearly improves the product they use every day.	Your buyers expect AI in the base price, as 81% did in G2's interviews, or AI's compute cost would eat into your margin with no line item to cover it.
06 Outcome-priced AI agents	The work is narrow and each result is easy to count, like a resolved ticket or a recovered chargeback.	Results are broad or hard to define, or you and the buyer would count them differently.
07 Ad-supported AI	You have a large, habitual audience, and many of their questions show buying intent, like shopping, travel or local services.	Few of your users' questions are commercial, or ads would make users doubt the answers.
08 AI-native services	Clients already hire a firm for this work, and AI can take over a growing share of each job while experts check the results.	AI can't continue to take over more of the work. If margins don't climb, the profitability might not be there.

Go further

To learn more about how to use these business models in practice: