Tactical insights for first-time founders to outsmart the burn, the churn & the breakdown.

Hey Founder,

Nobody knows how to price AI.

OpenAI charges for tokens. Anthropic charges for model usage. Intercom’s Fin charges per resolved conversation. Others charge per seat, workflow, agent, or some mix of all of it.

The biggest AI companies in the world still disagree on what the unit of value actually is. So if you’ve been feeling like you’re missing the obvious answer, you’re not. There isn’t one.

As a CFO, every time I work through AI pricing with a founder, I see the same thing: every model breaks somewhere.

One gets too cheap when the product gets good. Another makes costs unpredictable. Another sounds elegant in a pitch and ugly in the margins.

The challenge isn’t finding the perfect pricing model. It’s finding the one your customer understands, your economics can support, and your business can grow with.

This issue is about the four pricing models shaping AI today, where each one breaks, and how to choose one without destroying your margins.

Let’s dive in.

Strip away the branding and there are really only four ways to price AI: seats, usage, outcomes, and hybrid.

All four work. All four break.

The classic SaaS model charges per user, per month. It works for companies like Slack and Figma because costs stay relatively predictable while revenue scales. AI doesn't behave that way.

When one user can suddenly do the work of five people, seats stop measuring value. The better your product gets, the more leverage you give away.

The agency equivalent is hourly billing and traditional retainers. Revenue stays tied to time spent even as productivity improves.

If AI is increasing leverage, your pricing needs to capture some of that leverage too.

Usage-based pricing charges for consumption: tokens, messages, API calls, jobs processed. OpenAI’s API pricing is explicitly token-based.

The appeal is obvious:
• Revenue scales with usage.
• Costs and revenue stay aligned.
• Margins become easier to model.

Here’s the catch: more usage doesn’t always feel like more value.

Customers hate unpredictable bills. Once every interaction feels metered, they start optimizing for lower usage instead of higher adoption.

On paper, the model says: more usage, more revenue.
In reality, it often becomes: more anxiety, slower adoption.

Usage-based pricing works well for infrastructure and APIs. It's much harder when buyers want predictable budgets.

Outcome-based pricing charges for completed work: a resolved ticket, a booked call, a processed claim. For agencies, that means charging per asset, content piece, or campaign instead of billing for hours.

Products like Fin charge per resolution.

On the surface, it's the dream. Price aligns directly with value.

But very AI products deliver clear specific outcomes. Most are still copilots, not workers.

And even fewer deliver a clean result without messy dependencies. Results depend on prompts, workflows, data quality, human intervention, and customer behavior.

So you end up underwriting variables you don't control.

Average margins can look fantastic. Edge cases can destroy them.

Hybrid models combine multiple approaches:
• Platform fee (covers fixed costs),
• Included usage (covers variable costs),
• Outcome-based upside (scales with usage),
• Overage charges (sets ceilings so users aren’t shocked by the bill)

You see this everywhere now because it reflects reality better than any single model.

Intercom's Fin has a hybrid offer that combines outcomes- and seats-based pricing.

The danger is complexity. It's the easiest model to overbuild.

Every pricing lever solves a founder problem while creating a sales problem. Eventually customers need a calculator to understand the bill.

That's why hybrid pricing is usually something companies grow into, not something five-person startups should start with.

When you look at how AI leaders price today, the real takeaway is simple: Everyone is still experimenting.

So stop asking, "What's the right way to price AI?"

There isn't one.

The better question is: every model breaks somewhere, so which trade-off can you live with given your product, your stage, and what your customer will actually tolerate? 

Most AI pricing debates focus on "capturing value."
Your buyer is focused on friction:
A bill they can't explain internally.
An invoice that changes every month.
Feeling metered on every interaction.
Paying for outcomes they don't fully trust.

So ask: Where does this customer already feel pricing pain today?

And does our pricing reduce that friction, or add more of it?

If your sales team can't explain the pricing in one breath, your customer can't either. And a buyer who can't explain it won't champion it.

The goal isn't to maximize value capture, it's to make buying feel easy and paying feel worth it. 

Three things matter:
1. Outcome Clarity: Can you tie the product to a KPI the customer already tracks?
2. Human Involvement: Is the AI assisting a human or completing the work?
3. Cost behavior: Do your costs stay stable or rise with each unit of work?

If the AI mostly assists and the impact is hard to isolate, outcome-based pricing is difficult to defend.

If it delivers measurable work and costs stay controlled, outcome or hybrid models become much more viable.

That's why AI companies and AI-enabled agencies end up with different pricing structures. Their autonomy, attribution, and cost economics aren't the same.

Think of the following as a pattern map, not a rulebook.

You're not graduating from one model to another. You're matching your product and cost structure to a bill your customer can comfortably say yes to.  

If you're pre-seed, under ~$500K ARR, and have fewer than 30 customers, you probably don't need hybrid tiers, usage bands, or outcome ladders.

You need one thing: Can a prospect explain your pricing in one sentence and still want to buy? 

Think simple: One model. One metric.

Simple enough that your champion can explain it internally without opening a document.

The perfect pricing model can wait until $5M ARR. Right now, the question is whether customers can say yes quickly. 

A useful test: ask three early customers to explain your pricing back to you as if they were selling it to someone else.

If they can't, it's probably too complicated.

Superhuman didn't launch with complex tiers. It launched with a simple promise: roughly a dollar a day to get hours of your life back.

One number. One story. One yes.

A slightly underpriced deal that closes beats a beautifully engineered model sitting in someone's inbox. 

Pricing is the last step. Before debating tiers or token rates, answer three questions:
• Does the AI assist humans or complete work on its own?
• Can you tie the result to a KPI the customer already tracks?
• Do your costs stay stable (or increase) as usage grows? 

Those answers eliminate bad options quickly.

If the AI mostly assists and the impact is hard to measure, outcome pricing is probably off the table, no matter how attractive it sounds.

The right model depends on where you are, not where you want to be.

• Pre-seed (<30 customers): one model, one metric.
• Seed: one primary metric, maybe a second tier.
• Series A+: usage-based pricing with a base fee starts making sense.
• Growth: hybrid pricing can finally earn its place.

Most founders build pricing for the company they hope to become.

Price for the company you are today. 

AI pricing always looks cleaner in a deck than it does in front of customers.

Pick the simplest model that:
• converts,
• protects margins,
• and is easy to explain.

Run it with your first 10–20 customers without custom deals. Then watch where it breaks.
• Strong conversion but weak margins? Capture more value.
• Strong margins but weak adoption? Simplify.
• Clear customer segments emerging? You may have earned a second pricing structure.

The goal is finding where customers feel friction and where your economics start leaking. 

A founder asked me:

A quality user is not a signup. It’s someone who looks like future revenue.

Use a simple rule: Quality user = ICP fit + value moment + buying signal.
• ICP fit: the right customer with a problem worth paying for.
• Value moment: they do the thing that makes the product useful - connect data, invite teammates, run a workflow, turn on an integration.
• Buying signal: they ask about pricing, hit a usage threshold, book a call, or show a specific use case.

When testing a channel, ignore:
• visitors
• signups
• trials

Track only:
• quality users
• cost per quality user

If a channel can't produce a handful of quality users after a few weeks, it's probably not "early." It's probably just bad. So, move on. 

Send it my way, just hit reply.

Winning resources for founders: 

Before you go…

Nobody has solved AI pricing. The founders who get closest are usually the ones closest to their customers.

Every renewal, pricing objection, and “why did my bill spike?” conversation is feedback. Use it.

The company that learns fastest where pricing friction lives doesn’t just win the pricing debate. It wins the category.

That’s the moat.

See you next Thursday,

— Mariya

Hit reply and let me know. I read every single one (for real).

Hey, I’m Mariya, a startup CFO and founder of FounderFirst. After 10 years working alongside founders at early and growth-stage startups, I know how tough it is to make the right calls when resources are tight and the stakes are high. I started this newsletter to share the practical playbook I wish every founder had from day one, packed with lessons I’ve learned (and mistakes I’ve made) helping teams scale.

Mariya Valeva

Find me on LinkedIn