Imagine you're building an AI product.

It gets 10,000 users overnight.

Then AWS sends you a bill that is bigger than your revenue.

The AI companies that can't make economics work aren't necessarily the ones with bad products.

They're often the ones with bad pricing.

Meanwhile, the ones that are making money charge 10 times more than traditional software, and customers are paying.

Grammarly slashed their free AI offering by 95%.

Cursor jumped from 20 to $200 a month, a 10 times price gap that would kill any traditional SAS company.

Replet started charging based on computational effort, not features.

And it's working.

Today I'm doing a deep dive into which pricing models work for AI products, which ones destroy unit economics and margins, and how to price an AI product that will not make the founder go bankrupt.

Let's dive in.

I got inspired to make this episode after seeing Stripe's article, a framework for pricing AI products.

The framework describes three monetization formats most AI companies follow: consumption-based per API call per LLM token, workflow-based per completed task or outcome-based per successful outcome.

Simple, right?

Well, not quite.

In traditional SAS, margins are 80 to 90%.

Once infrastructure is deployed, adding an extra user costs you nothing.

In AI products, the margins are reportedly 50 to 60%.

And that is because every user that you add to your product burns tokens and computing power.

So your costs to maintain clients or your cost to onboard new clients can scale unpredictably.

I'll give you an example.

Let's say you're Figma and you have an enterprise customer, customer A.

You add a new AI feature where a user can generate designs based on their description.

Your customer, company A, loves the feature and they start using it multiple times every single day.

You were hoping that they'd use it, but maybe not as much.

And when they start doing it, your GPU usage shoots up and a week later you get your AWS bill and you see a 300% spike.

This situation was not possible before AI.

And what AI has done to the unit economics of businesses is that it changed annual recurring revenue from close to 100% predictable to 70% predictable, 30% consumption surprise.

We got to a point where your cost forecasting needs to move from annual planning to 90day cycle.

This is why most AI founders go bankrupt.

Well, it's not like they become homeless all of a sudden, but you know what I mean.

This is the reason why businesses fail and why AI founders cannot sustain their ventures for a long time because they price like SAS. But the costs behave like utilities.

Running marketing for a business is not for the faint of heart.

One tool for email, one for social media, another one for CRM, something else for WhatsApp, and then three separate apps that track analytics for three different use cases.

And all of it just to understand your customers.

But what if all of that just worked and you didn't have to spend a fortune?

Brebo asked the same question and they built the answer.

An all-in-one marketing and CRM suite for growing teams that don't want enterprise chaos or enterprise pricing.

With Brebo, you can run email, SMS, WhatsApp, and chat campaigns from one clean platform.

No app fatigue, no endless platforms to manage.

Just seamless communication with your customers wherever they find you.

Build automations in a few clicks.

Send a welcome series. Recover a cart.

Re-engage that lead who ghosted you in Q2.

And build that dashboard for conversations.

It will definitely make your support's life a lot easier.

And it's not just marketing.

Brevo includes a powerful customer data platform, which is essentially CRM and business analytics combined in one place.

It turns all of your customer data into a single source of truth and gives you a 360 view of every single customer you have.

Also, Brevo's Aura AI helps you write smarter, personalized emails and send them at the perfect time, and that feature is available even on the free plan.

Unlike other tools that charge per contract, Brevo charges only for what you send with transparent pricing that's up to 72% cheaper than competitors.

So, if you're ready to simplify your marketing, unify your team, and optimize your marketing spend, check out Brevo.

Huge thanks to Brevo for sponsoring this part of the video.

We received an email from a young founder whose name is Chandler and Chandler was asking for advice on how he should approach pricing for the app he's building.

I have been promising him this video for almost 2 months, maybe more.

So, Chandler, this is for you, but I am really sorry it took me 2 months.

Chandler is building a product called Serenity Notebook.

Check it out. The link is in the description.

What does it do?

Serenity notebook lets you use AI technology to create and share real-time visualizations with different types of data in just one place.

You can easily build charts for CSV data, explore computer a design models, and build datarrich notebooks that update in real time independent of programming language or framework.

Important clarification, the core product that Chandler is building works without AI.

It doesn't require AI.

target users. According to the founder, data scientists who can use code in any language they write in, feed it to serenity and it'll build the charts.

Engineers of various kinds, electrical, mechanical, chemical, biochemical, can visualize their work using computer a design models or any other type of specialized files.

business and analytics teams who can use Serenity to create web dashboards.

and scientists who can build custom visualizations or physics simulations as an example.

Chandler's pricing so far desktop version free for the time being unlimited usage.

Cloud plan starts at $8 a month.

The rationale behind this is to advertise hardware and hosting costs.

Chandler, if you're listening to this, and I really hope that you are, there are a couple things that I would recommend that you do, because your pricing strategy has fundamental flaws and risks that could cost you tens of thousands of dollars per year and possibly your entire business long term.

because your pricing strategy has fundamental flaws and risks that could cost you tens of thousands of dollars per year and possibly your entire business long term.

So, the three fundamental risks that you're signing up for in my view. Risk number one, desktop cannibalization.

If you make desktop truly free for life or for the foreseeable future with no limits, I do not see how on earth you're going to be selling your cloud component because why would anyone pay for cloud if they can use your desktop for free?

because why would anyone pay for cloud if they can use your desktop for free?

There is already evidence from your own comparisons that you provided.

Jupiter is free and it makes no money.

Jupiter is a nonprofit living off of grants.

Observable tried free private notebooks. Then they had to payw wall those and when they did the users revolted.

If you keep desktop free with no limits, it's not a business. It'll be a pet project.

My recommendation to you, keep it on premium, but impose limits.

For example, two to three local notebooks, not unlimited.

one computer aid file at a time, five AI credits per month, no cloud hosting, no collaboration features.

I would recommend that you look at how the company called Muro structures their free tier.

They give you enough to see the full value, but not enough to never upgrade.

Risk number two, underpricing.

At $8 a month, you leaving massive value on the table because here is what the market looks like. The license for a computer aid design commercialgrade software starts at $75 a month and goes all the way up to $4,000.

Your potential competitor Hex starts at $36 and goes all the way up to $75 per user.

Observable starts at $22 to $25 per user.

And most importantly, the prices for the engineering team tools start at $100 per user.

You are thinking to price at eight.

That's between $10 to $60 per month per user left on the table.

At 1,000 users, that is more than $100,000 per year in lost revenue.

My recommendation, start at $18 a month minimum.

If your users don't pay $18 for a tool that visualizes data in a language agnostic notebook with computer aid design capabilities, you have a product problem, not a pricing problem.

When I say product problem, I mean that it either doesn't solve the problem of your target user or there is a mismatch in terms of perception of the product, which is a marketing problem, different one but still, or the product is not strong enough to solve the problem all the way to the end.

But 18 bucks for your product is less than minimal.

Risk number three, AI cost spillover.

Your AI features, custom visualizers will have unpredictable compute costs.

The moment you get a power user, your unit economics will collapse and your margins will be buried.

My recommendation, credit-based AI pricing from the time you launch.

You can do something like this. You can give five free AI credits per month.

For a pro plan, you offer 100 credits per month.

You do the math.

These numbers are hypothetical. I don't know what your costs are, but you get what I'm saying.

This example is a Canva Grammarly example.

Users have adopted behavior when they understand credits and you as a founder avoid surprise bills.

To understand why your pricing won't work, I want to look at what's actually working in AI pricing right now.

Because the game and AI pricing has changed completely.

Premium as we knew it is dead.

The golden rule in traditional SAS has always been to offer 40 to 60% of core features for free.

The whole point was habit formation.

Users would come to your product, they would use it for a long time, they would hit feature limitations, and if they like the product enough, they go to upgrade.

Today's AI SAS companies have slashed their free tier features to just 20% of what they had before.

That is an 80% reduction in what users get for free.

The mindset shifted from try before you buy to taste before you subscribe.

So let's go through what is now being offered in modern AI premium models.

Grammarly, 100 monthly AI prompts for free versus 2,000 paid.

Notion, 20 total AI responses, no reset.

Monday.com, 500 monthly AI credits and additional packs at $200 per month.

Cursor is a perfect example of how AI creates premium pricing tiers.

In the free tier, they've got 2,000 completions and 50 slow requests.

In the pro tier, they've got unlimited completions and unlimited slow requests.

In the ultra tier, they've got 20 times more usage than Pro.

AI features create pricing gaps never seen in traditional software.

Cursor had a 10 times jump.

Copilot had a two times jump and so did JetBrains.

And the thing is, users aren't pushing back.

The market has mentally accepted this model where free comes with a meter.

Then Replet showed up and asked what if pricing actually matched the pain of your GPU.

Replet introduced probably the most logical pricing model among all AI apps on the market.

They call it an effortbased checkpoint system.

They completely departed from traditional software pricing and started charging users based on computational work that is being performed.

The model works like this. For simple tasks, they charge up to 15 cents.

For complex tasks, up to a dollar.

The price that they charge directly correlates with AI model inference costs.

And here is how it works. Technically, the system measures computational units in real time.

An agent analyzes task complexity before execution.

It estimates CPU cycles, memory usage, and required model calls.

It calculates the cost based on predicted computational effort.

After that, it tracks actual resource consumption during the task and then at the end it charges based on actual effort.

What Replet did is that they essentially aligned pricing to the pain and serenity is about to give away the pain for free.

Replet's example shows how far pricing can be pushed when you align it to raw compute.

But not every product can or should go that far.

Let me show you what works across different types of AI products and where Serenity fits.

When AI delivers a discrete and measurable outcome that directly replaces human work.

So for example, things like a document review or a customer support ticket that gets resolved.

What works is outcomebased pricing when you charge per successful outcome.

For example, medical imaging AI charges between $2 to $15 per scan.

AI tools for sales teams charge per qualified meeting booked, not per outreach attempt.

When AI automates repetitive workflows, compliance checks or claims processing, what works is subscription plus usage credits.

The base subscription covers platform access and credits kick in when automation executes.

A few examples for this type of pricing. Revenue cycle AI in healthcare 2 to 5% of recovered billing revenue.

Compliance AI offers 5 to $25,000 a month subscription plus credits per documents checked.

When AI augments individual productivity, writing assistance, design tools, code completion, what works is premium plus credit consumption.

You provide generous free tier limited by AI credits and then you package it as a tiered subscription, multiple seedbased plans with escalating AI quotas.

Examples: Chad GPT, free for basic, paid for advanced models and higher limits.

Canva free users get very limited AI prompts.

Microsoft Pilot integrates into Microsoft 365 but full features required paid subscription.

When AI delivers insights, forecasts or analytics, what works is subscription tiers.

You combine seedbased pricing with feature tiers.

Examples: marketing analytics AI from $100 to $1,000 per user per month for dashboard access plus credits per query.

HR performance AI from $10 to $100 per user per month.

The vast majority of successful AI companies follow one of these patterns and you need to do the same because this works.

Let me map Serenity to one of these verticals.

In my opinion, the way I see your product, you map to two of these categories. Productivity apps because you make technical professionals more productive with tools for charts, data, computer A designs, and instant AI visualization and analytical and insight platforms because you take raw data and produce visualizations.

and analytical and insight platforms because you take raw data and produce visualizations.

My recommendation based on what I just said, premium with credit consumption because your product is following a productled growth motion where acquisition, retention, and onboarding are driven by the product.

Credit based pricing for AI heavy features because you're going to need to offset compute costs so you don't drown in bills.

And finally, and I know you didn't ask for this, but I'm going to do it anyway because I'm going to regret if I don't.

I would like to challenge your thinking around B2C market orientation.

I did a quick audience analysis for you and I really really recommend that you reconsider.

Your product already has a lot of unique selling points.

If I'm not mistaken, you're the only one combining computer a designs with data visualizations in a notebook style app.

You are language agnostic which is a massive differentiator on the market.

And lastly, which I think is fantastic, you are not AI first.

Your product does not have AI first positioning.

You're not an AI app, which is a huge plus because your product solves a problem even if there is no AI.

I want to come back to your target audience.

You said engineers, data, machine learning, business analysts, and scientists.

My recommendation is to make engineering your primary target.

When I say engineering, I'm talking about all kinds of engineering. Electrical, chemical, civil, mechanical, IoT.

Basically, all kinds of engineering far beyond software.

Why?

Because your product solves their highest pain.

There is no unified computer a design plus notebook solution on the market today.

engineers that you're going to be targeting have very high purchasing power because engineer oriented products easily charge $25 to $50 per month per user and for team plans it's between $100 to $200 per month.

You're going to have a very high expansion potentials because entire engineering orgs need this tool for data machine learning deep learning.

Make them your secondary target.

Why?

because you're going to be solving medium pain and the pain is medium because they have alternatives like Jupiter or Mattplot lib.

You have a good expansion potential in this vertical and data science as a whole has a larger total addressable market than engineering.

Scientists and researchers make them your tertiary audience.

Why?

Because they have high pain but low budgets.

Academia relies and depends on grants.

They constantly fight for everything including software licenses.

The upside of this audience is that the free tier will work really well for them.

They will build a community for you.

And in general, academic community is fantastic for brand building and word of mouth.

They're not going to bring you a ton of money, but they're fantastic for branding.

And lastly, business analysts and all kinds of business use cases.

I recommend that you drop this because pain for these folks is not big enough.

You're going to be competing with at least Tableau, PowerBI, and Adobe Analytics.

If you spread too thin, you're going to be fighting massive competitors without going allin.

And remember, narrow is the moat.

Here is what I think you are missing and something that I really want to recommend.

Think about switching to B2B market orientation.

Why?

Let's go back to your audience.

When do mechanical engineers or data engineers use data visualization tools or 3D modeling software?

At work, not at home.

Engineering is collaborative by definition.

How many engineers work alone?

I've been in tech for almost 10 years. I haven't met a single engineer who wasn't on the team.

Where do scientists spend time?

research R&D labs.

PhD students basically live at those labs.

Primary use cases, visualizing experiments, research with collaborators, creating notebooks for publications, or managing lab data.

Scientists almost always work in groups.

The results of their work must be reproducible. They must be sharable and stored for compliance.

Every single user profile you're targeting will be using your product at work.

This is the definition of B2B orientation.

Your product caters to biotech, manufacturing, R&D labs, engineering consultancies, university PhD department, research labs, IoT and physical product development groups.

The revenue potential in B2B is enormous.

What you're building is a fairly niche, but not too niche of a product.

You already have product differentiation.

You clearly have taste and your website proves it.

That is a skill that not many B2B founders have.

I mean, have you seen B2B software?

This leads me to my final recommendation.

Start designing for teams.

Right now, you're clearly designing for a B2C use case, but I recommend that you start thinking in teams and network effects.

Collaboration will be your bread and butter.

Explore how Loom, Mero, Dropbox set up their network effects when they were starting out.

B2C can be your entry point for validation, data gathering, brand building.

But your revenue sits with B2B.

Let's do a quick math.

Current B2C focus, $8 per month per user.

At 1,000 users, that is 96,000 per year.

Alternative scenario, B2B focus, $50 per month per team seat.

That's a conservative estimate. Way below typical computer a design viewer pricing.

Let's say your target is 200 team seats across 40 companies, five seats on average.

That is $120,000 a year.

And here is the real opportunity.

Typical engineering team budget is between $100 and $200 per user for specialized tools.

If Serenity positions itself as an engineering tool, you can charge $100 per month per seat.

If you sell 200 seats, that's $240,000 a year.

Same product, different positioning, two and a half times revenue.

At $8 per month B2C orientation, you are leaving over $100,000 on the table.

If you compare yourself to the B2B engineering positioning conclusion, the companies that cannot make economics work, the companies that are going bankrupt aren't always the ones with bad products.

They're the ones with bad pricing.

Same product, different pricing, 10 times outcome.

To all founders, including Chandler, you're building something genuinely innovative.

Don't let your pricing be the reason it fails.

Get it right and build a sustainable business.

Build something that will last.

I hope this was helpful.

Thank you for listening.

We'll see you next time.

Bye.