Paid subscribers can listen to this in convenient podcast form and
download the Study Guide, Briefing Doc, and FAQ.

“AI Product Manager” is the new hot new title. It’s in demand, sounds exciting, and, in NYC and Silicon Valley, pay more. LinkedIn, podcasts, and pronouncements by Big Tech can give you a serious case of FOMO.

But is it sustainable?

To answer that, we need to understand how AI businesses actually work.

I’ve been there. I worked at an AI startup and launched a GenAI product. What I learned isn’t what the hype tells you.

Today, launching an AI app has never been easier. Every week we hear about someone who couldn’t code six months ago, now vibe coding a product, raising rounds, and talking about “changing the world.”

But peel back the hype, strip away the self-serving podcasts, investor decks, and VC-fueled valuations, and you see a more sobering reality:

Most AI businesses are terrible.

As a Product Manager — whether you’re building an AI product, considering joining an AI company, or watching your own company pour money into AI — you need to understand the economics.

Because your job and career depend on it.

So, today, I’m going to unpack the reality and talk about what it means for you.

Do you know why SaaS took off?

Not because of slicker UIs, better features, or Agile.

Customers weren’t clamoring for SaaS. In fact, many CFOs often preferred the old on-premise model because they could book it as an asset on the balance sheet instead of an expense on the income statement (where a SaaS purchase lives).

SaaS took off because investors and boards loved the economics:

The result? 70–90% margins.

Further product investments were meant to keep churn down, renewals high, and extend customer lifetime (LTV).

As long as lifetime value (LTV) was 3x+ acquisition cost (CAC), SaaS delivered hockey-stick growth.

AI apps flip that model. Every new user creates new costs:

This results in 30–60% margins at best. Even Anthropic’s Claude sits around 55%.

And some of that margin may come from “dead subs” — paying but inactive users, often acquired from bundled AI app deals. Those deals hide this, but those are the customers most likely to churn.

I saw this firsthand.

Our board wanted 85% margins. My analysis showed 60% max — and only if we priced high enough to lose the low-end of the market. That risked shrinking our addressable market and limiting our growth. More realistic was 35–45%.

The culprit: ML processing costs. More complex use cases meant more GPU resources, which meant skyrocketing costs.

That’s why:

This isn’t software as usual. Here, growth - more cost. Sometimes exponentially.

ChatGPT wants 1 billion users by EOY. Today: ~700–800 million users. Sounds impressive.

But only ~10 million pay for ChatGPT Plus. That’s ~2% conversion.

Even the rosiest estimates put it at 5–7%. Still shockingly low given the hype and reach. And abysmal compared to the average SaaS product.

And remember: free users aren’t free. Compute costs money.

This is where GenAI is right now:

Mass reach today comes more from media hype than product value. Even foundational AI companies are deeply unprofitable. Their current economic models can’t sustain themselves.

Sitting at the peak of the hype cycle isn’t correlated with long-term benefits.

Blockchain. Crypto. Augmented reality. All rode hype cycles to the moon before crashing down to earth.

Sure, LLMs will get better. Cars got better too. Today we have Teslas and Porsches, sedans and SUVs and trucks, EVs and plug-in hybrids. But they’re all cars. The fundamental economics of the car business remain the same, because the product delivers essentially the same value.

Can a software company stay unprofitable for years? Sure — if they keep raising cash off hype. But hype has limits.

Like mobile, social, and streaming, eventually LLMs will commoditize. Drastic changes in their pricing will be inevitable, as that will be the only way to fuel profitable product growth and maintain market share. (Just look at Netflix, Disney+, Apple TV+, Amazon Prime Video, and other streamers raising their prices or adding ads.)

AI may be transformative. But as a business model, AI-native SaaS isn’t the slam dunk many assume.

Some AI startups have found viable niches, especially in industries drowning in complex workflows like HR, sales, accounting, legal, finance, and healthcare administration.

Examples:

Maybe not billion-dollar moonshots. But the potential to be real, sustainable businesses.

The common thread? They solve a real, painful problem where AI itself isn’t the product, it’s the accelerant.

So, how do you take all this noise and apply it to your world? Let’s consider 3 areas: