The Real Cost of AI in SaaS

Carl Anderson

In the latest episode of the SaaS Growth Podcast, Carl flies solo to break down one of the most overlooked risks in modern SaaS: AI costs, and why they behave nothing like the costs we're used to.

Drawing on his work advising SaaS companies on AI adoption and spend, Carl unpacks why AI-based SaaS businesses are seeing gross margins as low as 25–45%, compared to the 70–85% typical of traditional SaaS. He argues that the real problem isn't that AI is expensive — it's that AI costs rise with usage, breaking the flat-cost, high-margin model SaaS has relied on for years. Carl also digs into where AI budgets are quietly wasted, from over-provisioned agentic AI to features that get shipped and never revisited, and lays out a more disciplined, adversarial approach to using AI without letting it erode your margins.

This episode is a must-listen for SaaS founders and product leaders trying to figure out whether their AI features are actually paying for themselves.

Key Insights from the Episode

Why AI Costs Are Fundamentally Different

  • Traditional SaaS margins improve with scale because costs stay flat while revenue grows — AI costs rise right alongside usage instead.
  • AI-based SaaS companies are averaging far lower gross margins than traditional SaaS, even among the fastest-growing AI startups.

The Shift to Usage-Based Pricing

  • Token- and usage-based pricing models move SaaS away from recurring revenue and closer to a series of one-off transactions.
  • Carl explains why this shift changes the underlying economics of the business, not just the price tag.

Where AI Budgets Get Wasted

  • Agentic AI is often over-built and run on premium models for tasks that don't require it.
  • "Ship it and forget it" — the classic SaaS playbook — becomes an expensive habit when applied to AI features that need ongoing cost review.

How to Use AI Without Letting It Eat Your Margin

  • AI is best used to prototype and test features fast, especially where requirements are hard to pin down up front.
  • Carl makes the case for staying deliberate — even adversarial — about how much AI stays in a product long-term versus getting replaced with cheaper, traditional logic.

Episode Highlights

  • 00:02 – Introduction: why AI costs deserve a closer look
  • 00:52 – Why AI costs are worse than other SaaS costs
  • 02:29 – The shift from subscriptions to token-based pricing
  • 03:13 – The problem with model cascading
  • 04:29 – How AI costs are being wasted on agentic solutions
  • 05:41 – Why "ship it and forget it" doesn't work for AI
  • 06:38 – The right way to use AI
  • 08:05 – Final thoughts and how to get in touch

Why You Should Listen

This episode offers a clear-eyed, practical framework for thinking about AI spend in SaaS — not as an inevitable cost of doing business, but as a decision that deserves the same scrutiny as any other line item. If you're building or scaling AI features, this is a chance to catch cost leakage before it quietly eats your margin.