A tale of two new SaaStr AI tools that reveals the future of enterprise AI economics
We just shipped two new AI-powered tools at SaaStr, and the LLM costs and token usage patterns tell a bimodal story about where enterprise AI costs are headed.
Both SaaStr AI tools are equally cool, and provide a ton of value, but deep down the second uses far more AI, even with a fraction of the usage so far (albeit it’s new). And is much more complex:
Tool #1: AI Startup Valuation Calculator
- Completions: 275,000+ valuations processed in < 30 days
- Token burn: Less than $50 in OpenAI credits
- Cost per use: ~$0.0002
Tool #2: AI VC Pitch Deck Analyzer
- Completions: 400+ pitch deck analyses in first 48 hours
- Token burn: $80 (and climbing)
- Cost per use: ~$0.20 per analysis
That’s roughly an 800x difference in token consumption per use case. Same company, same month, wildly different AI economics.
Now to be clear, I haven’t optimized token costs in the Pitch Deck Analyzer yet. I want to get past 1,000-2,000 founders using it first. There are ways to bring costs down. But only up to a point.
Why the Massive Gap?
The valuation calculator is elegantly simple. Feed it some basic startup metrics, get back a number. One API call, minimal context, job done. And deep down, almost all the “AI” was done upfront, synthesizing 4,000+ recent VC rounds into one rich table. The additional OpenAI calls are there, but fairly minimal.
The new pitch deck analyzer? That’s where things get interesting (and expensive). We’re running up to 5 passes through Anthropic’s API for a single analysis in some cases, and a really rich query, analyzing 10+ MB files in most cases:
- Initial deck parsing – Extract and structure content
- Market analysis – Compare against industry benchmarks
- Financial model review – Deep dive on unit economics
- Competitive positioning – Stack ranking against comparables
- Investment readiness scoring – Final synthesis and recommendations
- Comparison to 4,000+ other recent start-up VC deals and input from 800+ VCs – This actually consumes less tokens than rest, since we have this work from Valuation Calculator, but takes some tokens
Each pass builds on the previous one, creating a compound intelligence effect that delivers genuinely useful feedback. But it also creates compound token consumption.
The Paradox That’s Reshaping B2B AI Economics
Both trends are accelerating simultaneously:
On one hand, token costs are plummeting. What cost $50 last year might cost $5 next year. The efficiency gains are real and dramatic.
On the other hand, we’re finding ways to use 10x, 100x, even 1000x more tokens per use case as we discover what’s actually possible when you throw more and more AI horsepower at complex business problems.
As Amjad Masad CEO of Replit noted, we’re in this odd economic moment where we’re simultaneously:
- Driving unit costs down by orders of magnitude
- Driving usage up by orders of magnitude
What This Means for B2B Leaders
1. Budget for The Future, Not Just Your Product Today. Your AI costs aren’t going to follow a neat downward trajectory. They’re going to be lumpy, experimental, and occasionally shocking. Your team will optimize AI costs, but at the same time, if they are any good, they’ll also build new features than want to use 10x-100x more tokens.
2. Simple AI Will Get Cloned and Commoditized Fast If your AI feature can be replicated with a single API call and basic prompting, you’re building on quicksand. The valuation calculator works great, but it’s not defensible long-term. The real moats are being built with multi-step, context-aware AI workflows.
3. The Future Belongs to AI-Native Workflows The pitch deck analyzer isn’t just “AI-enhanced” or a co-pilot. it’s rethinking the entire process of startup evaluation from first principles. That’s where the 800x token premium gets you: genuinely new capabilities, not just faster versions of old ones.
Are you using AI to make existing processes 10% better, or to make entirely new processes possible?
The companies winning with AI aren’t just making their current workflows more efficient. They’re discovering entirely new workflows that were impossible at previous price points and capability levels. And the more AI advances, the more AI they are going to want to use.
Looking Ahead: The Token Economics of 2026+
My prediction — and Replit’s CEOs as well: we’ll see both trends accelerate:
- Simple AI tasks will approach zero marginal cost
- Complex AI workflows will justify premium pricing through genuinely differentiated outcomes
The winners will be the companies that figure out which problems are worth throwing serious AI horsepower at, and which ones should stay simple and cheap.
The 800x token difference between our two tools isn’t a problem to solve – it’s a feature to embrace. It’s the difference between AI as a cost center and AI as a competitive advantage.
