The four metrics that actually tell you whether your company is healthy. Adjust the assumptions, watch what hiring or unit economics shifts do to your runway. Runway alone is the metric that lies to you the most.
Four numbers. Computed live from the model below. Green is healthy, yellow is watch, red is fix-now.
Drag the sliders on the left. The P&L on the right updates live. Drag the inference cost to $0 to model BYOK (customers bring their own API key) and watch unit economics flip.
Most operators track runway. Runway alone is the metric that lies to you the most. Healthy burn is the intersection of four numbers, not one.
The existential question. At your current growth rate, do you reach breakeven before cash runs out, without raising another dollar? It collapses runway, growth, and burn into a single binary: are you on track to fix yourself, or aren't you?
Months until cash hits zero. Pairs with Default Alive/Dead: if you're alive, runway is effectively infinite (you reach breakeven first); if you're dead, runway tells you when. The "X mo at zero growth" sub-line is the worst-case anchor — what if growth stops tomorrow.
Net burn divided by net new ARR in the same period. Tells you how efficiently you're converting cash into growth. Two companies with identical burn and growth can have wildly different burn multiples, and the second one is a much better business.
For an AI-heavy product, this is the unit economics question. Every new customer either funds the business or accelerates burn — the sign of this number tells you which. Watch the trend monthly: are inference costs falling faster than usage is growing?
Gross margin in an AI-native company is shaped by three forces. The biggest single lever isn't pricing or efficiency, it's whether you're the inference middleman or not.
Frontier model pricing trends down significantly per generation. If you do nothing else, your unit costs improve over time. The question is whether customer behavior outruns the price drops.
As customers find more value, they use more tokens. Power users can multiply consumption within months. Without architectural changes (caching, smaller models for routine work, prompt compression), usage growth eats into margin.
If customers bring their own model API keys, you stop paying for inference at all. You become a software platform, not an inference middleman. Try the preset above. The change to your gross margin is the most powerful single architectural decision on this page.