Blog • Business Metrics
By Cemhan Biricik — Founder of ZSky AI
Most AI founders track the wrong metrics. Total signups, page views, social media followers — these are vanity numbers that feel good but do not predict whether your business will survive. Here are the metrics I watch daily at ZSky AI and why each one matters.
This is the most important metric for any AI product that serves free users. Every generation consumes GPU compute, which has a real cost. If your cost per generation exceeds what your revenue model can support, your free tier is burning cash instead of building a funnel. I track this daily and optimize relentlessly through inference optimization.
What percentage of free users upgrade to a paid plan? For AI products with generous free tiers, 2 to 5 percent is typical. But the raw number matters less than the trend. If conversion is climbing, your product is creating genuine value. If it is flat or declining, something in the free-to-paid experience is broken.
Not daily active users — daily active generations. I care about how many images users actually create, not how many times they log in. A user who generates 10 images daily has integrated your product into their workflow. A user who logs in and browses is still deciding. Generation volume is the truest signal of product-market fit.
What percentage of users come back within 7 days? Within 30 days? This metric separates products with genuine utility from those that attract curiosity-driven one-time visitors. A high return rate means people find ongoing value. A low return rate means your product is a novelty, not a tool.
Average revenue per user tells you whether your pricing strategy is working. Revenue per GPU tells you whether your infrastructure investment is paying for itself. Both numbers should trend upward as you optimize pricing and inference efficiency.
Total signups. A large signup number means nothing if those users never generate a single image. I would rather have 1,000 users who generate daily than 100,000 who signed up and forgot. Focus on depth, not breadth, especially in the early stages of an AI product.
Cost per generation, free-to-paid conversion, daily active generations, return user rate, ARPU, and queue wait time. Cost per generation is the most critical.
2-5% free-to-paid is typical with generous free tiers. Quality of converted users matters more than the raw percentage.
Daily active generation count and return user rate — not vanity metrics like total signups.