As AI usage grows across companies, we are faced with a significant issue of predicting pricing and profitability. In traditional SaaS, the cost of delivering software was predictable with compute and operational costs. With AI, that model has changed. Compared to gross margins of >80% for traditional SaaS companies, AI product margins can be lower than 60%. Invoices come in, spreadsheets are updated quarterly or monthly, with variability in the margins due to the changing AI costs. We created Bear Lumen to help teams make pricing decisions based on data instead of guesswork.
The AI Pricing Dilemma
When we began researching pricing for AI products with companies, we noticed a common pattern. Operational costs were rising sharply, yet no one had a clear picture of their gross margins on a per-customer basis. Traditional SaaS tools track infrastructure at the resource level, such as VMs and storage. However, AI products bring a whole new set of challenges:
- Inaccurate forecasting: Recent industry research shows that 85% of companies miss their AI related cost forecasts by more than 10%.
- Eroding gross margins: Companies report eroding gross margins of more than 6% from AI costs.
- More models, more moving parts: Businesses are rapidly increasing usage. Enterprise token usage has skyrocketed, and the average business now manages multiple different AI models at the same time. A misconfigured prompt, a faulty loop, an inefficient query, or a silent model or pricing change on the provider’s side can spike your bill overnight. The difference is catching it in hours, not weeks.
- AI cost is not just token usage: AI costs can come from data platform usage, network access to models, and LLM tokens, making it harder to forecast and control AI spend.
Most engineering teams try to cope with this complexity by updating spreadsheets weeks after the invoices arrive. This makes it nearly impossible to spot problematic accounts in real time.
Closing the Gap Between Cost and Revenue
We noticed a gap in the modern software stack. Current observability and billing tools manage the transactional part of a sale, but they struggle with complex pricing operations. They do not reconcile fluctuating provider costs with customer revenue, track sudden cost changes, or suggest better pricing options.
We founded Bear Lumen to fill this gap. Our goal was to create a platform that automatically tracks, attributes, and measures AI spending at the detailed workflow level. By using the Bear Lumen SDK to wrap AI workflows, teams can quickly capture compute and token volumes from major providers like OpenAI, Anthropic, Google Gemini, among others. Combining this data with revenue streams in tools such as Stripe allows you to see contribution margins broken down by specific customers, features, or models in real-time.
Pricing with Data, Not Guesswork
We want to help companies shift from guessing to making decisions based on data. We built our Pricing Intelligence engine so you can run historical backtesting, create profitable hybrid tiers, and understand how switching to cheaper open-source models affects your bottom line before you deploy code. You shouldn’t have to wait until the end of the month to find out if your AI product is actually making money.
Own Your AI Unit Economics
Stop struggling with delayed spreadsheets and unclear API invoices. You can see your exact unit economics, identify unprofitable accounts, and build sustainable pricing models in real time.