AI Cost Tracker

Origin & evolution

The need🫀NeedUserA user need, pain, desire, or constraintView reference → is specific to metered inference. A traditional web feature costs roughly the same to serve whether a user clicks once or a hundred times; an AI feature bills per token on every call, and a verbose prompt or a runaway agent loop can multiply that quietly. Practitioners now recommend token-level cost tracking that attributes spend to applications, users, and use cases, so a team can say a single AI-resolved support ticket🆘Support TicketCustomer SuccessCustomer support request or issueView reference → costs a known amount and set budgets and alerts per feature. The pressure is rising: API prices have been widely described as subsidised, with sizeable increases expected as vendors move toward sustainable margins, which makes per-feature cost a number worth watching before it moves.

How it works in practice

A product has two AI features: a cheap autocomplete and an expensive document summariser. The cost tracker tags every model call with its feature. After a month it shows autocomplete at £0.002 per use across 400,000 uses, and summarisation at £0.18 per use across 9,000 uses. Summarisation, the lower-traffic feature, is the larger line item. That single fact reroutes the roadmap🛣️RoadmapProduct SpecificationA strategic plan of features and milestonesView reference →: the team caps summary length and routes short documents to a smaller model, cutting that feature's cost by half without users noticing.

AI Cost Tracker vs. its neighbours

• AI Model🤖AI ModelAI & Machine LearningAn AI or ML model used within the productView reference → is what spends; the cost tracker is the meter on it. One model accrues cost across every feature and user that calls it, and the tracker is where those figures land.
• Cost Structure🧮Cost StructureBusiness ModelA cost category or structureView reference → is the product's full map of what it costs to run. The AI cost tracker is the component that feeds the AI portion into it, with the granularity that token billing demands.
• Unit Economics💹Unit EconomicsBusiness ModelPer-unit economic metrics (CAC, LTV, etc.)View reference → asks whether each unit of value earns more than it costs. The cost tracker supplies the cost half of that equation for AI-powered units, where the cost is variable and easy to underestimate.

In the graph

In the Unified Product Graph, an AI Cost Tracker sits in the AI region and bridges into the business region. The model it measures connects through ai_model_costed_by_ai_cost_trackerAI Modelcosted byAI Cost Trackerhierarchy, and its output flows upward through ai_cost_tracker_feeds_cost_structureAI Cost TrackerfeedsCost Structurecross-domain. That second edge is the point: it wires a technical observability artefact directly into the product's economics, so a model's token bill is never a fact stranded in an engineering dashboard🎛️DashboardData & AnalyticsAn analytics dashboardView reference →.