LLM Cost & Token Economics: Pricing AI Features

What you'll build

What you will be able to do

Each lab ends with a concrete result that you build yourself and the checker confirms, cell by cell.

• Estimate the tokens of one Fernhill Assist support reply from word counts, then see how a six-turn chat multiplies the input tokens because the history is sent again every turn.
• Look up the input and output prices of five Fernhill Assist features from a model price table and calculate the cost per request, per 1,000 requests and per dollar.
• Build Fernhill Assist's monthly LLM bill from three usage tiers, and find out how much of it a small group of power users drives.
• Price Fernhill's internal policy bot with and without prompt caching, add a batch discount for requests that can wait, and find the cache hit rate at which caching starts to pay.
• Route a growing share of Fernhill Assist's reply drafts from the large to the small model, calculate the blended cost and quality of each split, pick the cheapest split that keeps quality high enough, and chart the trade-off.
• Price Fernhill Assist's help-centre answers built with retrieval-augmented generation: the one-off embedding of 4,000 articles, the context tokens of every question and the full monthly bill.
• Compare Fernhill's support desk with and without the AI assistant: human cost per ticket, AI cost per resolved ticket, the monthly saving and the resolution rate the AI needs to break even.
• Price Fernhill Assist's three add-on plans: cost per seat at five usage levels, the gross margin of every plan at every level, the usage limit that keeps a 70% margin, and a chart of margin vs usage.