> ## Documentation Index > Fetch the complete documentation index at: https://docs.getvector.me/llms.txt > Use this file to discover all available pages before exploring further. # Unit economics and pricing > Work out what it really costs you to serve a customer, see the standard cost-per-unit metrics, and compare pricing strategies for your own AI product. **This area is still in development.** The product marks Modeling as **WIP**: screens, labels and behavior described here can change, and some steps may not work yet. Two questions sit under almost every pricing decision: what does it actually cost you to serve a customer, and which pricing model turns that cost into a healthy business? Vector's unit economics tools are built to answer both for your own product, using the assumptions in your model. ## Count the full cost of serving AI Serving an AI feature costs more than the inference call itself. Alongside direct inference — transcription and LLM tokens, for example — your model can account for a **harness markup**: the surrounding AI infrastructure a real product needs, such as vector storage, embeddings, agent orchestration, caching, and logging or tracing. Without it, cost of goods sold understates what serving a customer actually costs and margin looks better than it is. The markup is configurable — turn it on or off, and adjust an overall percentage or its individual components — and it appears as its own line in the COGS breakdown and unit-economics table, not folded into inference cost. Change it and gross margin and per-account contribution recalculate immediately. Turned off, the numbers match plain inference cost, so nothing breaks if you don't use it. ## Read what each unit costs you Once your AI-COGS is complete, Vector expresses it in standard cost-per-unit terms: cost per processed item (per meeting or per query, for example), cost per paid seat per month, and the share of revenue AI-COGS consumes (or the inverse — how much of what you spend converts into revenue). These sit in your KPI summary and the model's dependency graph, where you can see what each one is built from. The vocabulary follows the FOCUS 1.4 billing standard, so the numbers line up with what you'd see from an [AI provider](/getting-started/connect-a-provider)'s own billing or a FinOps report. Months with zero volume don't break these metrics; they show as zero or not applicable rather than an error. ## Compare pricing strategies Beyond per-seat, freemium, and usage-based pricing, Vector's strategy catalog adds a hybrid of a base fee plus usage, tiered plans built around an AI quota, and feature-gated add-ons. Apply any strategy and gross margin and lifetime value per plan recalculate automatically. Apply two strategies to the same assumptions and compare them side by side — same funnel, same costs, different pricing mechanics — to see which one produces healthier margin and LTV for your product. When a quota-based strategy is active and a customer's usage goes past their included allowance, the overage is reflected in revenue and cost according to that strategy's rules. ## Design a credit-based tier If your pricing model is credit-based, design it directly from **Design a credit-based tier**, at the bottom of the pricing area: a catalog of actions with a cost and a price in credits, an included allowance, a price per credit, and an overage rule. Once turned on, it's a real input to your forecast, not an isolated preview — the allowance, overage, and how usage concentrates across customers (a small share of heavy users driving disproportionate consumption) all feed into projected revenue, AI-COGS, and margin. Change the price per credit or the allowance and the forecast recalculates; turn the tier off and the forecast behaves as it did without it. ## Test ways to cut inference cost Your model can also include levers for the engineering choices that affect inference cost: the share of requests served from a prompt cache, the share processed in batch, the mix of higher-cost "frontier" models versus cheaper "mini" models, and your input-to-output token ratio. Adjusting any of these recalculates AI-COGS and margin, so you can see the effect of a decision — like shifting traffic to a cheaper model — before committing to it. ## Where to go next The actuals that ground these numbers. Turning the conclusions into work.