Web Analytics Made Easy - Statcounter
Skip to content

Cost Intelligence

Cost intelligence explains where AI spend goes and whether that spend produces useful outcomes. It connects token usage, model cost, latency, tool calls, evals, users, workflows, and business KPIs.

What Cost Intelligence Does

Anosys can attribute cost by model, prompt, agent, tool, user, workflow, team, product area, customer, and business outcome. It can also detect anomalies and connect spikes to releases, retries, provider behavior, or user activity.

When To Use It

Use cost intelligence when:

  • Token spend is growing faster than usage.
  • Model escalations or retries are expensive.
  • Teams need cost per workflow or customer outcome.
  • You need to optimize model choice without reducing quality.
  • Finance or leadership needs explainable AI cost reporting.

Prerequisites

  1. Capture token usage and model names.
  2. Add dimensions such as user, team, workflow, customer tier, and business outcome.
  3. Connect evals if you want cost-versus-quality views.

Step-By-Step Setup

  1. Instrument model and agent calls.
  2. Confirm token usage and model fields appear in traces.
  3. Add cost widgets to a dashboard.
  4. Group by model, workflow, user, team, or outcome.
  5. Create cost anomaly alerts.
  6. Use pipelines to send weekly cost summaries.

Example Event Fields

1
2
3
4
5
6
7
8
9
{
  "workflow": "checkout-agent",
  "model": "gpt-4o-mini",
  "input_tokens": 1842,
  "output_tokens": 420,
  "cost_usd": 0.042,
  "user_tier": "enterprise",
  "business_outcome": "checkout_completed"
}

What Appears In The Console

You will see token usage, cost by model, cost by workflow, cost anomalies, model escalation tables, token-versus-latency views, and drill-downs to the traces behind spend.

Common Mistakes

  • Tracking cost only by model name.
  • Ignoring retries, tool calls, and escalations.
  • Reducing cost without checking eval score or customer outcome.
  • Missing user or workflow attribution fields.