Track AI spend by feature, not just totals, to cut costs 30-40%

How granular cost tracking uncovered hidden AI waste across teams

When a vice president asked for a single number to quantify productivity gains from AI investments, most organizations drew a blank. After analyzing 200,000 AI requests in six months, a new tool now delivers per-feature, per-developer spend visibility and claims 30-40% savings by automatically routing simple tasks to cheaper models.

From cost dashboards to performance attribution

Early versions of the tool focused solely on expense tracking. The upgrade transformed it into a system that answers three critical questions:

• Which features consume the most AI resources?
• Which developers drive the highest AI-related costs?
• How does AI spending correlate with actual output?

Instead of vague statements like “We spent $4,200 on OpenAI,” teams now see granular breakdowns such as “Code review costs $340 per month, customer support consumes $89, and classification tasks tally $12.” This granularity shifts conversations from budget approvals to product decisions.

Smarter routing slashes expenses without code changes

The system introduces intelligent routing that automatically forwards simple prompts to lower-cost models. Teams previously sending every request to premium models reduced their OpenAI bills by 30-40% without modifying a single line of code.

The integration process remains minimal:

import { CostLens } from 'costlens';

const costlens = new CostLens({
  apiKey: 'cl_...'
});

const openai = costlens.wrapOpenAI(new OpenAI());

For coding assistants like Claude Code, Kiro, and Cursor, the setup requires only:

{
  "mcpServers": {
    "costlens": {
      "command": "npx",
      "args": ["-y", "@costlens/mcp-server"],
      "env": {
        "COSTLENS_MCP_KEY": "your_key"
      }
    }
  }
}

Productivity tracking ties spend directly to output

Beyond cost controls, the tool measures developer productivity through the Model Context Protocol (MCP). It logs sessions, code commits, pull requests, and time-to-delivery, then compiles weekly reports that CFOs or VPs can forward to justify AI investments.

During the six-month analysis:

• 52% of requests were simple tasks better suited for cheaper models.
• Code review emerged as the most expensive feature, three times higher than initially estimated.
• 15% of spending came from abandoned or looping sessions.

Teams leveraging the productivity tracking can now correlate AI spend with tangible engineering outcomes, moving beyond cost containment to strategic resource allocation.

Instant safeguards prevent runaway spending

A $50 weekend burn by a background agent prompted the introduction of a Slack kill switch. When spending velocity spikes, a real-time alert appears with a “Pause” button. One click halts the specific agent session for 30 minutes, automatically resuming afterward. This feature prevents accidental overruns without locking down entire teams.

Access options balance individual and team needs

The core SDK and MCP server are available at no cost for individual developers. Team plans, including advanced reporting, smart routing, and budget controls, start at $99 per month. The free tier enables evaluation and small-scale adoption while ensuring larger organizations can scale governance without friction.

As AI adoption accelerates, tools that deliver both cost visibility and productivity insights will define which investments pay off—and which simply drain budgets.