How do I track cost per business outcome in Cognocient?

The X-Cost-Outcome header links each AI call to the business result it produced — a resolved ticket, a signed contract, a generated report. Cognocient aggregates these into cost-per-outcome metrics that answer the CFO's core question: what did we get for this AI spend?

Why cost per outcome changes the conversation

Without outcome tracking, you have AI spend. With it, you have AI ROI.

Without outcomes: "We spent $12,000 on AI last month. Is that a lot?" — No answer. You have a cost with no context.

With outcomes: "We spent $12,000 on AI and resolved 28,500 support tickets at $0.42 each. Our human agents cost $8.40 per ticket." — AI delivers 20× cost efficiency per ticket. ROI is obvious.

Quick setup — X-Cost-Outcome header

The fastest way to track outcomes is the X-Cost-Outcome attribution header. Add it alongside your existing X-Cost-Feature header:

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": prompt}],
    extra_headers={
        "X-Cost-Feature":  "ticket-resolver",
        "X-Cost-Outcome":  "ticket-resolved",   # ← outcome tag
        "X-Cost-User":     f"user_{user_id}",
    }
)

The X-Cost-Outcome tag feeds simple call-level analytics (visible alongside your other attribution tags), separate from the outcome-definition system in Step 1 below. If you want the dedicated cost-per-outcome dashboard, define an outcome and record events for it as described next.

Step 1 — Define an outcome

Go to Executive → Outcomes → Define Outcome. An outcome definition has four fields:

Name — What you're measuring. Examples: "Support ticket resolved," "Contract drafted," "Document processed."

Description (optional) — Free-text notes for your team.

Correlation Method — By Session or By User. This determines what correlation_key you'll send when recording an event: a session ID or a user ID.

Window (minutes) — How far back Cognocient looks to match AI calls to the outcome event, correlated on the matching session or user tag. Default 30 minutes.

Step 2 — Record outcome events

Call this endpoint from your application when a positive outcome occurs — ticket resolved, contract signed, document approved. Authenticate with your Cognocient proxy key.

import httpx
 
COGNOCIENT_KEY = "sk-cog-YOUR-PROXY-KEY"
 
def mark_ticket_resolved(session_id: str, ticket_id: str):
    """Call this when a support ticket is marked resolved in your ticketing system."""
    httpx.post(
        "https://api.cognocient.com/api/outcomes/events",
        headers={"Authorization": f"Bearer {COGNOCIENT_KEY}"},
        json={
            "outcome_name": "Support ticket resolved",  # must match a defined outcome
            "correlation_key": session_id,               # session or user ID, per the outcome's Correlation Method
            "metadata": {
                "ticket_id": ticket_id,
                "resolution_time_minutes": 4,
            },
        },
    )

import fetch from 'node-fetch'
 
async function markContractDrafted(sessionId: string, contractId: string) {
  await fetch('https://api.cognocient.com/api/outcomes/events', {
    method: 'POST',
    headers: {
      'Authorization': `Bearer ${process.env.COGNOCIENT_KEY}`,
      'Content-Type': 'application/json',
    },
    body: JSON.stringify({
      outcome_name: 'Contract drafted',
      correlation_key: sessionId,
      metadata: {
        contract_id: contractId,
        contract_value_usd: 45000,
      },
    }),
  })
}

The endpoint returns 202 Accepted. outcome_name must match an outcome you've already defined in Step 1 — it isn't created implicitly on first event.

Step 3 — Reading outcome data in the dashboard

The Outcomes list (Executive → Outcomes) shows your outcome definitions — name, correlation method, window, and status. Click into one to see its metrics for the selected period:

There's no success-rate, cost-per-failed-outcome, or trend line on this view today — it reports cost against recorded events only, not against a denominator of attempted-but-abandoned sessions.

Industry benchmarks for cost per outcome

Use these benchmarks to calibrate your own metrics and frame the ROI conversation for your board:

Benchmarks are illustrative ranges. Actual costs vary by model, prompt complexity, and task definition. Your Cognocient data will give you the precise figure for your implementation.