---
title: Baseline Cost per Workflow — Step 1 of the Cromus 4-Step Framework
description: Baseline Cost per Workflow is the fully-loaded cost of running one AI-assisted process — including API costs, human review, error correction, and overhead. Step 1 of the Cromus measurement framework.
canonical: https://cromus.ai/baseline-cost-per-workflow
source_html: https://cromus.ai/baseline-cost-per-workflow
---

# Baseline Cost per Workflow

> **Step 1 of the 4-step AI workflow measurement framework.**

Baseline Cost per Workflow is the **fully-loaded cost** of running one AI-assisted process from start to finish. It goes beyond raw API spend to account for every cost component a team actually bears.

---

## What "fully-loaded" means

Most teams track only the API invoice. Baseline Cost per Workflow captures the complete picture:

| Cost component | What it includes |
|---------------|-----------------|
| **API cost** | Token spend across all model calls in the workflow |
| **Human review** | Time spent by people reviewing, correcting, or approving AI outputs |
| **Error correction** | Cost of retries, fallbacks, and failed runs (expected failure rate × correction cost) |
| **Orchestration overhead** | Compute time for the orchestration layer, state management, and tool calls |
| **Latency cost** | Time-to-completion for time-sensitive workflows where delay has a dollar value |

---

## How Cromus computes it

From a compiled SKILL.md, Cromus calculates Baseline Cost per Workflow deterministically:

1. Sum token estimates per step (input + output) × per-model unit cost from the verified registry
2. Apply the declared failure rate and retry policy from ETHOS.md to add expected error correction cost
3. Apply the step dependency graph to compute end-to-end latency
4. Surface per-run and monthly (100-runs/day) projections for each cost-quality mode

The computation uses the **verified model registry** (62+ models, weekly price checks) — not published list prices, which are frequently stale.

---

## Example output

```
Workflow: Customer Support Triage SOP
Cost-quality mode: Balanced

Per-run cost: $0.031
  → API (3 steps): $0.019
  → Expected retry (8% failure rate, 1 retry): $0.008
  → Tool call overhead: $0.004

Monthly (100 runs/day): $93.00
```

---

## Why it matters before execution

Without a baseline, teams have no way to:
- Budget for AI workflow costs accurately
- Identify which workflows justify investment in optimization
- Measure the ROI of switching models or restructuring steps

Baseline Cost per Workflow is the input to Croms™ analysis: you cannot measure *preventable* waste without first knowing what you're actually spending.

---

## Part of the 4-step framework

| Step | Name | What it produces |
|------|------|-----------------|
| **0** | Workflow Classification | Tier map of AI touchpoints |
| **1** | **Baseline Cost per Workflow** | Fully-loaded per-run cost |
| **2** | Croms Analysis | Preventable waste score |
| **3** | Total Cost of Workflow Ownership (TCWO) | True per-outcome cost over time |

---

## Related pages

- [Workflow Classification →](/workflow-classification)
- [Croms™ — preventable AI workflow waste →](/croms)
- [Total Cost of Workflow Ownership →](/total-cost-of-workflow-ownership)
- [Interactive Demo →](/demo)