What is Workflow Classification?

Workflow Classification maps AI touchpoints by complexity across four model tiers — Mythos/Capybara, Frontier, Balanced, and Lightweight — ensuring the right model handles every task.

What percentage of AI tasks need frontier models?

Typically only 30-40% of AI touchpoints require Frontier or Mythos-tier models. The remaining 60-70% can use Balanced or Lightweight models, saving 40-60% on API costs.

AI Workflow Classification Framework — Cromus

Workflow Classification is Step 0 in Cromus's 4-step measurement framework. It maps your AI touchpoints by complexity to determine which model tier should handle each task — preventing the common mistake of using expensive frontier models for simple tasks.

The four classification tiers: Mythos/Capybara tier (next-generation models for the most complex, nuanced tasks requiring state-of-the-art reasoning), Frontier tier (GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro for complex multi-step reasoning, creative generation, and expert-level analysis), Balanced tier (Claude Sonnet 4.6, Gemini 3 Pro, GPT-5.4 Mini for standard business tasks with good cost-quality tradeoffs), and Lightweight tier (GPT-5.4 Nano, Claude Haiku 4.5, Gemini 3 Flash for simple classification, extraction, routing, and formatting tasks).

Classification criteria include: Task complexity (single-step vs. multi-step reasoning), Output sensitivity (how much quality matters for the downstream process), Latency requirements (real-time vs. batch tolerance), Volume (how many executions per day/month), and Cost sensitivity (budget constraints per workflow).

Proper classification typically reveals that 60-70% of AI touchpoints in an organization can be handled by Balanced or Lightweight tier models — saving 40-60% on API costs without measurable quality loss.

This page connects to: Baseline Cost (Step 1), Croms (Step 2), and TCWO (Step 3) in the Cromus measurement framework.