Radical Geek field guide
Radical Geek's Guide to Measuring AI Adoption And Impact
A practical guide to accountable outcomes, engineering-capacity evidence, AI attribution, RAG impact and full human-machine economics.
Why this guide
Most organisations begin with the numbers that are easy to obtain: licences, training sessions, prompt counts and tool usage. Those figures show activity. Operational impact appears in changed workflows, accepted outcomes, sustainable verification effort and credible economics.
Version 2026.08.03 follows each work item from operational qualification through execution and review to an accountable outcome. It adds complexity-aware engineering work classes and a whole-system gate for establishing capacity movement, followed by counterfactual comparison designs and observed/associated/attributable grades for AI’s contribution. It also measures the six-link RAG impact chain — authoritative evidence available, required evidence retrieved, answer supported, answer accepted, workflow improved, accountable outcome changed — alongside consequence, delegability, full human-machine economics, tail reporting and stopping rules.
It is part of the Radical Geek guide system, providing the evidence and economics layer for the same governed work-class lifecycle.
Establish engineering-capacity movement with comparable demand and whole-system evidence, then grade how confidently the movement can be attributed to AI.
What you will leave with
A guide built to be used.
- 01
Measure each AI-supported work item through to an accepted, accountable business outcome.
- 02
Establish sustained engineering-capacity movement using comparable demand and a whole-system claim gate.
- 03
Grade AI's contribution as observed, associated or attributable using a credible comparison design.
- 04
Classify consequence and delegability separately before expanding a governed work class.
- 05
Measure the six-link RAG impact chain from authoritative evidence through retrieval, support, acceptance and workflow improvement to accountable outcome.
- 06
Combine model, token and retrieval cost with human verification, rework, exceptions, elapsed time, support and risk.
- 07
Use Proof of Operation, tail metrics and stopping rules before expanding delegation.
Inside the guide
The working ground it covers.
- Accountable outcomes, baselines and Proof of Operation
- Engineering-capacity evidence, counterfactual designs and AI-attribution grades
- Workflow impact, exception concentration and knowledge cooperation
- Quality, risk, shadow context and retrieval-augmented work
- Consequence, delegability and correct abstention
- Work-item ledgers, tail-aware economics and stopping rules
- Token economics and practical scorecards