Upfront, recurring, review, rework, integration, governance, and risk.
Original Framework / Research and Product Design
AI Economics Decision Framework
How the Running Before Crawling research became a transparent, recalculable operating tool instead of another static ROI slide.
The problem
Activity is not the same as an accepted business result.
AI business cases often lead with model price, licenses, prompts, or estimated hours saved. Those measures do not show whether the business received a usable, compliant, complete outcome—or what review, correction, integration, risk, and operating work was required to produce it.
The questionHow can an executive argument about AI sequencing become a transparent, recalculable operating tool rather than another static ROI slide?
The approach
Design for challenge, not persuasion.
Select a part of the system to see how the argument stays evidence-led, transparent, and maintainable.
Operating evidenceBuild the argument around what can be supported
The framework separates adoption from enterprise impact, technical activity from verified outcomes, and released capacity from realized cash savings. Externally sourced claims are mapped to a source ledger and explicit limitations.
Calculation modelConvert the thesis into three connected engines
AI TCO covers implementation, recurring operations, human review, rework, infrastructure, governance, and risk. Verified outcomes separate first-pass, reviewed, corrected, failed, and unresolved work. The business-case engine adds baseline comparison, ROI, payback, NPV, break-even, and capacity.
Transparent logicExpose every material assumption
The calculator labels evidence quality and returns “not comparable” or “not calculable” instead of zero. Capacity never becomes cash by default, expected loss is protected from double counting, and the tool never recommends headcount reduction.
Durable systemMake every result reproducible
Dense factual graphics become semantic HTML. Formula versions, dated pricing snapshots, pure calculation modules, and exported inputs make every result reproducible.
The framework
Three engines. One verified denominator.
First-pass, reviewed-pass, corrected, failed, and unresolved.
Hard ROI, risk-adjusted value, payback, NPV, break-even, and capacity.
Governing formulas
The arithmetic stays visible.
Both calculator modes use the same accepted-outcome denominator. Advanced adds cost composition without changing the comparison logic.
- 01Current cost per accepted outcome
current-process cost ÷ accepted current-process outcomes - 02Accepted AI-assisted outcomes
total AI-assisted attempts × accepted AI-assisted rate - 03AI cost per accepted outcome
total AI-assisted cost ÷ accepted AI-assisted outcomes - 04Equivalent-output AI cost
accepted current-process outcomes × AI cost per accepted outcome - 05Selected-period savings or loss
current-process cost − equivalent-output AI cost - 06Break-even accepted rate
AI cost per attempt ÷ current cost per accepted outcome
Illustrative worked example
Monthly ecommerce content production.
This is a teaching example—not a client engagement, benchmark, forecast, or claimed result. The accepted unit is one approved product-detail page (PDP).
- Current process
- 1,800 accepted PDPs at $126,000
- AI-assisted process
- 2,500 total attempts at a 78% accepted rate
- AI-assisted cost
- $82,000 for the same month
- Reasonable source
- PIM approvals, team and agency cost, generation logs, and a reviewed pilot
Current unit cost$126,000 ÷ 1,800 = $70.00
Accepted AI-assisted PDPs2,500 × 78% = 1,950
AI cost per accepted PDP$82,000 ÷ 1,950 = $42.05
Equivalent-output AI cost1,800 × $42.051282… = $75,692.31
Monthly modeled savings$126,000 − $75,692.31 = $50,307.69
Annualized decision view$50,307.69 × 12 = $603,692.31
Break-even accepted rate($82,000 ÷ 2,500) ÷ $70 = 46.9%
Quick versus Advanced
Use only the detail the decision needs.
Use five known planning values.
Choose Quick when you have accepted current volume, current cost, total AI attempts, an accepted rate, and an all-in AI cost for one consistent period.
Open the blank Quick EstimateBuild totals and show their composition.
Choose Advanced when finance or operations needs labor components, implementation allocation, direct or time-derived review, and explicit completeness notes.
Open the blank Advanced EstimateDeliverables
A decision system, not a single artifact.
- Board and C-suite decision framework
- Source ledger and evidence controls
- Reviewed visual system with factual qualification
- AI Cost Reality Calculator specification
- Five-input public estimate and advanced model-economics plan
- Formula registry, JSON data schema, and acceptance test vectors
- Static-site content, structured data, analytics, privacy, and release plan
Current status
Built, tested, and deliberately bounded.
Formula 1.1.0 is implemented as an isolated calculation engine and validated against acceptance vectors plus boundary tests. The interface remains static and private: calculations stay in the browser and are not submitted.
Interpretation and limitations
Decision support—not evidence of a client result.
The illustrative example shows how to read the model: it supports a more detailed pilot and operating review; it does not prove savings. This work demonstrates a research and product-design system and does not claim a client engagement, typical AI ROI, production adoption, or audited financial outcome. Results depend on the entered acceptance definition, completeness of cost, representative volume, and quality of evidence. The calculator is not accounting, legal, investment, employment, cybersecurity, or regulatory advice.
Research and decision tool
Read the argument. Challenge the economics.
Start with the primary paper, then use the calculator to name the accepted unit and make every assumption visible.