The Foundry Method

The Foundry Method for defensible AI roadmap decisions

The Foundry Method is a six-step framework—Scan, Source, Evidence, Review, Prove, Build—combining evidence modeling, 52 operator patterns, Impact × Feasibility × Moat scoring, and explicit readiness gates.

1

Evidence enters through human review, not blind auto-acceptance.

2

Operator coverage reveals what opportunities are grounded versus underspecified.

3

Scoring balances impact, feasibility, moat, quality, risk, and time-to-value.

1. Scan

Start outside-in. Before interviews begin, dated public customer, channel, competitor, supplier, regulatory, hiring, investor, and industry evidence can seed a source-linked discovery agenda. Aggregate consumer and platform-policy sources support the same workflow for B2C companies. Read the full demand intelligence methodology.

2. Source

Add company websites, documents, spreadsheets, annual reports, interview evidence, and optional consent-first workflow capture. Machine-drafted facts remain proposals until a person reviews them.

3. Evidence

Build a structured model of workflows, pains, assets, constraints, relationships, and cited financial-report signals before ranking opportunities. A self-contained evidence and provenance map makes the reviewed model, open gaps, and opportunity support visible outside the app.

4. Review

52 operator patterns map evidence to candidate opportunities—from core automation and forecasting through financial-report signals, domain orchestration, physical and R&D plays, analytics and revenue, software estate, outside-in demand gaps, constraint triggers, and prerequisites. Operator coverage highlights which patterns fired and what evidence is missing. Candidates receive written Impact × Feasibility × Moat reasons, then split into quick wins, strategic bets, prerequisites, evidence gaps, and dropped generic ideas.

5. Prove

Selected opportunities become business cases, agent blueprints, eval suites, safety checks, and pilot plans. Fixed readiness gates return ready, needs work, or unsafe instead of allowing an LLM to approve its own output.

6. Build

After the required gates clear, export starter code and a production handoff package. Use Case Foundry prepares the package; the customer's technical owner supplies credentials, verifies integrations, and decides what goes live.

Ready to apply this to your own AI roadmap?

Use a sample workspace now, or contact us to discuss your assessment workflow.