90-day roadmap · Four phases, thirteen weeks
Set standards, build consensus, and redesign one real execution unit.
Ninety days does not promise a complete transformation. It completes one cycle: standards for recurring judgments, shared language among core members, and at least one redesigned unit with evidence from both success and failure.
Designed scopeDesigned for up to 20 core members. The accountable leader must personally run the Feynman checks.
Larger organizationsKeep the sequence and reduce the scope: choose one division or independent business unit.
PrerequisiteThe accountable leader must reserve two to three hours a day. If the calendar has no room, do not start.
- Use spending records to identify the limiting resource: cash, judgment, or consensus
- Count recurring judgments from the past month and select the three most frequent
- Locate paying users on the adoption curve and identify the current product path
- Deliverable
- One-page current-state map
- Most common failure
- Turning the audit into a rally before any usable standard exists.
- 02
Weeks 3–6
Start the standards engine
- Have the accountable leader draft eight-layer standards v1 as condition + action + acceptance
- Create 3-, 6-, or 9–12-month ROI cards appropriate to the organization's stage
- Encode the three highest-frequency judgments and measure median adoption
- Deliverable
- Eight-layer standards v1 + ROI cards
- Most common failure
- Optimizing wording or replacing first-author accountability with a vote.
- 03
Weeks 7–10
Run the consensus engine
- Name key standards and place them in context every member can retrieve
- Run individual Feynman checks: explain plainly, answer boundaries, and guide a decision
- Freeze a forward or consensus-reverse product path and install five development gates
- Deliverable
- Core-member Feynman checks + product-path decision
- Most common failure
- Mistaking an email or training session for shared understanding.
- 04
Weeks 11–13
Run two pilots
- Choose a small, real, painful execution unit for AI or outsourcing redesign
- Run a single-variable product iteration with a stable core and rollback plan
- Freeze stop conditions before launch and close with evidence from success and failure
- Deliverable
- Two pilot reviews + standards revision
- Most common failure
- Continuing through unexplained quality loss or moving acceptance criteria after failure.
Three boundaries
The first cycle is not the whole transformation, and a roadmap is not a guarantee.
- Most organizations need three to five cycles before new defaults stabilize.
- Every pilot freezes stop conditions before launch; failures and successes both enter the standards library.
- Evidence comes mainly from early-stage companies and mechanism comparisons, not thousand-person stress tests.
