01 / operating architecture

Build the decision system—not another layer of software.

Operating architecture is the connected evidence, workflow, interface, control, and ownership required to make one recurring decision reliable.

02 / order of operations

AI belongs at the end of the operating sequence.

A model cannot repair an undefined metric, missing ownership, or a broken handoff. Build the deterministic operating loop first. Add probabilistic capability only where it earns its place under evaluation.

  1. 01Decision
  2. 02Definitions
  3. 03Data
  4. 04Workflow
  5. 05Interface
  6. 06AI
03 / system anatomy

Each failure has a place to be found.

01

Decision

The accountable owner, operating cadence, required inputs, acceptable delay, and action that follows.

02

Evidence

Source records, shared definitions, quality rules, lineage, and refresh timing that make a conclusion traceable.

03

Workflow

The handoffs, approvals, exceptions, and escalation paths through which evidence becomes action.

04

Interface

The report, alert, or application through which a responsible person reviews and acts.

05

Control

Access, tests, monitoring, auditability, fallback, and human review carried with the system.

06 / optional

AI, when justified

Bounded retrieval, classification, summarization, or recommendation added only after a defined evaluation passes.

04 / delivery method

Make the smallest durable intervention.

The method begins with a named decision and ends with a client-owned operating asset. It does not presume a platform, model, or company-wide transformation.

  1. 01
    FrameName one recurring decision, its owner, cadence, inputs, acceptable delay, and current operating baseline.
  2. 02
    TraceFollow definitions, records, transformations, handoffs, and exceptions from source to action.
  3. 03
    GovernSet ownership, access, quality thresholds, review requirements, and acceptable use before capability expands.
  4. 04
    BuildConnect the minimum durable data, workflow, and interface required to make the decision reliable.
  5. 05
    Prove and transferTest known cases, measure the agreed after-state, document the system, and place routine operation with the client.
05 / design principles

Controls travel with capability.

I

Begin with the decision

A tool has no operational value until a responsible person, recurring choice, required evidence, and resulting action are clear.

II

Define before connecting

Shared names are not shared meanings. Reconcile the grain, owner, time boundary, and source of every consequential metric.

III

Keep deterministic work deterministic

Arithmetic, eligibility rules, reconciliations, and hard controls should remain explicit, testable, and reproducible.

IV

Make AI earn its place

Use AI only for a bounded task with a known evaluation, documented fallback, accountable reviewer, and measurable benefit.

V

Carry controls with capability

Access, lineage, tests, monitoring, exceptions, and review belong inside the design, not in a later assurance phase.

VI

Transfer the system

Durable value remains legible and operable after the consultant leaves. Documentation and handoff are part of the build.

Diagnose one decision before funding the build.

See the diagnostic