Tactical AI Audit

Matthew Arthurs · Be A Decisive Leader

U.S. Army LTC · PMI AI (CPMAI) · PMP · PMI-ACP · LSSBB · MBA · MSIT

EXECUTION ONLY Anti–analysis paralysis · Next right move

BLUF: Inventory shadow AI, classify work by decision doors, then name one jump-master per critical workflow—before you scale spend or liability.

PHASE 1 The SITREP

Audit shadow AI tools & flow metrics—what ships without governance visibility.

✓	Row item	Capture
	Shadow tools register	List every AI/automation touchpoint in production (approved or not): vendor, owner, data classes touched.
	Flow metric — cycle time	Baseline hours/decisions from trigger → customer-visible outcome for top 3 AI-assisted workflows.
	Flow metric — defect escape	Count rework / compliance exceptions tied to AI outputs (last 30 days).
	Single source of truth	Where does human authority override model output today? Document the escalation path in one sentence.

Categorize tasks as Single-Door vs. Double-Door—automate only after the door math is honest.

✓	Door type	Rule	Your examples (fill)
	Single-Door	Reversible or cheap to unwind; bounded downside—candidate for assisted automation with monitoring.
	Double-Door	Irreversible or asymmetric risk (legal, safety, brand, capital)—human judgment + explicit sign-off.

Name one workflow you will stop AI-augmenting until double-door controls exist.
Name one single-door workflow where speed wins if metrics hold for two sprint cycles.

Assign human-in-the-loop accountability—one named owner per critical AI-assisted jump.

✓	Jump (workflow / decision)	Jump-master (name + role)	Proof they own (metric or artifact)

Jump-master ≠ committee. One throat to choke per row.
If you cannot name the jump-master, the workflow is still classified as shadow ops.