One decision pathway. Many decision jobs.

The next dollar goes to whichever channel claims the most credit, not to where it would create incremental profit.

Platform-reported ROAS, spreadsheet reallocations, and monthly meetings that relitigate attribution.

Counterfactual comparison of allocation moves — including no change — validated against margin floors and shift ceilings, routed to the budget owner.

Allocation choices argued from shared evidence, restraint recorded when doing nothing wins, and a scored prediction on every move.

Media lead within the delegated cap; Finance jointly above it.

"The creative is tired" is the default explanation for every dip — and refreshes are expensive when the real cause is elsewhere.

Eyeballing CTR curves; rotating creative on schedule or on instinct.

Creative fatigue tested as one hypothesis among several — against engagement stability, audience overlap, and experience telemetry — before any rotation is recommended.

Rotations happen when the evidence says fatigue, not when the calendar does; wrong-lever interventions decline.

Creative or brand owner approves rotations; automatic within a standing rotation grant.

Each platform optimizes itself; nobody optimizes across them with margin and inventory in view.

Per-platform automated bidding plus manual cross-channel judgment calls.

Cross-channel moves evaluated on one evidence base, with in-platform automation kept in place and bounded by organizational policy.

Channel shifts that respect the whole system — economics, stock, and site experience — with the reasoning preserved.

Media lead; escalation to Finance when shifts touch committed budgets.

Problems are found at the weekly review — days of spend after they started.

Dashboards checked when someone has time; alert fatigue from ungoverned notifications.

Continuous SENSE with qualified signals only: movements are diagnosed and arrive as evidence-backed decisions, not raw alarms.

Attention lands where evidence says it should; the noise envelope absorbs the rest, and the record shows both.

None to observe; the acting stages carry their own authority map.

Media buys revenue that the margin math quietly loses — discovered at month close, if at all.

Finance reviews after the fact; marketing optimizes to revenue targets it can see.

Margin floors are policy inside VALIDATE: any move that would breach them is vetoed before it becomes a recommendation, with the check recorded.

Margin protection stops depending on someone catching it; the veto record shows finance the rail held.

Finance owns the floor; overrides require the finance owner by name.

Campaigns accelerate demand into products that cannot be fulfilled — buying stockouts and disappointed customers.

Manual coordination between media and merchandising, when calendars align.

Inventory position is operating context: constrained SKUs shape PLAN's candidates and VALIDATE's limits automatically.

Spend leans into stock that can sell and away from stock that cannot — as a system property, not a meeting outcome.

Merchandising or operations owner for constrained-SKU exceptions.

Promotions launch on instinct, and nobody can later say whether the lift paid for the discount.

Calendar-driven promotions with retrospective, contested readouts.

Each promotion candidate is planned with a stated expected effect and a no-promotion baseline, then scored against its own prediction in LEARN.

A compounding record of which promotion mechanics work under which conditions — argued from outcomes, not memory.

Commercial owner; Finance where discount depth crosses the configured floor.

Conversion slips and four teams each blame a different cause — creative, audience, site, price.

Parallel investigations in separate tools; the loudest theory wins.

All candidate causes tested on one evidence base in REASON, with eliminated hypotheses preserved so the argument ends.

One diagnosis, owned by evidence; the fix lands on the actual cause the first time more often.

Owner of the diagnosed surface — site, media, or merchandising — authorizes the fix.

Commercial incidents — tracking breaks, latency spikes, feed failures — reach the right owner late, via whoever noticed.

Ad-hoc escalation through chat; severity judged by volume of complaints.

The diagnosis names the owning function, and DECIDE routes there directly — with the evidence attached and severity grounded in commercial impact.

Incidents arrive diagnosed, at the right desk, with the business case for urgency already made.

Owning function acknowledges; remediation actions carry their own authority map.

Fix queues are ordered by recency and noise, not by what the delay actually costs.

Ticket triage by gut feel; commercial impact estimated, if at all, by the requester.

Each queued item carries the graph's evidence of commercial impact, so prioritization is argued from the same record as everything else.

Engineering and operations effort flows toward the delays that cost the most, visibly and defensibly.

Queue owner; reprioritization above a scope threshold routes to the operations lead.

Decisions that span media, site, inventory, and finance die in the handoffs between them.

Standing meetings and shared documents that go stale the day after.

One decision record spans every function: each owner sees the same evidence, applies their own constraints, and signs their own stage.

The handoff is the record itself — nothing is re-explained, and accountability is explicit at every step.

Each function approves its own constraint domain; joint authorization where scopes overlap.

Forecasts arrive as numbers without a track record — nobody knows how much to trust them.

Confidence by seniority of the presenter.

Every prediction the system makes is scored against its realized outcome, so confidence is a measured property with history — not a feeling.

Leadership weighs recommendations by demonstrated calibration, and sees calibration improve — or not — in the open.

None — this is a property of the record; consuming it requires no authority.

"Why did we do that?" has no answer six months later — the reasoning left with the people and the slide decks.

Institutional memory, email archaeology, and reconstruction after the fact.

Replay: any decision re-walked end to end — evidence as it stood, alternatives, checks, authority, action, outcome.

Reviews, audits, and post-mortems argue from the record instead of memory; lessons survive personnel changes.

Read access governed by role; replay changes nothing and needs no action authority.

Headcount and budget follow the best storyteller, because outcome evidence is scattered and contested.

Annual planning on assembled narratives.

Outcome memory shows which decision classes, channels, and interventions actually delivered against their predictions — the raw material for allocation.

Planning starts from scored history; the burden of proof shifts from rhetoric to record.

Executive owner of the planning cycle; the record itself needs none.

Strategy assumes capabilities and constraints that the operating reality quietly contradicts.

Top-down plans reconciled with reality once a year, painfully.

The operating context — real constraints, real capacity, real decision outcomes — is inspectable when strategy is set, and strategy's assumptions become policies the system then enforces.

Strategy and operations share one substrate: the plan constrains the decisions, and the decisions inform the next plan.

Executive sponsor sets the policies; changes are versioned and attributed.