Pre-Execution Economics: Governing the Spend That Hasn't Happened Yet

Every AI cost dashboard on the market answers one question: what did this cost? The invoice arrives, the chart renders, and finance learns about the overrun after the overrun. That is post-execution accounting, and it is where most of the category still lives.

The harder question — the one an enterprise actually needs answered — comes before any of it: should this workload happen at all, and what does it put at stake if it does?

This week PromptKing shipped both halves of that answer. We call it pre-execution economics: not post-execution control.

Half one: workload feasibility, before the spend

AI Capacity Forecasting takes the trailing thirty days of recorded consumption per seat and projects it forward. Every seat in the fleet now carries a forward position: daily burn, projected thirty-day consumption, a projected capacity gap against its plan, and a sprint completion risk band — will this seat reach its capacity ceiling before the work cycle completes?

Three things about how it's built matter more than the feature itself.

First, it computes from real recorded columns. No new tables, no shadow data model — the projection reads the same usage records the billing reconciliation reads.

Second, the capacity denominators are explicit. Where a seat's plan capacity is recorded, that number wins. Where it isn't, the model falls back to PromptKing's documented per-vendor denominator conventions — and every convention-based figure is labeled as one, on the page, because a planning convention presented as a vendor commitment is how forecasting tools lose trust.

Third, when a vendor's capacity unit isn't comparable — request allowances, interaction quotas, resource units — the model says "capacity unavailable" instead of inventing comparability. A forecast that refuses to fabricate a number is worth more than one that always produces one.

Half two: blast radius, quantified

Knowing a workload is feasible tells you it can run. It doesn't tell you what's exposed if it runs badly.

That's the second ship: every governed trajectory now carries a blast radius score — a 0-to-100 measure of the projected downstream impact if a flagged trajectory were allowed to continue. Decision systems need three inputs: cost, risk, and impact. PromptKing has had cost since day one and risk since the policy engine shipped. Blast radius closes impact.

The methodology is published on the score itself. Five weighted factors, all computed from signals already in the governance record: cycle waste on the trajectory, budget exposure against the organization's own soft and hard caps, containment exposure, archetype risk, and loop intensity. Every factor renders with its subscore, its weight, and a plain-English line explaining what it saw. If a CFO asks why a trajectory scored 37.5, the answer is on the screen — not in a support ticket.

One more thing, because honesty is a feature: our first scored trajectory is a seeded governance scenario, built deliberately to demonstrate cycle detection and containment end to end. It says so, in orange, directly on the page — and it is scored by the identical methodology that scores live trajectories. If a governance vendor won't label its own demo data, ask what else it won't label.

Why this pairing matters

Put the two halves together and the shape of the platform changes. Before an agentic workload executes, PromptKing can now answer: is there capacity for this? What does it put at stake? What policy applies? And — through the governance consultation tools that shipped alongside — an agent can ask those questions itself, through a standard interface, before spending a token.

Simulation is the control plane. The circuit breaker is the safety net. Pre-execution economics is what runs on top of both: the ability to shape which trajectories are even possible, instead of narrating the ones that already happened.

Post-execution accounting will always have a place. But the organizations that get ahead of AI spend won't be the ones with the best charts of last month. They'll be the ones whose governance layer answered the question before the meter started running.

That layer is live at promptking32.com.