AI is now a line item your board cares about. That's new, and it changes the standard of proof. A cost that used to sit inside "cloud infrastructure" is now large enough, and volatile enough, to warrant its own slide — and its own questions.
The board doesn't want a tour of your model stack. It wants defensible answers to four questions. Get all four right and AI spend becomes a governed line like any other. Get any of them wrong and it becomes the thing that gets flagged every quarter.
Here are the four questions, the answer each one needs, and the artifact that makes the answer defensible.
Question 1 — "How much are we spending on AI?"
This sounds like the easy one. It isn't, because the honest answer is usually a range, and a range is what makes boards nervous.
What the board wants is one number, trending, by month. Not "somewhere between $40K and $70K depending on the quarter." One figure for the current month, the same figure for the prior twelve, and a line that shows the trajectory. AI spend compounds faster than any other cloud category, so the trend line matters more than the point estimate — a flat number hides a curve that's about to bend.
The problem is that most organizations can't produce this cleanly. Spend is spread across direct provider invoices (Anthropic, OpenAI), per-seat tools (Copilot, Cursor), and embedded SaaS AI — each on its own billing cycle, each in its own currency of measurement. Stitching them into one monthly figure is a spreadsheet exercise that breaks the moment someone changes a provider.
A spend control plane produces this figure as a byproduct. Every call passes the gateway, and every call is recorded in the Token Ledger with its actual cost. One number, any period, always current — because it's a query against the ledger, not a reconciliation project.
Question 2 — "On what, and who?"
The second question is where a blended provider invoice fails completely. "We spent $58,000 on Anthropic in June" is not an answer a board can act on. It's an operations cost with no owner.
What the board wants is attribution: by team, by product, and by customer. Which business unit generated the spend. Which product feature. Which end customer, if you're an embedded-AI company whose margin depends on it. That's the difference between a cost you report and a cost you can manage.
This is exactly what the Token Ledger's five-dimension attribution is built for. Every governed key answers five questions before a token is spent — WHO (the person or agent), HOW MUCH (the budget), WHY (the project and task), WHERE (the location), WHEN (the timing). Because attribution happens at the moment of the call, not in an after-the-fact analysis, the roll-up is authoritative. The search team's spend is the sum of the calls tagged to the search team — traceable down to the individual request.
That attribution is what turns a provider invoice into a chargeback. Finance gets per-team, per-product, per-customer cost, reconciled to the invoice total, exported as a chargeback CSV they can post to the general ledger. The board stops hearing "we spent X on a vendor" and starts hearing "this product line spent X, and here's its margin."
Question 3 — "What's the return?"
A number without a denominator is just a bill. The board's third question is the one that decides whether AI spend is an investment or a leak: what are we getting for it?
The unit that matters here is cost per successful task, per feature, tied to the outcome. Not cost per token, not cost per API call — cost per unit of business value. If your support-triage feature costs $0.14 per resolved ticket and each resolved ticket saves twenty minutes of agent time, that's a defensible ROI story. If it costs $0.14 per ticket and half the calls are retries that resolve nothing, that's a different conversation — and one the board would rather have now than in the annual review.
Per-request and per-task cost attribution makes this computable. Because the ledger records cost against the project and task that generated each call, you can divide spend by the outcome that matters for that feature. The finance frame the board understands — cost per unit, trended over time — becomes available for a category that most companies still report as an undifferentiated lump.
Question 4 — "What stops it running away?"
The last question is the one that's actually about risk, and it's the one that keeps a CFO up at night. AI spend has a failure mode no other line item has: a single misconfigured agent can 100× the bill overnight, and the native provider dashboards show you the damage a day after it's gone.
The board wants to hear the control story: hard caps and anomaly alerts, so a rogue agent can't blow the quarter. Not "we monitor it closely." A mechanism.
A spend control plane is the mechanism. Every key carries a hard budget cap enforced at the gateway — when the budget is exhausted, the next call returns a 402 before it reaches the provider, not a line on next month's invoice. Spend-anomaly alerts fire the instant a call pattern deviates from plan, while the spend is happening, not after it posts. Every allocation and every denial is written to an append-only audit trail. That's the answer to "what stops it": the spend is pre-authorized against a budget before a token is spent, and the cap is a wall, not a warning.
The four-slide board deck
Four questions, four slides. Build them once and reuse the shape every quarter.
Slide 1 — The number. One headline figure for the current month. A twelve-month trend line behind it. A single annotation on the trajectory ("up 18% QoQ, driven by the new agent feature"). Nothing else.
Slide 2 — The attribution. A stacked bar or table: AI spend by business unit, with the top product and top customer called out. This is the slide that proves the number has owners. Sourced directly from the chargeback CSV.
Slide 3 — The return. For the two or three AI features that matter, cost per successful task next to the outcome it drives. This is the slide that reframes spend as investment.
Slide 4 — The controls. The caps that are in place, the budget headroom remaining, and the anomaly-alert posture. One line: "hard caps enforced at the gateway; no single workload can exceed its budget." This is the slide that lets the board stop worrying.
The mapping
| Board question | The artifact that answers it | |---|---| | How much are we spending? | Token Ledger monthly total, trended — one number per period | | On what, and who? | Five-dimension attribution → chargeback CSV (team / product / customer) | | What's the return? | Cost-per-successful-task, from per-request attribution tied to the outcome | | What stops it running away? | Hard gateway caps (402 before the call) + spend-anomaly alerts + append-only audit |
Generate it, don't assemble it
The reason most AI-spend board slides are late, hedged, or wrong is that they're assembled by hand — someone pulls three invoices, reconciles them in a spreadsheet, guesses at the team split, and hopes nobody asks a follow-up. That model doesn't survive a board that's paying attention.
The alternative is to generate all four artifacts from the ledger. The monthly number, the attribution, the per-task cost, and the control posture are all queries against the same authoritative record of every call — not a spreadsheet that goes stale the day after you build it. When a director asks "why did the search team's number jump in May," the drill-down is already there.
If you want to see the attribution and per-call cost recorded live, the playground runs real calls through the gateway and shows the ledger entry each one produces — the same record the four slides are built from.