The questions your board will ask about AI.
They are predictable, there are about six of them, and most companies can currently answer two. Here they are, and here is where each number comes from.
What should a board pack say about AI spend?
Six things: total AI cost for the period and the trend, what it is as a share of revenue and of cost to serve, which parts of the business it sits in, what it returned, what next year costs under stated assumptions, and how much of the current spend could be saved. Camaze produces all six from attributed data, so the AI section is generated and then edited rather than assembled from several sources each quarter.
The AI slide is the one that gets pushed.
Every other section of the pack has a system behind it. Revenue comes from the CRM, headcount from the HRIS, cash from the ledger.
More detail
The AI section is assembled by hand from invoices, cloud bills and a spreadsheet somebody maintains, which is why it is thin and why it is late.
Then the question arrives that the slide does not answer, usually a version of what did we get for it, and the answer is qualitative in a room where nothing else is.
- A total assembled manually and rarely complete
- No breakdown by product or department
- No unit economics, so no answer on margin
- A forecast with no stated assumptions and no range
- No figure for how much of the spend could be saved
Illustrative product view. Figures are examples.
Six questions, six numbers, generated.
Each of the standard board questions maps to a specific output. Once the underlying data is attributed, the section assembles itself and the work becomes editing rather than authoring.
- Total and trend, complete across every provider, tool and self-hosted workload
- AI as a share of revenue and as a component of gross margin
- Split by product, department and customer segment
- Return measured per unit of work where the workflow supports it
- Forecast with a range and with the assumptions stated
- Identified savings, and how much has already been realized
What did it cost, and where does it sit.
A complete total, including the parts that usually go uncounted: cloud-billed model access, self-hosted GPU time and AI features inside existing SaaS contracts.
More detail
Then split by product and department.
The split is what stops the conversation from being about AI in general. Spend concentrates, and showing that two workflows account for most of it turns a vague concern into two specific decisions with owners.
Illustrative product view. Figures are examples.
What is it as a share of revenue.
For any company selling an AI-backed product this is the question that matters most, because AI is a variable cost that scales with usage and therefore compresses gross margin directly.
More detail
Reporting AI cost per account and per segment against revenue answers it, and it answers the follow-up too, which is whether the ratio is improving or deteriorating as you scale.
| Segment | Revenue per account | AI cost per account | AI as share of revenue |
|---|---|---|---|
| Enterprise, tier 1 | $18,400 | $2,180 | 11.8% |
| Enterprise, tier 2 | $9,200 | $1,640 | 17.8% |
| Mid-market | $3,100 | $291 | 9.4% |
| Growth | $980 | $142 | 14.5% |
| Self-serve | $79 | $31 | 39.2% |
Illustrative product view. Figures are examples.
What did it return, and what does next year cost.
Return is answered per unit of work where the workflow supports it: cost per resolved ticket, per reviewed document, per enriched record, against whatever that unit is worth.
More detail
Not every workflow has a clean denominator, and where one does not it is better to say so than to construct one.
The forecast goes in with a range and with its assumptions visible, plus scenarios for the decisions actually in front of the board, such as taking a feature to general availability.
Illustrative product view. Figures are examples.
How much of this could cost less.
Most companies find they can get the same AI output for roughly half the cost. A board that hears this without a plan attached will reach for a cap.
More detail
A board that hears it alongside an itemized ledger, an owner per finding and a realized figure already banked will treat it as a program.
Identified, in progress and realized are reported separately, because the difference between them is what tells a board whether the program is real.
| Optimization | Workflow | Status | Savings |
|---|---|---|---|
| Open-weight model covers the classification step | Support triage | Done | $15,600 |
| Cache the system prompt, 78% of input tokens | Document workflow | In progress | $8,700 |
| Put a ceiling on the agent retry loop | Internal agents | In progress | $7,400 |
| Move batch-eligible work off synchronous | Data enrichment | Planned | $5,100 |
| Committed-use tier now earned | All providers | Planned | $14,200 |
| Right-size a GPU endpoint at 11% utilization | Self-hosted models | Under review | $6,900 |
| Reclaim unopened seats on two AI tools | Engineering | Under review | $3,900 |
| Total | $61,800 a month |
Illustrative product view. Figures are examples.
Generate the AI section.
Bring last quarter's board pack and we will show you the generated version alongside it.
What a good AI section looks like
Four slides, and the discipline to stop there.
The number and the trend
Total AI cost, twelve month trend, split by product and department, with the two largest movements called out and explained in a sentence each.
The margin picture
AI as a share of revenue and as a component of cost to serve, by segment, with the direction of travel. This is the slide that gets questions.
The forecast
Next four quarters with a range, assumptions stated, and prior forecast accuracy shown so the range can be weighted. Scenarios for the decisions actually pending.
The optimization program
Identified, in progress and realized savings, with the largest findings named. This is the slide that converts a cost conversation into an investment one.
Keep reading
For CFOs and finance leaders
Owning the AI line, attributing it and defending it in a budget review.
Read moreFor executive teams
The evidence to expand AI investment rather than cap it.
Read moreReporting and exports
Close packs, chargeback journals and board summaries, generated on your calendar.
Read moreQuestions people ask
What if we cannot measure return on a workflow?
Say so. Some workflows have a clean denominator such as tickets resolved or documents processed, and some do not. A board is considerably more receptive to a stated limitation than to a constructed metric that does not survive a follow-up question. Report cost per unit where the unit is real, and report cost with a qualitative note where it is not.
How far ahead should we forecast AI for a board?
Four quarters with a range, and be explicit that the range widens considerably beyond two. Model pricing changes several times a year in both directions, which is a genuine source of uncertainty rather than an excuse. Showing prior forecast accuracy alongside the projection is what makes the range credible.
Should we present the savings figure to the board?
Yes, but only with the program attached. A savings figure presented alone invites a cap. The same figure presented with an itemized ledger, owners and an already-realized amount is evidence that the cost is being managed, which is the outcome you want.
Can the pack be generated automatically?
The quarterly summary is generated as a PDF or slides with the trend, drivers, unit economics, forecast and savings program. Most teams then edit the commentary. The intention is to remove the assembly work, not the judgment.