Board packs are read quickly, often the night before, by people who are not going to ask a clarifying question about EC2. An infrastructure slide has one job: answer the obvious questions before they are asked, and leave no ambiguity that invites a follow-up you cannot answer in the room.
The four numbers
Total against budget. The absolute figure and the variance. No commentary yet.
The ratio. Infrastructure cost as a proportion of revenue, or cost per customer. This is the number that determines how the rest of the page reads.
Top three drivers of change. In business language, with a figure each.
Forecast with a range. Next quarter, as a band rather than a point.
That is the whole report. Everything else is available on request.
The ratio is the argument
If you present only one thing, present cost as a ratio.
Infrastructure spend up 60% year on year reads as a problem. Infrastructure spend up 60% while revenue grew 90%, taking cost from 14% of revenue to 12%, reads as operating leverage — and it is the same fact.
The ratio also pre-empts the question behind the question. What a board is really asking is whether spend is growing because the business is growing or because something is out of control. The ratio answers that in one line, and it answers it in the unit they already use to think about every other cost in the business.
Pick the denominator your board already sees in other slides. Consistency matters more than which one you choose.
Name drivers as business activities
"Compute up $40k" is not a driver, it is a restatement. A driver is a decision or an event:
- "Three enterprise customers onboarded; their data volume added $40k a month in processing and storage."
- "Moved inference to a larger model in March to reduce error rate; $18k a month, permanent."
- "A batch job ran unchecked for six days after a failed migration; $9k, one-off, detection now in place."
Each names what happened, what it cost, and whether it recurs. Three of those and the page is complete.
The forecast should be a range
A single number will be wrong, and being visibly wrong twice costs you the credibility of every future forecast.
Present a band with the assumption stated: "£420k–£480k next quarter, assuming customer growth continues at the current rate and no new model migrations." That is defensible when it lands anywhere inside the range, and when it does not, the stated assumption tells you which input broke.
A range also signals that you understand the uncertainty, which reads as competence rather than hedging.
What to leave out
Provider names. Whether it is AWS or GCP is not a board-level fact.
Service-level breakdowns. A table of twenty services is the fastest way to generate questions that do not matter.
Anything needing translation. If a term requires you to explain it verbally, it does not belong on a page read without you.
Optimisation projects in progress. Report them when they have landed and produced a number. Reporting intentions invites a status question every quarter thereafter.
Have one layer of backup
You will be asked one hard question. It is usually a version of "can this be reduced" or "what happens if growth doubles".
Have two things ready but not on the page: the largest single line and what it buys, and a stress case at double the growth assumption. Producing either on request is impressive; having them on the slide is clutter.
Make it a by-product, not a project
The reason these reports are painful is that they are assembled from several consoles a few days before the meeting, with the causes reconstructed from memory. That is also why the forecast is weak — it is extrapolated from a total rather than built from unit economics.
If spend is consolidated with a categorisation that already maps to how the business thinks, and drivers are recorded when they happen rather than remembered afterwards, the quarterly report becomes a summary of things already known.
StackSpend consolidates every provider into one total with P&L-friendly categories, forecasts month-end with a confidence range from a time-series model, and records anomalies with dates as they occur — which is where the drivers section comes from.
FAQ
What should a board infrastructure cost report include?
Total against budget, cost as a ratio of revenue or customers, the top three drivers of change in business terms, and a forecast presented as a range. One page is enough.
What ratio should I use for infrastructure spend?
Whichever denominator your board already uses elsewhere — revenue, customers or transactions. Consistency with the rest of the pack matters more than the specific choice.
Should I present a single forecast number or a range?
A range, with the assumption stated. A point estimate will be wrong, and being visibly wrong twice undermines every forecast you present afterwards.
How much detail should go in a board cost report?
Less than feels comfortable. Service-level tables and provider names generate questions that do not change any decision. Keep depth in reserve for the one hard question you will be asked.
