Checklist · 4 minute read
AI Board Reporting Checklist: What to Put in the Pack
Board reporting on AI works when the same short set of numbers appears every quarter: what is deployed, what it costs, whether quality is holding, what went wrong, and what risks are open. Consistency is what makes a crisis conversation start from shared understanding rather than from scratch.
Board reporting on AI works when the same short set of numbers appears every quarter. This checklist covers what belongs in the pack, drawn from FISTA Solutions' AI enablement governance work.
What belongs in the report?
Six sections, each on one page or less.
| Section | What it answers |
|---|---|
| Inventory | What is running and what changed |
| Cost | Full operating cost and trend |
| Quality | Is performance holding? |
| Incidents | What went wrong and what changed |
| Risks | What is open and who owns it |
| Accountability | Who is answerable for each system |
Inventory
Establishes scope, which everything else depends on. See AI service catalog template.
- Total count of deployed AI systems
- Change since last report: added, retired, materially changed
- Split between internal and customer-facing
- AI features inside purchased software included
- Systems making automated decisions about individuals identified
- Systems with agent capability identified separately
- A note on how the inventory is maintained
Cost
Full operating cost, or the correction comes later. See the operating cost of intelligence.
- Model and infrastructure spend for the period
- Human review hours costed
- Evaluation and maintenance effort costed
- Trend against the previous three periods
- Cost per unit of work where a baseline exists
- Forecast for the next period with assumptions stated
- Variance against budget explained
Quality
A trend, with the methodology stated.
- Quality metric per significant system, with the measure defined
- Trend over at least three periods
- Material changes explained
- Proportion of output reviewed by a person
- Human correction or disagreement rate
- Any system where quality is not measured, named as such
- Methodology stated so the number is interpretable
Incidents
Reported plainly, which is what builds credibility.
- Incidents in the period with severity and duration
- Customer impact stated honestly
- Root cause in plain language
- What changed as a result
- Open actions from previous incidents
- Near misses worth noting
- Trend in incident count and detection time
Risks
Open items with owners and dates, not a generic register.
- Top risks listed with owner and mitigation
- Changes since last report
- Regulatory developments relevant to your sector
- Vendor concentration or continuity concerns
- Capability gaps that constrain the programme
- Risks accepted, with who accepted them
- Anything requiring a board decision, stated clearly
Accountability
A name per significant system.
- Named owner for each customer-facing system
- Confirmation that owners have authority to pause
- Governance forum and its cadence
- Policy changes made in the period
- Training or capability investment
- Audit or assurance activity completed
- Escalations that reached the board and their outcome
What are the most common failures?
A different deep dive each quarter. Cost that omits review hours. Quality as a snapshot. Incidents omitted until one is unavoidable. And no named owner, which is the question directors ask first.
Who should own this?
An executive owns the report — typically whoever is accountable for the AI programme. Preparation should be distributed, but a single person answers for the numbers in the room.
How often should it run?
Quarterly for most organisations, with an interim note if a material incident occurs. Annual reporting is too infrequent for an estate that changes this quickly.
What evidence should it produce?
The reports themselves, with consistent metrics across periods. A four-quarter series showing the same measures is itself evidence that the programme is managed.
What if the numbers are not available?
Report that plainly rather than estimating. A line saying quality is not measured for these three systems is uncomfortable and it is the correct disclosure.
It is also the fastest way to get the instrumentation funded, because a gap stated to a board tends to close. Estimated figures presented as measured create a much worse problem later. See what boards will ask about AI.
What should you do first?
Build the inventory and put a count in the next pack. It is the question directors ask first and the one most organisations cannot answer.
How FISTA Solutions helps
FISTA Solutions builds and operates production AI systems through AI agents, AI enablement, and forward deployed engineering: reporting built from generated figures rather than hand-assembled ones, with the same measures each period so trends rather than snapshots reach the board, decisions documented with their reasoning, and handover that leaves your team able to maintain what was delivered. The record is 150+ projects for 50+ companies across 12+ countries.
To adapt this checklist to your environment, message FISTA on WhatsApp, or read what boards will ask about AI.
Share-ready article cover
Download the generated social format.
Clear answers
Questions raised by this field note.
Straightforward guidance for evaluating scope, fit, and the next step.
01Why report the same numbers each time?
Because trends are the information. A one-off deep dive impresses and teaches nothing; the same five measures over four quarters show whether the estate is under control.
02What should lead the report?
The inventory: how many AI systems are deployed, what changed this quarter, and which are customer-facing. It establishes scope, which every other number depends on.
03How should cost be presented?
Full operating cost including review time and maintenance, not just the model bill. A cost figure that omits the largest line invites a difficult correction later.
04Should incidents be reported?
Yes, plainly. Boards that hear about problems routinely trust the reporting; boards that only hear good news until a crisis do not, and that trust matters exactly when it is tested.
05What makes a report credible?
Numbers generated from systems rather than assembled by hand, a stated methodology, and consistency with what was reported last time. Hand-assembled figures drift and get caught.
Continue exploring
Related capabilities
Start with the hard problem
Need the outcome owned, not merely analyzed?
Tell us where delivery is constrained. We’ll map the fastest credible path from intent to verified production.