Table of Contents
- Why Boards Demand a Structured AI Board Pack in 2026
- Slide 1: AI Spend & Unit Economics
- Slide 2: Deployment vs Scale Status
- Slide 3: Model & Vendor Concentration
- Slide 4: Agent Incident Log
- Slide 5: Governance Posture
- Slide 6: Talent & Capacity
- Slide 7: The Two Decisions
- Reporting Cadence & What to Do When the Numbers Are Bad
- Summary & Next Steps
Why Boards Demand a Structured AI Board Pack in 2026
By 2026, the AI conversation in the boardroom has moved from curiosity to capital allocation. Directors no longer ask whether the company is using AI—they want to know exactly how much is being spent, what it’s returning, and where the next material risk is sitting. For a CTO, the quarterly update can no longer be a few slides on “AI initiatives” buried inside the engineering report. It needs to be a standalone, CFO-grade board pack that speaks the language of unit economics, concentration risk, and governance maturity.
This shift is especially acute for mid-market brands, private-equity portfolios, and scale-ups that are moving fast with agentic AI, public-cloud re-platforming, and multi-model architectures. Their boards—often composed of operators who have seen technology cycles come and go—expect the same rigor they apply to financial reporting. That’s why fractional CTO and AI program leadership engagements now routinely start with a board-readiness exercise: the pack a CTO owes when the board asks, “Show us the AI numbers.”
PADISO founder Kevin Kasaei has built the firm’s CTO as a Service practice around this exact moment. Whether working with a US-based private equity firm consolidating portfolio companies or an Australian scale-up preparing for a Series B, the deliverable is always the same: a board pack that connects AI investment to EBITDA, risk, and strategic decisions. The eight slides that follow form the backbone of that pack—and they’re what every CTO should have ready before the next board meeting.
flowchart TD
A[AI Board Pack] --> B[Slide 1: Spend & Unit Economics]
A --> C[Slide 2: Deployment vs Scale]
A --> D[Slide 3: Model & Vendor Concentration]
A --> E[Slide 4: Agent Incident Log]
A --> F[Slide 5: Governance Posture]
A --> G[Slide 6: Talent & Capacity]
A --> H[Slide 7: The Two Decisions]
B --> I[Reporting Cadence]
C --> I
D --> I
E --> I
F --> I
G --> I
H --> IThese eight slides don’t just inform—they drive decisions. They give a board the clarity to approve a new model migration, to double down on a high-performing agentic workflow, or to pause a deployment that’s burning cash without scaling. And when the numbers are bad, they provide the structure to have an honest conversation without panic.
Slide 1: AI Spend & Unit Economics
The first slide sets the financial baseline. It answers a simple question: “What are we spending on AI, and what are we getting back?” For mid-market boards and private equity operating partners, this slide is often the most scrutinized—because AI spend can balloon quietly inside cloud bills, API invoices, and headcount allocations.
Breaking Down AI Spend
Spend should be categorized into no more than five line items: foundation model API costs (across providers like Claude Opus 5, Sonnet 5, GPT-5.6 Sol, and Gemini 3), fine-tuning and training infrastructure (often on AWS, Azure, or Google Cloud), inference hosting for open-weight models, internal tooling and observability, and people—engineers, prompt architects, and governance leads. Avoid lumping everything under “cloud.” That’s how AI costs become invisible.
Platform engineering in San Francisco teams regularly see inference costs overtake training costs once agentic workloads hit production. That’s a leading indicator the board needs to see, because it signals a shift from experimentation to operational dependency.
Unit Economics That Matter
For each AI-powered product, workflow, or internal tool, the pack must show a simple unit-economic view: cost per inference, cost per transaction, cost per agent task completion, and the revenue or efficiency value attached to that unit. Where AI is replacing a manual process, show the fully loaded cost of the manual alternative. Directors understand comparisons, not abstractions.
A board that sees a $0.03 inference cost delivering a $4.70 gross margin uplift per transaction will fund more of it. One that sees a $1.20 agent task cost with no measurable outcome will demand a rethink. That’s the conversation you want.
When the Numbers Are Bad
If unit economics are underwater—or worse, unknown—don’t hide it. Present the gap, the diagnostic timeline, and the decision you’re asking for: a 60-day optimization sprint, a model switch to a lower-cost tier like Haiku 4.5 for non-critical tasks, or a pause on a specific agent deployment. Boards respect a CTO who brings a problem with a plan. AI strategy and readiness engagements at PADISO often begin with exactly this kind of unit-economic triage, turning an opaque cloud bill into a board-ready P&L.
Slide 2: Deployment vs Scale Status
Boards confuse “deployed” with “scaled.” Slide 2 eliminates that confusion by showing a clear, color-coded matrix of every AI capability in the organization: pilot, limited production, scaled, and retired. This slide is where the CTO proves they’re running a portfolio, not a science fair.
The Scale Funnel
Plot each initiative against two axes: user or transaction volume, and business impact. A conversational AI agent handling 10,000 customer queries a day with a 92% deflection rate belongs in the top-right quadrant. An internal document summarizer used by three people in legal belongs in the bottom-left. The board needs to see the concentration of value.
For PE-backed companies executing roll-ups, this slide is doubly important. A private equity roll-up strategy often involves deploying the same AI capability across multiple portfolio companies. Slide 2 shows whether the playbook is actually working—or whether each acquisition is reinventing the wheel. PADISO’s Venture Architecture & Transformation practice regularly builds these deployment dashboards so that operating partners can see, at a glance, where AI is creating portfolio-level value versus where it’s stuck in pilot purgatory.
What to Do When Scale Stalls
If a capability has been “in pilot” for two quarters, flag it. The board should ask: is it a data problem, a change-management problem, or a model-performance problem? Be specific. In many cases, the bottleneck isn’t the model—it’s the integration layer or the absence of a platform design and engineering foundation that can serve inference reliably at scale. Name the bottleneck and the investment needed to clear it.
Slide 3: Model & Vendor Concentration
In 2026, the model landscape is a multi-polar field. The Claude 5 family—Opus 5 and Sonnet 5 with 1M-token context windows, and Fable 5 as the most capable widely released model—competes with GPT-5.6 Sol and Terra, Gemini 3, Kimi K3, and a growing ecosystem of open-weight alternatives. A board that doesn’t understand its model concentration is flying blind on cost, performance, and supply-chain risk.
Mapping the Model Portfolio
Slide 3 lists every model in production, its provider, its use case, and its cost tier. It also shows the percentage of total inference volume each model represents. A healthy portfolio in 2026 rarely relies on a single model family. Instead, it tiers workloads: Opus 5 for high-stakes reasoning, Sonnet 5 for broad agentic tasks, Fable 5 for customer-facing experiences where capability and safety are paramount, and Haiku 4.5 for high-volume, low-latency classification and routing. Open-weight models may handle sensitive on-premise workloads.
For CTOs overseeing AI and agents automation, vendor concentration is a board-level risk. A sudden price increase from a hyperscaler or an API deprecation can destabilize unit economics overnight. The board pack should quantify that exposure: “42% of our inference spend goes to a single provider” is a line that will get attention—and it should.
The Migration Readiness Question
Alongside the concentration chart, include a one-sentence migration readiness statement. If you had to move 30% of inference volume to a different provider within 30 days, could you? Many teams discover they can’t, because prompt engineering, guardrails, and evaluation suites are tightly coupled to a specific model’s behavior. That’s a technical debt item the board needs to fund remediation for. Fractional CTO advisory in New York often surfaces this gap during diligence for fintech and media companies that have built on a single model stack without an abstraction layer.
Slide 4: Agent Incident Log
Agentic AI means software that acts—not just generates text. When an agent can send an email, update a CRM, or trigger a financial transaction, incidents are inevitable. Slide 4 is the agent incident log, and it’s the slide that separates serious AI operators from everyone else.
What Counts as an Incident
Define incidents clearly: an agent that hallucinated a customer commitment, a workflow that executed an unintended action, a guardrail bypass, a data leakage event, or a model output that required human intervention to prevent harm. Each incident gets a row: date, agent ID, severity, business impact, root cause, and remediation status. The board doesn’t need a novel—it needs a table it can scan in 90 seconds.
Trend Lines, Not Just Events
One incident is a bug. Three of the same type is a pattern. Slide 4 should show a trailing-six-month trend line of incidents by severity and category. A rising trend in “guardrail bypass” incidents demands a governance response. A flat or declining trend in a growing deployment base signals maturity. This is the kind of operational rigor that AI strategy and readiness engagements bake into the operating model from day one.
When the Log Is Ugly
If the incident log is growing faster than deployment volume, say so. Then present the decision: invest in a dedicated AI observability layer, tighten agent permissions, or slow the rollout of a particular agent class until the safety envelope is proven. Boards will fund safety—if they understand the cost of not funding it. Security audit readiness via Vanta often surfaces agent-permission gaps that would otherwise become incidents, which is why PADISO integrates SOC 2 and ISO 27001 controls directly into the agent lifecycle.
Slide 5: Governance Posture
Governance is no longer a compliance checkbox—it’s a board’s primary line of defense against AI risk. Slide 5 shows where the organization stands against a recognized framework, and it’s the slide that general counsel and audit committee chairs will lean into hardest.
Aligning to the NIST AI RMF
The NIST AI Risk Management Framework provides the most widely adopted structure for AI governance in the US and, increasingly, in boardrooms across Canada and Australia. The full AI RMF 1.0 document defines four core functions—Map, Measure, Manage, and Govern—that map directly to board-level oversight. Slide 5 should show a maturity score for each function, with evidence: documented risk assessments, model cards, bias testing results, and human-in-the-loop protocols.
The NIST AI Resource Center offers implementation support materials that CTOs can use to accelerate this posture. For mid-market companies that don’t have a dedicated AI governance team, a fractional CTO in Boston can often stand up the core artifacts in a matter of weeks, using the NIST AI RMF Playbook as a practical guide.
The Governance Gap That Worries Boards
Most boards care less about framework alignment in the abstract and more about one question: “Could an AI decision trigger a regulatory action, a lawsuit, or a reputational event?” Slide 5 must answer that directly. If the organization is using AI in lending, hiring, or healthcare, the governance posture needs to reflect domain-specific controls. The Brookings analysis of the NIST AI RMF underscores that voluntary frameworks are becoming de facto standards, and boards that ignore them are taking on unnecessary liability.
From Framework to Audit-Readiness
For companies pursuing enterprise deals, governance posture must translate into audit-readiness. PADISO’s security audit service, built on Vanta, maps AI governance controls directly to SOC 2 and ISO 27001 criteria. That means a board can see, on a single slide, not only the AI RMF maturity level but also whether the organization is on track to pass its next audit. For private equity firms driving portfolio value creation, that dual visibility—AI governance plus compliance posture—is often the difference between a company that sells at a premium and one that doesn’t.
Slide 6: Talent & Capacity
AI doesn’t run itself. Slide 6 addresses the board’s growing concern: “Do we have the people to execute this plan?” It covers headcount, skills gaps, and the build-vs-buy decisions that determine whether the AI strategy is resourced or aspirational.
The Talent Stack in 2026
A modern AI team isn’t just machine learning engineers. It includes prompt architects, agent-ops engineers, AI product managers, and governance specialists. Slide 6 should show current headcount against target, with a clear hiring or contracting plan for the gaps. For mid-market companies that can’t attract or afford a full-time CTO with AI depth, CTO as a Service provides a fractional leader who can build and mentor that team while keeping the board pack current.
Build, Borrow, or Buy
Not every capability needs to be in-house. Slide 6 should categorize each critical AI function as build (internal hire), borrow (fractional or contract), or buy (platform or managed service). For example, a company might build its prompt engineering muscle internally, borrow a fractional CTO in San Francisco for architecture oversight, and buy an observability platform for agent monitoring. This framework prevents the board from approving a headcount budget that’s misaligned with the actual operating model.
When Talent Is the Bottleneck
If AI initiatives are stalling and the root cause is talent, the board pack should say so explicitly—with a specific ask. That might be approval to engage a venture studio and co-build partner to accelerate a critical agentic workflow, or to invest in an internal upskilling program. The AI Maturity Guide 2026 highlights that talent readiness is one of the strongest predictors of AI ROI, and boards that treat it as a line item rather than a strategic investment consistently underperform.
Slide 7: The Two Decisions
Every board pack must end with a decision, not just information. Slide 7 presents exactly two decisions the board is being asked to make. Two is the right number: one is too narrow, three is too many. Two forces prioritization.
Decision One: The Investment Ask
This is typically a resource decision: approve a budget for a new model migration, fund a dedicated AI governance hire, or greenlight the expansion of an agentic workflow from pilot to scale. The ask must be concrete—a dollar amount, a timeline, and a clear link to the unit economics on Slide 1. If the board approved the last ask, show the outcome. Directors remember what they funded, and they want to know it worked.
Decision Two: The Strategic Choice
This is a direction-setting decision: commit to a multi-model architecture, adopt an open-weight-first policy for sensitive workloads, or authorize a third-party AI audit ahead of a fundraise or exit. Strategic choices often don’t have immediate P&L impact, but they shape the company’s risk profile and competitive position for years. CTO advisory in Melbourne regularly frames these choices for insurance and health scale-ups navigating APRA and LIF compliance, where the strategic decision on model hosting can determine whether a product ever reaches market.
Structuring the Ask
Each decision slide should contain: the recommendation, the alternatives considered, the expected impact, the cost of inaction, and the specific vote being requested. Boards operate on clarity. A well-structured decision slide turns a quarterly update into a governance event, which is exactly what PE portfolio value creation demands when multiple companies are moving at different speeds.
Reporting Cadence & What to Do When the Numbers Are Bad
A board pack is only as good as the cadence that sustains it. For mid-market companies and PE-backed businesses, the right rhythm is usually quarterly, with a monthly one-page dashboard for the CEO and the operating partner. The full eight-slide pack goes to the board, but the monthly snapshot keeps leadership aligned between meetings.
The Quarterly Deep-Dive
Each quarter, the CTO walks the board through all eight slides, with particular focus on the slides that have changed materially. If unit economics have shifted due to a model-price change from a hyperscaler, Slide 1 gets the time. If an agent incident triggered a customer escalation, Slide 4 leads. The board pack is modular by design—it flexes to the quarter’s reality.
The Monthly One-Pager
Between board meetings, a single page with four metrics keeps the conversation grounded: total AI spend vs. budget, top-line unit economics for the highest-volume workload, incident count and severity, and a red/amber/green status on the two decisions from the last board meeting. This prevents surprises and gives the CTO a regular forum to flag emerging issues.
When the Numbers Are Bad
Every CTO will eventually present a board pack with red ink: unit economics that have inverted, an incident that made it to a customer, a model concentration that became a single-point-of-failure. The worst thing to do is bury it. The board pack should surface the bad news with the same structure as the good: what happened, why it happened, what’s being done, and what’s being asked of the board. In many cases, the board’s role is to provide air cover—to approve an unplanned investment, to absorb a short-term margin hit, or to communicate with external stakeholders. That only works if they’re told early and clearly.
Case studies from PADISO’s work with mid-market brands show that boards which receive honest, structured AI reporting are far more likely to increase AI investment over time. Trust is built in the quarters when the numbers are bad, not when they’re good.
Summary & Next Steps
The AI board pack a CTO owes in 2026 is not a status report—it’s a decision-making instrument. It translates technical reality into the language of spend, risk, and strategic choice that boards are built to process. The eight slides—spend and unit economics, deployment vs scale, model and vendor concentration, agent incident log, governance posture, talent and capacity, and the two decisions—form a complete narrative that any director can follow, regardless of technical background.
For CEOs and boards of US and Canadian mid-market companies, for private equity firms running roll-ups, and for scale-up founders preparing for their next round, having this pack ready is a competitive advantage. It signals operational maturity, it accelerates board decisions, and it protects the organization from the AI risks that are now front-page material.
PADISO was built for this moment. From fractional CTO advisory in Sydney to AI strategy for financial services, the firm’s engagements produce exactly the kind of board-ready artifacts that turn AI from a boardroom anxiety into a boardroom asset. Explore our services or book a call to start building your AI board pack before the next quarterly meeting. The board is waiting—and now you know exactly what to show them.