An AI Capital Allocation Framework for Executives

Most AI budgets are a list of projects with a total at the bottom. That structure cannot answer the questions a CFO or board will ask: what is infrastructure, what is a bet, what is a commitment, and what is the cost of keeping it all running. This guide gives executives a capital allocation framework with four categories, the return logic and evidence test for each, and a rebalancing rule.

What are the four categories?

Category	What it funds	Return logic	Evidence test	Typical error
Platform	Gateway, agent identity, connectors, evaluation harness, observability, inventory	Appreciates: every agent reuses it; time to next agent falls	Adoption; time to first agent for a new team	Treated as overhead and starved
Committed outcomes	Production agents with baselines, owners, and dates	Returns: measured cost, cycle time, and quality deltas	Baseline, target, pass rate, production metrics at each gate	Funded all at once with no gates
Options	Small, time-boxed experiments on uncertain use cases	Informs: converts uncertainty into a decision	A hypothesis, a decision date, and a decision made	Never reach a decision; become permanent pilots
Operations	Monitoring, scheduled evaluation, governance, reskilling, residual review	Protects: prevents drift, incidents, and credibility loss	Detection time, drift caught, quality trend	Cut after launch because the build is "done"

FISTA's AI funding models for executives piece covers where the money comes from; this piece covers how it is divided.

Why does platform appreciate?

Because each agent that uses it makes the next one cheaper and safer. The first agent pays for the gateway, identity, and evaluation harness; the tenth inherits them. Platform is the only AI investment whose return rises with use, which is why treating it as overhead to minimize is the most expensive allocation error. The CIO's guide to AI and agentic AI describes what the platform contains; the agentic AI value chain piece explains which platform layers to build versus adopt.

How should committed outcomes be funded?

In tranches at evidence gates: discovery (baseline and specification), build (evaluation set exists), deployment (pass rate meets threshold), run (production metrics against baseline). Each gate is a decision, and outcomes that miss a gate stop. Funding comes from the business unit that owns the outcome, so ownership follows money. The how to hold teams accountable for AI outcomes guide describes the review that administers the gates.

What makes an option an option rather than a pilot?

A hypothesis, a decision date, and a small budget. "Can an agent handle first-pass contract review at our quality bar? Decide by end of quarter on an evaluation of fifty real contracts." At the date, the option converts to a committed outcome, is closed, or is extended once with a stated reason. Options that never reach a decision are pilots, and a portfolio of them is the definition of AI theater.

Why is operations a category, not a line item?

Because it is where AI programs lose credibility. An agent launched and then unmonitored drifts, fails, and is blamed. Operations funding covers scheduled evaluation, monitoring, drift response, governance reviews, reskilling, and the residual human review of consequential actions. It scales with agents in production and should be protected from the instinct to cut it once builds are finished. The AI observability explained for executives piece explains what the money buys.

How should the mix change over time?

Year one is platform-heavy, with two or three committed outcomes and a handful of options. Year two shifts toward committed outcomes and operations as agents reach production and the platform stabilizes. By year three, operations is a substantial, stable line, committed outcomes dominate, options are a disciplined small share, and platform is a modest maintenance and roadmap cost. Programs whose mix never shifts are stuck in year one.

How is the portfolio rebalanced?

Quarterly, in the same review that decides autonomy and tranches. Outcomes that pass gates draw further funding. Options at their decision date convert or close. Operations funding is adjusted to the number and tier of agents in production. Platform funding follows the roadmap implied by committed outcomes. Money moves on evidence; nothing is protected by sunk cost. The AI operating rhythm for leadership teams guide places this in the cadence.

What should executives ask?

• What share of our AI spend is platform, outcomes, options, and operations?
• Which options have passed their decision date without a decision?
• Is operations funding scaling with agents in production, or was it cut after launch?
• What did the last quarterly rebalance move, and on what evidence?
• Was this year's budget built from committed outcomes or from a percentage?

How can FISTA Solutions help?

FISTA Solutions structures engagements to fit the framework: platform work that clients keep, committed outcomes delivered in evidence-gated tranches, time-boxed options with real evaluations, and operations handoff with monitoring in place, through its AI enablement and AI agents practices. Since 2017, FISTA has delivered 150+ projects for 50+ companies across 12+ countries.

To classify your current AI spend into the four categories and see what the mix reveals, talk to FISTA on WhatsApp, or read the AI business case template for the per-outcome artifact.