Leadership ¡ 4 minute read
AI Funding Models for Executives
Companies fund AI through four models: a central fund, business-unit budgets, a platform-plus-product split, and stage-gated tranches. Each produces different behavior. The mix that works funds the shared platform centrally, funds agents from the business units that own the outcomes, and releases money in tranches tied to evidence.
Strategy documents describe what a company intends to do with AI; the funding model determines what it actually does. Money shapes ownership, sequencing, and what counts as done. This guide compares the four funding models executives use, the behaviors each produces, and the mix that funds platforms centrally while tying agent spend to evidence.
Why does the funding model matter so much?
Because ownership follows money, and outcomes follow ownership. A central innovation fund produces projects that business units did not ask for and will not maintain. Business-unit budgets produce ownership but also five incompatible platforms. All-at-once funding produces pilots that drift because nothing forces a decision. The funding model is the operating model expressed in dollars, and it is one of the few levers executives control directly. FISTA's CFO's guide to AI and agentic AI covers the economics; this piece covers the structure.
What are the four models?
| Model | How it works | Behavior it produces | Best for |
|---|---|---|---|
| Central fund | One budget, allocated by a central group | Platform gets built; projects lack business owners; adoption is pushed | Early platform construction |
| Business-unit budgets | Each unit funds its own AI | Strong ownership; fragmented platforms; duplicated effort | Mature platform, many owners |
| Platform plus product | Central funds shared layers; units fund agents | Shared infrastructure with owned outcomes | Most companies past the first agent |
| Stage-gated tranches | Money released at evidence gates regardless of source | Decisions forced; drift prevented; theater starved | All agent funding |
The models combine. The recommended mix is platform-plus-product for the source of funds, with stage-gated tranches for the release of agent funds.
How does platform-plus-product work?
The platform (model gateway, agent identity, governed connectors, evaluation harness, observability, inventory) is funded centrally, because every agent uses it and no unit would build it alone. It is run as a product with a roadmap and adoption metrics. Agents are funded by the business units that own the outcomes, from their own budgets, which makes the business owner real. The platform team charges nothing or a simple allocation for usage, so that adoption is not taxed. The CIO's guide to AI and agentic AI describes the platform this funds.
How do stage gates work?
Agent funding is released in tranches tied to evidence:
- Discovery tranche: small; produces a baseline, a specification, and the start of an evaluation set. Gate: baseline and spec approved, owners named.
- Build tranche: funds the build on the platform. Gate: evaluation pass rate meets the threshold.
- Deployment tranche: funds supervised deployment and the human review it needs. Gate: production metrics against baseline; incident record.
- Run funding: an operating line, owned by the business unit, reviewed monthly on cost per task.
Projects that miss a gate stop, with the discovery output kept for later. The structure starves theater, because money cannot flow without evidence. The how to structure an AI pilot agreement guide applies the same logic to external partners.
How should run cost be treated?
As a separate operating line, owned by the business unit, tracked as cost per task against baseline. Run cost covers inference, monitoring, maintenance, scheduled evaluation, and the residual human review the agent still needs. It grows as agents take on volume, and within a year or two it is usually the larger line. The most common funding failure is cutting run funding for a working agent because the build is "finished"; the agent then drifts, degrades, and is blamed. The AI total cost of ownership guide lists the run-cost drivers.
How much should be budgeted?
Bottom-up from committed outcomes, not top-down from a percentage. For each committed outcome: discovery, build, deployment, and run at expected volume. Plus the platform. Plus operations, reskilling, and governance. Top-down budgets produce portfolios sized to the money rather than to the outcomes, which is how a company ends up with thirty pilots and nothing in production. The AI business case template provides the per-outcome structure.
What should executives ask?
- Who funds the platform, and is it run as a product with adoption metrics?
- Which business unit funds each agent, and does that unit own the outcome?
- What evidence gates release agent funding, and has any project been stopped at one?
- Is run cost a separate line, owned by the business, tracked per task?
- Was this year's AI budget built from outcomes or from a percentage?
How can FISTA Solutions help?
FISTA Solutions structures its engagements to fit stage-gated funding: discovery that produces a baseline and specification, builds gated on evaluation, and supervised deployment with production metrics, through its AI enablement and AI agents practices. Since 2017, FISTA has delivered 150+ projects for 50+ companies across 12+ countries.
To design a funding model that produces production agents rather than pilots, talk to FISTA on WhatsApp, or read how to budget for digital FTEs.
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.
01Should AI be funded centrally or by business units?
Both, for different things. The shared platform (gateway, identity, connectors, evaluation, observability) is funded centrally because every agent uses it and no unit would build it alone. Agents are funded by the business units that own the outcomes, because ownership follows money. A purely central fund produces projects nobody owns.
02What is stage-gated AI funding?
Releasing money in tranches tied to evidence: a small tranche to establish a baseline and specification; a build tranche once the evaluation set exists; a deployment tranche on the evaluation pass rate; and run funding once production metrics show the outcome. Each gate is a decision, and projects that miss a gate stop rather than drift.
03How should AI run costs be budgeted?
As a separate operating line from build cost, owned by the business unit running the agent, tracked as cost per task against the baseline, and reviewed monthly. Run cost includes inference, monitoring, maintenance, evaluation, and residual human review. Expect it to grow as agents take on volume and to become the larger line.
04How much should a company budget for AI?
There is no useful percentage. Budget from the committed outcomes: the build cost to reach production for each, plus the platform, plus the run cost at expected volume, plus the operations and reskilling that follow. Companies that budget top-down produce portfolios of pilots sized to the budget rather than to outcomes.
05What funding mistakes do executives make with AI?
Funding breadth instead of depth; funding builds without run cost estimates; central innovation budgets with no business owners; releasing all the money at once with no evidence gates; treating AI as a one-time project rather than an operating capability; and cutting run funding for agents that are working because the build was "finished."
Continue exploring
Related capabilities
Start with the hard problem
Need the outcome owned, not merely analyzed?
Tell us where delivery is constrained. Weâll map the fastest credible path from intent to verified production.