The seat that asks the first real question: who funds this, and what do we get back? Funding models, a business case the board will accept, the pilot-funding cliff, AI FinOps, vendor financial terms, and turning risk into dollars.
A frame for the budget conversation, not financial advice. Models and ranges here structure the decision; your own budget office sets the numbers.
The fight every budget cycle isn't whether to fund AI. It's who pays. Four funding models, when each fits, and the catch.
Best whenEarly days, and for shared infrastructure (a gateway, enterprise licenses) that no single unit should own.
One front door; prevents duplicate buys; signals institutional commitment.
Becomes a free buffet, demand outruns the fund unless paired with prioritization.
Best whenMature programs where colleges own their own AI initiatives and budgets.
Spend sits with the unit that gets the value and feels the cost.
Fragments into duplicate tools and uneven access; the have-nots fall behind.
Best whenConsumption-priced AI at scale, where usage varies widely by unit.
Aligns cost to use; showback alone often curbs waste without billing anyone.
Metering overhead, and chargeback can suppress the very experimentation you want early.
Best whenResearch-driven or innovation pilots with a defined external source.
Extends reach without central dollars.
The classic cliff, the grant ends and there's no operating line to catch it.
The Compass scores impact; finance needs it in dollars. A one-page value case any sponsor can bring and any CFO can defend.
Net the two, state the payback period, and attach the mission value the numbers don't capture. Tie the score back to the Strategic Compass so finance and strategy tell the same story.
Promising tools die because the pilot got one-time money and the program needed recurring. Plan the landing before the jump.
Approve one-time pilot money and a provisional recurring line in the same breath, so a success doesn't die for lack of an owner's budget.
Define, before the pilot, what evidence justifies recurring funding, and what result means you stop.
Decide up front whether a scaled tool moves to central, unit, or chargeback funding. 'We'll figure it out later' is how the cliff forms.
A pilot with no off-ramp becomes permanent by inertia. Decide what retiring it looks like before you start.
Consumption pricing means cost scales with success. Four controls that keep that a feature, not a surprise.
Know your cost per user, per seat, or per query, consumption pricing means spend scales with success, for better and worse.
Hard caps and spend alerts at the gateway turn a runaway loop or leaked key from a five-figure surprise into a notification.
Inventory who's paying for what. Overlapping AI features across the LMS, suite, and point tools are quiet duplicate spend.
Showing a college its own consumption changes behavior more reliably than a policy, often before any chargeback is needed.
Procurement's legal teeth get the data right; these get the money right. The terms that decide whether a cheap pilot becomes an expensive dependency.
Per-seat rewards broad light use; consumption rewards narrow heavy use. The wrong model can double your cost at the same volume.
Cap annual increases. AI pricing is volatile, and renewal is where a 'cheap' pilot becomes an expensive dependency.
Know what happens when you exceed the tier, uncapped overage is how budgets blow up mid-year.
Multi-year deals buy a discount and a risk. Weigh the savings against how fast this market changes.
Switching costs and egress fees are real. Price the exit before you're locked in, not after.
Your board speaks in dollars; the risk register speaks in qualities. Translate, so governance spend defends itself.
When the cost of one incident dwarfs a year of governance, the business case for governance makes itself. Attach a dollar range to high-severity risk-register entries so the trade-off is explicit.