In my last article, I argued that public sector institutions face a uniquely difficult set of conditions for getting AI right: flat budgets, constrained workforces, constant scrutiny, and a mission they can’t walk away from. I offered three principles for working inside those constraints. One was to invest creatively—easy to say, but hard to do when dedicated AI funding remains uneven across the public sector.
It’s worth taking a second to describe the backdrop. The AI conversation has shifted from what these systems can do to what they cost. Will an AI-driven workflow pay for itself once its full building and running costs are counted? In the public sector, where there’s no slack to absorb underperformance, the answer is particularly important. It also makes the funding case and the value case the same case, which is why every move below starts from an outcome and its economics, not the technology.
Let me start with the most useful thing I’ve learned watching public sector organizations try to fund AI: the ones that succeed treat AI as a capital-allocation question owned at the top and funded like an investment with a return, rather than as a technology line item. That framing is harder in government, where appropriations rules often tie money to specific purposes and budget office scoring rules often prevent agencies from getting “credit” for projected returns. But the constraint is usually conceptual. The same money that wouldn’t be spent solely on “AI” would likely fund tangible outcomes, such as a 20 percent reduction in benefits processing time. The money for AI is, in some cases, already on the table—just allocated to the way the institution works today. When it isn’t, agencies can pursue outside funding.
Here are the moves I see working.
Find it in what you already spend
Two plays can surface AI funding from within existing budgets. In each case, the money follows an outcome the institution already cares about—AI is just how you get there faster.
Play 1: Ride along with work that’s already funded
Fold AI into work that is already underway, like a legacy-system replacement, a cloud migration, or a benefits modernization.
Canada’s Benefits Delivery Modernization program is a useful example. Employment and Social Development Canada launched it in 2017 to replace the aging technology behind Old Age Security, the Canada Pension Plan, and Employment Insurance. The effort moves those programs onto a modern common platform, and the department says AI and automation will be integrated over time to speed decisions and manage workloads. It’s not a clean success story. Costs have ballooned and timelines have stretched. The point, however, is that embedding AI in an already-funded mandate gave it a path that a standalone “AI project” almost never gets.
Play 2: Recover the slack already in the workforce budget
Do so not by cutting the budget, but by freeing capacity trapped in work people shouldn’t do manually.
Most public sector institutions carry hidden labor costs: backlogs, overtime, surge staffing, manual document review, and reliance on contractors. Quantify them. If a benefits office pays overtime to clear claims, fund an effort against the work creating the pressure. The promise should be specific: this investment will cut X hours of manual review, Y dollars of overtime, or Z weeks of backlog within a defined period.
For example, one government agency faced a widening gap between its workload and its workforce. Rather than ask for headcount it wasn’t going to get, the agency went after a specific bottleneck that was tied to an outcome: drafting requests for proposals faster. An agentic tool now drafts roughly 200-page solicitations in under ten minutes instead of two weeks, with early results showing RFP development time down 15 to 30 percent. In steady state, the agency projects that almost 70,000 staff-hours will be recovered each year, redirecting capacity straight back to work that the team couldn’t otherwise cover.
Sometimes, freed-up capacity shows up as cashable savings. Captured, those savings can seed the next AI investment. In the US, federal law already allows for this: since a 2017 change in tech-funding rules, an agency can set up an IT “working capital fund”—an internal account that lets it hold its own savings and unspent year-end money across budget years, rather than surrender them at the close of the fiscal year. The Small Business Administration used it first, ring-fencing about $6 million for a handful of enterprise priorities. The Department of Labor used the same mechanism to shift more of its IT spending toward modernization—from roughly 10 percent in the early 2020s to about 25 percent by 2024. This approach may not be quick and easy, but, in some cases, it’s a feasible option to pursue.
Explore outside funding
When the money isn’t inside the institution, it may be outside it. It’s worth exploring the following two plays in times of need, though I recognize they can be challenging.
Play 1: Shift to a capex funding model
Where there’s a genuine capital case, finance it like one. For substantial modernization, structures that repay the investment from the value they create belong in the conversation. The clearest is outcomes-based, or “pay-for-success,” financing: an outside investor provides capital, the institution repays only if agreed results materialize, and the risk of underperformance rests with the investor, not the taxpayer. It’s well-established outside of AI—used for workforce, health, and justice programs for over a decade—and it fits AI-enabled modernization, where savings are real but upfront capital isn’t available.
Istanbul offers an early example in an adjacent domain. Its development agency funded ICT training through an impact bond in which private investors provided the capital, and the government repaid only when job-placement and retention targets were met.
While this approach is still nascent for AI, where the capital case is real, it’s worth considering.
Another option is debt financing. Governments have borrowed to fund capital assets—roads, buildings, systems—for as long as they’ve existed. A large AI-enabled modernization can qualify for the same treatment when it produces reliable, measurable savings. In fact, some public universities have taken this approach for significant IT modernization efforts. The University of Washington, for example, used a mix of debt and equity to fund a more than $300 million finance transformation. Done well, public institutions using this approach can potentially repay the debt out of the efficiencies the system generates.
Play 2: Find an external partner
Financing isn’t the only way to bring in outside capital. When internal dollars are tight, partners can get the first projects off the ground on gentler terms. One route is a foundation, where funders’ priorities often align with what public-sector institutions need to do. The move is to co-develop a scalable playbook with the funder and to bring in a partner who bridges the gap between the grant and the working system.
Such models exist. The $20 million Public Benefit Innovation Fund—backed by the Ballmer Group, the Gates Foundation, and others—pairs its money with hands-on technical assistance rather than just writing checks. In its December 2025 round, it funded New Jersey’s Office of Innovation to scale a shared GenAI platform across departments such as Human Services and Labor, simplifying how residents access programs such as Medicaid and unemployment insurance—piloting first, then scaling what works. Other funders are moving in the same direction.
Universities are a version of the same idea. A university partner can supply model-testing capacity, data-science talent, responsible-AI review, and grant-writing muscle that a short-staffed agency doesn’t have. The University at Albany’s Center for Technology in Government is one working example: its AI in Government Lab built a grant-funded system pairing sensors and AI to help the small city of Schenectady, New York, monitor its infrastructure in near real time, with students doing technical work the city couldn’t staff.
None of this is the end state. The truth is that AI is an evergreen operating cost that public-sector budgeting rules haven’t yet caught up to. They will, but slowly. Until they do, the organizations that make real progress will get creative within the rules they have, funding outcomes rather than technology.
This article was originally published on LinkedIn on August 6, 2026.