Year-End Funds, Fast: Obligating AI Dollars via SDVOSB Sole-Source
If you have unobligated FY2026 dollars and an AI modernization need, the fastest defensible way to put that money on contract before September 30 is a single-source award to a certified SDVOSB. It clears in weeks, not quarters, because it skips full-and-open competition. It rides a statutory exception, not an urgency justification, so its protest profile is clean. And it counts toward the 5% governmentwide SDVOSB goal that most agencies are still chasing. One award, every box checked.
Two pressures collide on the fourth quarter of every fiscal year, and they collide harder in FY2026. One is the calendar: one-year money expires at midnight on September 30, and it doesn’t come back. The other is the scorecard: the governmentwide SDVOSB goal jumped from 3% to 5%, and the early numbers say FY2025 may have missed it. The SDVOSB sole-source award is the rare instrument that relieves both at once. This piece is about the timing and the tactics — when to start, how late is too late, and how to spend fast without buying shelfware. (The rewrite that reshaped this authority is in how the FAR Part 19 overhaul lets you direct-award AI; the full sole-source workflow is in the $5M sole-source how-to; the broad picture is in the pillar.)
Why is September the deadline — and why does that matter for AI?
The deadline is a statute, not a habit. Under the bona fide needs rule, annual appropriations are available only for the needs of their fiscal year of availability. For one-year money — most operations-and-maintenance and salaries-and-expenses accounts — that period ends September 30. Unobligated balances then sit in an expired account for five years, available only for adjustments to obligations already made, before they cancel and revert to Treasury. They cannot start new work. That is the entire mechanism behind “use-it-or-lose-it.”
And agencies wait. The canonical study (Liebman and Mahoney, American Economic Review, 2017) found procurement spending in the final week of the fiscal year runs 4.9 times the rest-of-year weekly average — and, crucially, that IT projects obligated in that last week are two to six times more likely to score below the median on quality. Mercatus analysis puts 16.3% of executive-branch obligations in September alone, roughly double a uniform distribution.
That quality finding is not a footnote. It is the risk. AI obligations are IT obligations, and IT is exactly where rushed year-end money goes bad. So the operative question for a CO or PM in May is not “can I obligate this” — it’s “can I obligate this without buying shelfware.” That is the production-path diagnostic: if the honest answer to “who runs this after delivery” is “we’ll figure it out next year,” you’re funding a pilot that will stall.
What kind of AI work can year-end money actually buy?
This is where the bona fide needs rule shapes scope. Services are generally a need of the year they’re performed, so two structures are clean and one is a red flag:
Non-severable deliverable. A single defined output — a trained and documented model, a delivered evaluation pipeline, a pilot-to-production package, a RAG application — is treated as a bona fide need of the year of award. You may obligate the full amount with FY2026 funds even if performance runs into FY2027.
Twelve-month severable run. Recurring support that can be cut at any point — model monitoring, ongoing fine-tuning, MLOps run-rate — is severable. One-year FY2026 funds can cover up to 12 months of it crossing into FY2027 under the severable-services exception.
Open-ended “AI transformation services” with no defined deliverable, obligated at the last minute on one-year money, is the textbook GAO red flag. Don’t scope it that way. Pick the deliverable or cap the run.
Why is the SDVOSB sole-source path the right tool right now?
Because from May or June, most other paths can’t finish in time. GAO found DoD median procurement lead time around 32 days for simpler actions, but competitive higher-dollar awards routinely stretch many months — the Army’s own benchmark is roughly 210 days for sole-source versus 240 for competitive, and lead times for higher-value competitive contracts grew by about 70 days across FY2019–FY2022. A competitive AI procurement above the simplified acquisition threshold started in late spring is unlikely to award by September 30 unless it’s an order off an existing vehicle. Sole-source compresses that timeline to the days it takes to document market research, draft the justification, and issue the award.
And it’s the only mechanism that relieves both year-end pressures in a single action:
| Pressure | What an SDVOSB sole-source delivers |
|---|---|
| Time | No solicitation, no competitive evaluation, no Q&A cycle. PALT collapses to package-prep time. |
| Statutory cover | Authorized by 15 U.S.C. 657f under FAR 6.302-5(b)(6) — “authorized by statute,” not a 6.302-2 urgency J&A. |
| Goal credit | Counts toward the 5% governmentwide SDVOSB goal and your agency’s negotiated target. |
| Bona fide needs | Scope as a non-severable deliverable or a 12-month severable run — both clean. |
The four-condition test — no reasonable expectation of two or more SDVOSB offers, within the dollar ceiling, responsible contractor, fair and reasonable price — plus the market-research and price-reasonableness mechanics live in the $5M sole-source how-to. The one thing not to do under time pressure: shortcut the market research. That’s the protest vector. (For the credentialing argument — why SDVOSB is leverage and not just access — see SDVOSB is leverage. Use it.)
How much goal pressure is actually on the table?
More than the headline suggests. Section 863 of the FY2024 NDAA raised the governmentwide SDVOSB goal from 3% to 5% — a 67% increase, and statutory, not an OMB aspiration. FY2024, the first year at 5%, cleared it: 5.14% / $32.8 billion governmentwide on the SBA scorecard released July 2025.
But look underneath the number. VA, under its Veterans First program, put 23.63% of its prime dollars — $10.2B — to SDVOSBs in FY2024. DoD, the largest buyer in government, earned an “A” overall but landed SDVOSB attainment in the low-3% range, well short of 5%. The governmentwide figure only clears the bar because VA overshoots so far it carries DoD’s shortfall. Strip VA out and most of the government is under the goal.
Now the FY2025 signal. The official scorecard isn’t out yet — expect it around July 2026 — but preliminary FPDS analysis (Fed-Spend Research Team, February 11, 2026) puts FY2025 governmentwide SDVOSB prime awards at $28.6 billion across roughly 52,000 actions, a $4.2B drop from FY2024. If the official scorecard confirms it, FY2025 is the first miss of the new 5% goal. That is precisely the kind of public report-card embarrassment that has agency leadership instructing contracting officers to find more SDVOSB obligations before the fiscal year closes. The two pressures — expiring dollars and a missed goal — point at the same award.
One more tailwind on the supply side: certification is now mandatory (self-certification ended December 22, 2024), so awarding to a VetCert-certified SDVOSB is an audit-proof goal-credit transaction — and SBA cleared its certification backlog in November 2025, cutting processing to about 12 days. The eligible vendor pool is real and verifiable.
What does FY2026 specifically change?
The timing math is unusually tight this year. FY2026 opened with the longest full government shutdown in U.S. history — 43 days, October 1 to November 12, 2025 — followed by further partial and DHS-specific lapses into spring. All twelve FY2026 appropriations bills were finally enacted by April 30, 2026. Two consequences:
Funds are unusually fresh. Many civilian agencies didn’t receive full-year apportionments until well after October 1 — some not until spring 2026 — compressing a 12-month obligation calendar into roughly four to eight months. The year-end spike will likely be sharper than normal, with more dollars chasing fewer working days.
Contracting capacity is thin. Acquisition-workforce vacancy rates are approaching 40% at some major agencies. Fewer COs, more dollars, less time — the exact conditions where a fast, defensible statutory authority is both most necessary and most justifiable.
Practical calendar: what’s still achievable before September 30?
Assuming a $1M–$4.9M services AI scope, an SBA-certified SDVOSB already identified, and a CO willing to use the authority — the realistic latest-start dates:
-
May–June (now)
Everything is open. Run market research (FAR Part 10), draft the PWS or SOO, build the IGCE, route the J&A through legal, synopsize per FAR Part 5. Target award by early August; performance plausibly begins in FY2026, so bona fide needs is uncontroversial.
-
July
Still viable. SDVOSB sole-source PALT is typically 30–60 days with a complete package — but the package needs to be in legal review by mid-July.
-
August
Tight. Achievable only if market research is documented, the IGCE is built, and the firm is VetCert-certified and SAM-active. Scope must be a defined non-severable deliverable to keep FY2027 performance clean.
-
First half of September
New stand-alone awards are largely out of time. A task order off an existing SDVOSB-held IDIQ, BPA, or Schedule (with an order-level set-aside) is the realistic move — see which contract vehicle should you buy AI on? for the vehicle decision.
-
Last two weeks of September
Micro-purchases, simplified-acquisition orders, or orders off existing SDVOSB vehicles only. New contract awards in this window are exactly the ones the research flags as lowest-quality.
Frequently asked
Can I use expiring one-year money for AI work performed mostly next year?
Is "the money expires" a valid justification to sole-source?
Will this actually count toward our small-business goal?
How late can I start?
What's the biggest risk?
Working with Truvisory
If you’re a CO, PM, or OSDBU specialist with FY2026 AI dollars to obligate before September 30, the cleanest version of this is a fixed-scope, non-severable deliverable from a verified SDVOSB that ships inside the fiscal year. That’s the Truvisory model — working software in federal-grade environments, delivered in 30-to-90-day increments, scoped to a defined output you can obligate against with confidence.
Verify all of it in SAM.gov and at search.certifications.sba.gov before you write the J&A. When you’re ready, start with a scoping call — and see the $5M sole-source how-to for the award workflow, the FAR Part 19 direct-award piece for the regulatory backdrop, and the federal AI modernization pillar for the full path.