When the numbers don't add up

How new Scorecard math and AI glitches are squeezing 8(a) firms

Photo by Aaron Lefler on Unsplash

Posted on LinkedIn by the former State Department OSDBU Director, George Price, SBA has a new Scorecard methodology. SBA reportedly shared the methodology with other agencies to show how the new Scorecard will assign grades, starting with FY26:

High Level FY26 Scorecard Factors
• Prime Contract awards to Small Businesses – 50%
• Subcontracts Awards to Small Businesses – 25%
• Serving our Veterans – 15%
• Providing Competitive Value to the Taxpayer – 10%
• Partners with SBA to reduce fraud – 10%

Take a second look at those numbers. They don’t add to 100. Maybe SBA is echoing Don Zimmer and telling agencies to “give 110%” for small businesses. Never mind that, the math isn’t actually important here. I’ll get to where math matters a lot more —errors in 8(a) suspensions.

But first, the Scorecard: The changes are not likely to make a big difference.¹ GAO found that prior changes to the Scorecard methodology had “little impact on small business procurement opportunities.” More significant, GAO wrote, are Category Management policies that result in fewer opportunities for small businesses. And these numbers aren’t contracting goals themselves. They are a formula that SBA uses to weight those goals and, in the end, assign a letter grade to agencies’ efforts to prioritize small-business contracting.

The new formula centers small-business GovCon around veterans. Veterans already are counted in the first two categories—prime contracts and subcontracts. Now they have their own category, “Serving our Veterans,” that evaluates growth in spending with veteran-owned businesses. SBA might be overcorrecting on the close call on the 5% service-disabled veteran-owned goal—more on that below. The new math makes veterans, all told, about 25% of the Scorecard (that’s the new 15% category, plus 5% from prime, 2.5% from subcontracting, and an unknown amount from the competitive-value category). Will it matter? The much bigger mover of veteran-owned contracting is the VA. The reason veteran-owned spending dipped in 2025 wasn’t the Scorecard. It was because DOGE cut so many veteran-owned contracts at the VA.

The new competitive-value metric discourages agencies from using 8(a) sole-source contracts. That’s consistent with what the FAR Overhaul has done in 8(a). SBA might go further in the forthcoming “Fraud, Waste, and Abuse Reforms” proposed rule. SBA’s publicly stated rationale is that taxpayers get the most value when agencies hold competitions, instead of going straight to sole-source contracts.

Let’s be clear: what SBA is doing won’t make a dent in the use of sole-source contracts. Most sole-source contracts are awarded outside of the 8(a) program. Even if you got rid of 8(a) sole sources entirely, the government would still have awarded $266 billion in sole-source contracts in 2025:

Of course, the publicly stated rationale isn’t really what’s going on. SBA is trying to shrink the 8(a) program. That’s why the agency still hasn’t approved an 8(a) application since last August. But, in changing the Scorecard methodology, the agency is assuming those 8(a) sole-source dollars will go to other small businesses, preferably veteran-owned small businesses.

I don’t see why that would happen. Agencies aren’t using 8(a) sole source because of DEI or the Scorecard. They want to finalize contracts quickly, and 8(a) allows them to do that. But 8(a) isn’t the only way to do that. There’s also OTAs, CSOs, SBIR Phase III, and plain old limited-source justifications. Take a look at the top recipients of sole-source contracts; there aren’t any 8(a) firms on there:

So, as SBA narrows the path to 8(a) sole source—whether by the Scorecard or through regulatory changes—agencies are not likely to respond by using veteran-owned small businesses. Instead, they may bypass small businesses altogether.

Where bad math does damage

While math on the SBA Scorecard is not a big deal, it is elsewhere. In SBA’s fraud enforcement, the SBA’s AI is failing at math.

SBA sent suspension notices to hundreds of 8(a) firms, accusing their owners of not being economically disadvantaged. But most of them are. The concept of economic disadvantage would seem to be an area where AI enforcement could work. Economic disadvantage uses three objective thresholds—$850,000 for net worth, $400,000 for three-year average personal income, and $6.5 million for assets. The information to determine economic disadvantage comes from tax returns and financial statements.

And yet. The companies receiving these letters say that the calculations are wildly off. The AI doesn’t split joint assets. It doesn’t apply regulatory exceptions.

In some cases, the companies had been investigated previously and received notices from SBA in 2025 that they were, in fact, economically disadvantaged. But then the AI ignored that past decision. It went ahead with no new facts. And it suspended the companies despite the work that SBA staff—human beings, mind you—had already completed.

This wouldn’t be a problem if SBA used safeguards. Sometimes AI counts to 110 instead of 100. That’s what hallucinations are. But a human checks the math. Or the human looks at the company’s history to see that it’s already been cleared based on the same facts.

The way SBA is deploying its AI is doing immense damage. Because the 8(a) suspension takes immediate effect, companies are losing contract awards. They are in danger of being pulled off of big GWACs, like OASIS+ and 8(a) STARS III.

Meanwhile, press releases are smearing each company as a “fraudulent participant that took opportunity away from legitimate and eligible small business owners at taxpayers’ expense.” But it’s the AI that’s finding totally legitimate companies and inventing fraud where there isn’t any.

This problem is only going to get worse. The new SBA Inspector General is looking closely at pandemic loans, of which there are millions. The Senate-passed SBIR reauthorization bill lets agencies deny an SBIR award because of a “security risk for any reason.” AI tools are very likely to be used for both purposes.

What needs to happen is more human oversight of these tools. We’ve already seen they aren’t so great at counting. I’d rather not see what else they can get wrong.

A correction on disabled-veteran contracting

Speaking of math, I have to correct some of my own. A few months ago, I predicted that the Federal government would miss the 5% goal for disabled veteran-owned contracting for FY 2025. But I called it too early. Since my prediction, the data has changed. New data added about $60 million in disabled veteran contracts and removed a lot of other contracts. That raised the disabled veteran-owned contracting percentage from 4.71% to 4.87%.

That’s still not 5%. But 4.87% is close enough that, with more data that gets added every year, the government will probably surpass that 5% goal for 2025. I don’t know where that $60 million came from. But I think it’s more likely that it is late-arriving data instead of anything nefarious.

The women-owned and HUBZone percentages—at 4.3% and 2.55%, respectively—are still way too low to have a shot of meeting the 5% women-owned and 3% HUBZone goals.

One other thing is for certain: The number of small businesses participating in contracting went down again in 2025. This should be no surprise: 2025 was the year of DOGE, USAID’s contraction, and consolidation memos. Small businesses continued to leave contracting.

Whether they come back is uncertain. SBA isn’t helping matters with its roughshod enforcement push. But the agency can still turn things around.

All it takes is giving 110%.

---

With 20 years of Federal legal experience, Sam Le counsels small businesses through government contracting matters, including bid protests, contract compliance, small business certifications, and procurement disputes. His website is www.samlelaw.com.

This article is for informational purposes only and does not constitute legal advice.

1. In comments on George’s post, some people asked why they couldn’t find this on the SBA website. SBA doesn’t typically publish the Scorecard methodology until the Scorecard itself is released. So this one wouldn’t come out until mid-2027. The current version on the SBA website is 2024’s methodology.