Buy vs build: what 2.17 million SBA loans actually say

Everyone romanticizes the startup. The garage, the founder myth, the zero-to-one.

The data tells a quieter, more profitable story.

The SBA publishes the loan-level record of its two flagship programs, 7(a) and 504. That's 2.17 million loans since 1991. On newer vintages, each loan says what the borrower was doing: buying an existing business, running one, or starting one from scratch. And each loan says how it ended.

We sorted every resolved loan by that field and counted the failures. The result isn't close.

Buying a business fails at roughly a third of the startup rate. Same lender, same loan program, same economy. The only variable is whether there was already a working business underneath the loan.

Why the gap is so wide

It's not luck. It's structure.

• Revenue exists on day one. You're not praying for product-market fit. You inherited it, with customers who've been paying for 15 years.
• The systems are already built. Pricing, suppliers, staff, a name people trust. You're optimizing a machine instead of welding one from scratch.
• The bank can underwrite reality. Lenders see real cash flow, so the loan is structured against something that actually throws off money to service it.

Starting from zero, you carry every risk at once: demand, operations, hiring, brand, and debt on top. Buying, you carry one. Run well what someone already proved works.

The catch nobody mentions

Here's the part that keeps most people on the sidelines: the good businesses aren't listed.

The owner who's quietly ready to sell, 60-something with 20 years in and no kid to take over, isn't on a marketplace getting bid up by ten other buyers. They haven't told anyone. By the time a business hits the broker sites, it's been picked over, marked up, or it's the one nobody wanted.

So the real edge isn't deciding to buy instead of build. The data settles that part. The edge is finding the established, profitable, owner-ready business before everyone else does. That's a data problem, and it's the one we built Mainferret to solve.

What to do with this

1. Stop comparing "start a business" to "buy a business" as equals. One fails at 5.4%, the other at 14.6%. Anchor your decision there.
2. Buy established. Failure risk falls with every year a business survives. The 15-year-old company that still hums is a different asset class from anything under two years old.
3. Go off-market. The listed deals are the leftovers. The owners worth buying from are reachable, if you know which signal points to them.

How we computed it, honestly

Every number above comes straight from the public record. No survey, no vendor data.

• Source: the SBA's official 7(a)/504 FOIA loan-level dataset (data.sba.gov), files as of March 2026. 2.17 million loans, FY1991 to present.
• The business-age field uses today's four categories on roughly 595,000 loans. SBA adopted these labels broadly around FY2018, partially back to FY2004, so this cut skews to recent vintages. Older vintages use a different vocabulary with no acquisition category, so they can't answer buy vs build.
• A loan counts as failed if the SBA charged it off. The rate is charge-offs divided by resolved loans (paid in full or charged off). Cancelled loans never disbursed, and loans still being repaid are too young to judge, so both are excluded.
• The exact levels move a little depending on how you slice vintages. The ranking never does. In every cut we ran, acquisitions were the safest category and startups the riskiest, with acquisitions failing at roughly a third to half the startup rate.
• The analysis is reproducible: one command over the public files. Full query logic and caveats are in our methodology.