Methodology

The source

All loan statistics come from the U.S. Small Business Administration’s 7(a) & 504 FOIA dataset, a loan-level record of every loan the agency has guaranteed. It is public-domain federal data, refreshed quarterly. Our current build uses the extracts published as of 2026-06-30.

We resolve the file URLs from the SBA’s dataset page at ingest time rather than hardcoding them, because the SBA stamps the as-of date into each filename and rotates them every quarter.

What we include, and what we leave out

• Fiscal year 2010 onward only. The SBA publishes back to 1991. We exclude the older extracts because their column sets differ and because lending conditions before the financial crisis tell you little about who will fund your business now.
• Both programmes. 7(a) and 504 are combined in totals. Where a metric only exists for one - the extract carries a borrower interest rate for 7(a) but not for 504 - we say so next to the number and compute it from 7(a) alone.
• Approvals, not disbursements. The dataset records approvals. Some approved loans are never drawn. Approval volume is the standard measure used by the SBA’s own reporting and by the industry, so it is what we use.
• Zero-dollar and malformed rows are dropped. As are interest rates outside a 0.5% to 30% band and terms above 360 months, which are data-entry errors rather than loans.

After those filters the current build covers 1,052,071 loans totalling $507,095,985,191.

Lender name normalisation

The same institution appears under several spellings across years - “TD Bank, National Association”, “TD BANK NA”, and “TD Bank N.A.” are one bank. We collapse unambiguous legal suffixes (national association, N.A., FSB, SSB and similar) so that a lender’s history aggregates into one profile.

This is deliberately conservative. We do not strip geographic or descriptive words, because there are many genuinely distinct banks called “First National Bank” and merging them would be worse than leaving a large bank split across two profiles. Some fragmentation therefore remains, particularly among small community banks and among institutions that have been acquired - an acquired bank’s pre-acquisition loans stay under its own name.

Franchise brand normalisation

The SBA records franchise brand as free text, and the conventions changed over time - older vintages shout (“SUBWAY SANDWICH SHOP”), newer ones do not (“Subway”). We normalise case and punctuation, strip contract boilerplate and parenthetical qualifiers, take the leading segment of slash-joined labels, and apply a small hand-maintained alias table for high-volume brands the SBA labelled two different ways. Administrative placeholder values that are not brands are excluded outright.

This will not be perfect. Some brands remain split; a co-branded location may be attributed to its lead brand. Treat franchise counts as close rather than exact.

Medians, not averages

We publish medians for loan size, rate, and term rather than means. SBA loan sizes are heavily right-skewed - a handful of multi-million-dollar approvals drags a mean far above anything a typical borrower would see. Where a mean is the more informative figure we label it “average” explicitly.

We also publish the 10th and 90th percentiles as a “typical range”, because a median alone hides how wide the distribution is.

Charge-off rate - the number we are most careful with

A charge-off rate is easy to compute badly and defamatory when you do. Ours is defined as:

The share of a lender’s FY2010-FY2019 approvals that have reached a terminal status (paid in full, or charged off) and were charged off.

Two constraints matter. First, the vintage cut-off: a loan approved last year is almost certainly still current, so including recent approvals would flatter every lender that grew recently and punish every lender that did not. Second, the resolved denominator: we divide by loans that have actually ended, not by all loans, so the figure is a realised loss rate rather than a snapshot of an immature book.

Where a lender, industry, or brand has fewer than 25 resolved loans in that cohort, we publish a dash rather than a rate. A charge-off percentage computed on eight loans is noise presented as a fact.

What a charge-off rate is not: a measure of service quality, or a warning. It largely describes the credit box a lender chose. A lender serving startups and thin files will show higher realised losses than one writing large collateralised acquisition loans, and that is a description of its mission rather than a criticism of its competence.

For reference, the national figure on this definition is 7.6% across 463,245 resolved FY2010-FY2019 loans.

Minimum thresholds for a page to exist

We only generate a profile where there is enough history to say something true. A lender needs 25 loans, an industry 40, a franchise brand 10, and a state-and-sector pair 20. Below those thresholds the medians are noise. That currently yields 1,519 lender profiles.

How rankings are ordered

Every league table is a straight sort on the column named in its header - dollars approved, loan count, median loan size, median rate, or charge-off rate. Rate and charge-off sort ascending because lower is better; nulls always sort last. There is no composite score, no weighting, and no editorial adjustment.

Commercial relationships are not an input to any ranking. No lender can pay for position. Paid placement exists elsewhere on the site, is labelled, and is rendered outside the ranked tables. See advertiser disclosure.

Partial fiscal years

The most recent fiscal year in any extract is incomplete. Every chart marks the final bar in a different colour and every comparison of “the last complete year” uses the year before it. A partial year read as a full one looks like a collapse in lending and is not.

Cost bands on funding-type pages

The cost ladder uses editorial bands rather than computed figures, because no public dataset covers merchant cash advance or fintech pricing. They are our judgement, based on published rate ranges across each category, expressed in APR-equivalent terms so that products quoting factor rates and discount fees can be compared with products quoting interest. They indicate relative expense; they are not quotes.

What we deliberately do not do

• No star ratings. A five-star score on a lender implies a judgement we have no basis for. We publish the numbers and let them speak.
• No predictions. We do not estimate your approval odds, your rate, or your eligibility. Anyone who does, without your financials, is guessing.
• No borrower reviews. We do not host testimonials or ratings. Unverifiable reviews in a category this high-stakes attract manipulation, and the FTC’s rule on fake reviews exists for good reason.
• No modelled or estimated data. If we do not have it from the source, we do not publish it.

Verification and corrections

Before each release we cross-check total approvals and dollar volume for the most recent complete fiscal year against the SBA’s own published activity report. If our figures diverge materially from the agency’s, the release does not ship.

If you believe a figure is wrong, write to hello@fundingforbusinesses.com with the page and the number. We will check it against the source file and correct it if we are wrong. We do not remove accurate figures on request.

Regulatory notes we track

Two items bear on how sites like this operate and we mention them because readers evaluating any financing source should know they exist. New York’s Commercial Finance Disclosure Law requires providers of commercial financing at or below $2.5 million to give standardised cost disclosures before a specific offer - if you are offered financing in a state with such a law, you are entitled to see an estimated APR and total cost in writing. Separately, New York Senate Bill S3177, which would license persons making or soliciting commercial financing products in the state, remains pending in committee. Neither obligation applies to us, because we neither provide financing nor solicit it - but both are useful to know when someone else offers you money.