How this was made
Methodology and sources
Every figure on this site traces to a named public file with a retrieval date. This page says which files, what was removed from them, where the judgement calls were, and the specific ways this data misleads people who read it quickly.
Last reviewed 4 September 2026
1. The datasets
Primary: SBA 7(a) FOIA loan-level extract. Published by the United States Small Business Administration at data.sba.gov/dataset/7a-504-foia. Two files were used, both stamped as of 30 June 2026: FOIA_7a_FY2010_FY2019_asof_260630.csv and FOIA_7a_FY2020_Present_asof_260630.csv. Retrieved 4 September 2026. Together they contain 934,089 loan approval records. This is a public federal record and the SBA rotates the filenames quarterly, stamping each with its as-of date.
Reference: 2022 NAICS codes and titles. United States Census Bureau, 2-6 digit 2022 Codes. Retrieved 4 September 2026. Used for sector and industry-group titles so that names are the official ones rather than paraphrases.
Reference: state names and USPS codes. United States Census Bureau, state.txt. Retrieved 4 September 2026. The SBA file carries two-letter codes only.
Secondary: announced-transaction comparables. A set of 10,212 announced transactions maintained by QEIP LLC, snapshot 2026-08-18. Used on a single page, the multiples page, and only in sector aggregate. No individual transaction, buyer or seller from that set appears anywhere on this site.
Programme reference. The $5,000,000 maximum 7(a) loan amount is stated by the SBA at sba.gov/funding-programs/loans/7a-loans, which also lists changes of ownership as an eligible use.
2. What counts as an acquisition
The SBA records a BusinessAge field on each 7(a) approval. One of its values is Change of Ownership. A loan carrying that value was approved to finance the purchase of an existing business, which the SBA confirms as an eligible use of the programme. Those rows, and only those rows, form the spine of this site.
This is the whole reason the site exists. The SBA file is normally read as lending data. Filtered to change of ownership it becomes something rarer: a near-complete public register of completed small business acquisitions, each with a price component, a term, a rate, a lender, an industry, a county and an eventual outcome.
No other BusinessAge value is included. Loans coded as startup, new business, or one of the several existing-business age bands are expansion or working capital lending and are not acquisitions.
3. Why the record starts at FY2019
The SBA did not use the Change of Ownership code before FY2018. Every 7(a) approval from FY2010 through FY2017 carries zero instances of it. The whole business-age vocabulary changes during FY2018, which contains both the old scheme and the new one at once, and in which Unanswered rises to 8,779 records. FY2019 is the first year in which the coding is stable.
This matters more than it sounds. A chart of change-of-ownership approvals starting at FY2010 would rise from exactly zero to several thousand a year, and would look like a transformation of the market. It is a change in a form field. Any analysis of this file that runs a trend from FY2010 is wrong, and the error is invisible unless you check the codes by year.
Even within the usable window there is a residual effect: FY2019 carries 9,191 Unanswered business-age records against 3,919 in FY2020, so the FY2019 acquisition count is likely to be a slight undercount relative to later years. Comparisons across years should be read with that in mind.
FY2026 is a partial year. The extract closes 30 June 2026 and the SBA fiscal year ends 30 September, so roughly a quarter of the year is missing. Every chart on the site marks that column and every page carries the note. The final column is never a decline.
4. Approvals, cancellations and disbursement
An approval is not a transaction. Of 46,696 change-of-ownership approvals from FY2019 onward:
- 40,612 funded. Loan status of
PIF(paid in full),CHGOFF(charged off) orEXEMPT(active). These are treated as completed acquisitions and are the basis of every figure on the site. - 3,965 cancelled. Status
CANCLD. Approved and then never disbursed. Excluded from every measure except the cancellation rate, which is published as a finding in its own right. - 2,119 committed. Status
COMMIT. Approved but not yet disbursed, overwhelmingly recent approvals. Excluded from both the numerator and the denominator of the cancellation rate, so that recent years are not penalised for loans that simply have not funded yet.
Rows with a funded status but a gross approval of zero or less are dropped as data errors.
The cancellation rate is therefore cancelled divided by cancelled plus funded, and it is 8.9% nationally.
5. How each measure is computed
Deals is a count of funded loans. Capital is the sum of GrossApproval, which is the total loan amount including the guaranteed portion. It is not the purchase price and not the guaranteed amount.
Median, quartiles and deciles are computed on the full sorted array of gross approvals for the group with linear interpolation between adjacent ranks, not on a sample or a bucketed approximation.
Median rate uses InitialInterestRate, the rate at approval. Values outside 0.5% to 30% are treated as data errors and excluded. Most 7(a) paper is variable, so this is the rate at origination and not the rate paid over the life of the loan. A median is published only where at least ten loans in the group carry a valid rate, and each page states how many did.
Median term uses TermInMonths, restricted to 1 to 360 months. Published only where at least ten loans carry a valid term.
Term bands. The distribution is not smooth. 25,192 loans run exactly 120 months and 6,907 exactly 300 months, with very little between them. Those are the programme's maturity ceilings, which depend on what secures the loan, so the site groups terms as under ten years, exactly ten years, ten to twenty, and twenty or more, and reads the longest band as an indication that real property was financed alongside the business. That is an inference from the shape of the distribution and is described as one wherever it appears.
Jobs is the sum of JobsSupported, a figure reported by the lender at approval. It is an estimate made before the transaction completed and should be read as such. Franchised share is the proportion of funded loans carrying a non-empty FranchiseName.
6. Charge-offs and why they are a floor
Charge-off is reported for a single fixed cohort: loans approved in FY2019, FY2020 and FY2021, of which 16,602 funded and 455 had charged off as of 30 June 2026. That is 2.74%.
This is a cumulative share at a point in time, not a lifetime default rate, and the difference is large. Those loans are five to seven years into terms that are usually ten years, so a substantial part of their risk period has not happened yet. The lifetime figure will be higher. The number published here is a floor.
The site deliberately does not publish charge-off rates for FY2022 onward. Those cohorts are younger still, and a naive comparison across approval years would show charge-offs falling to almost nothing, which reflects nothing but the passage of time. A group is only given a charge-off figure where its cohort contains at least 100 funded loans.
7. Reporting floors and what has no page
A page exists because enough rows exist behind it to say something that is not noise. The floors are: 25 funded acquisitions for a sector, 25 for a four-digit industry group, 25 for a state, 30 for a county, 50 for a lender, and 30 for a state and sector combination.
That produces 19 sector pages, 147 industry group pages, 52 state pages, 315 county pages, 151 lender pages and 328 state-by-sector pages.
Below the floors, and named rather than silently dropped:
- Sector 92, Public Administration: 6 acquisitions.
- U.S. Virgin Islands: 12 acquisitions.
- Guam: 1 acquisitions.
- 158 further four-digit industry groups.
- 2,103 further counties.
- 1,434 further lenders.
- 531 further state and sector combinations.
These rows are still counted in every parent total. They are excluded from having a page of their own, not from the data.
8. Geography and lender names
A loan is assigned to ProjectState, the location of the business being acquired, falling back to BorrState where the project state is missing or not a recognised code. County uses ProjectCounty, normalised to title case. County names in the SBA file are not accompanied by FIPS codes, so counties are matched on name within state. Where a state contains two similarly named counties the file does not distinguish them and neither can this site.
Lender names are taken from BankName and lightly normalised: trailing National Association, N.A., FSB, SSB and corporate suffixes are stripped, and whitespace is collapsed. Without this one institution appears as several. No attempt is made to consolidate banks that merged during the period, so a lender acquired in 2022 may appear under both its old and new names. Rankings should be read with that in mind.
9. The disclosed-multiple layer
One page uses a second dataset. Of 10,212 announced transactions, 826 disclose an EBITDA multiple, and 811 carry both a multiple and a deal value after removing multiples above 60x as data errors. Sectors with fewer than 20 usable observations are named but not given a quartile range.
That page exists to make a corrective point rather than to supply a valuation. The median disclosed transaction in the set is worth $97M and only 6.2% are under $5M, so the multiples are earned by companies far larger than the businesses this site otherwise describes. The median deal value is published in the same table row as the multiple for exactly that reason.
10. What was considered and left out
SBA 504 loans. The 504 programme finances fixed assets, principally real estate and heavy equipment, through a certified development company. Its extract does not carry a business-age field that identifies changes of ownership, so it cannot be filtered to acquisitions the way the 7(a) file can. Including it would have added volume that is mostly not acquisition activity.
7(a) loans before FY2010. The 1991-2009 extracts contain no change-of-ownership coding at all and their column sets differ.
A firm-count denominator. Dividing acquisitions by the number of firms in a sector would give an apparent turnover rate, and Census County Business Patterns would supply the denominator. It was left out. SBA-financed acquisitions are one financing channel among several, so the ratio would be a fraction of an unknown whole, and the NAICS vintages of the two files do not align cleanly. A plausible-looking ratio that is quietly wrong is worse than no ratio.
Individual loan records. The SBA file contains borrower names and street addresses. Nothing at that grain is published here. Every page on this site is an aggregate of at least 30 loans, and usually far more.
Purchase prices. They are not in the file and are not estimated. Anywhere a page might be read as giving a sale price, it says explicitly that the figure is a loan amount.
11. What this data cannot tell you
It is not the whole market. It is one financing channel, capped at $5,000,000 per loan. Acquisitions funded conventionally, by strategic buyers, or by investment firms are absent entirely. The site is a complete census of a defined slice, not a sample of the whole.
Loan amounts are not prices. A buyer combines bank debt with their own equity and frequently with a note from the seller. The SBA records only its own loan. Enterprise values are higher than the loan amounts shown here, by an amount this data cannot measure.
It cannot value your business. A median across hundreds of businesses in an industry group says nothing about any one of them. Nothing on this site is a valuation, and no page should be used as one.
Reported fields are reported, not verified. Jobs supported, franchise name and NAICS code are entered by the lender. NAICS in particular is sometimes approximate, and a business can be classified into a group that does not quite describe it.
12. Reproducing this
The two ingest scripts in the repository read the raw files and write the derived JSON that every page imports at build time. No page queries a database and no figure is computed at request time, which is also why every page here is static.
The derived files are committed so that a build is reproducible without re-downloading 436MB of federal CSVs. The raw downloads are not committed. Re-running the ingest against a newer SBA extract regenerates everything, and the as-of date stamped on every page comes from the file rather than being typed in.
If you find an error, the corrections page explains how to report it and how corrections are recorded.
Related: Corrections, Disclosures, About, Terms, Privacy.
When the file changes
The SBA restates its 7(a) extract periodically. When this site is rebuilt on a newer extract, subscribers get one email describing what moved, what was added and what fell below the reporting floors. That is the entire list: occasional, tied to the data, and nothing else is sent.
This is a reader list. It is separate from the buyer register on the contact page, and neither one feeds the other.
Your address is used for this list and nothing else. Every email carries an unsubscribe link. Privacy.