USDA Business and Industry Loans in Public Data: What USAspending Shows and What It Hides

A lender sizing a rural business loan, or a borrower preparing a USDA Business and Industry (B&I) application, will sooner or later want to know how much of the program has gone to a given state or county, and in which years. USDA's own web pages look like the place to find out, but the public record that anyone can query, download and rerun lives elsewhere: in USAspending, the federal spending site. This article sets out what that record holds at its true grain, which of its fields can be trusted for which purpose, why its totals drift away from USDA's own budget figures, and what a feasibility study can and cannot draw from it.

The method described here is the one behind the USDA tables of the Loan Analytics loan dataset. Those tables were rebuilt from USAspending downloads pulled on November 10, 2026, and every B&I count and dollar figure printed below is rendered from the dataset's committed manifests rather than typed. Where a figure comes from USDA's budget tables, it is a secondhand transcription made by the site, not a primary copy, and the text says so each time it appears.

Where the Public B&I Record Lives, and Why USAspending Is the Route

The legal root is the Federal Funding Accountability and Transparency Act. It requires one searchable website, free to the public, carrying a defined set of data elements for each federal award ^[1]. The statute counts loans among federal awards, alongside grants and cooperative agreements ^[2], so a loan guarantee such as B&I appears on that site because the law puts it there, not because an agency chose to publish it. The same section asks the site to hold data from fiscal year 2007 onward ^[3]. USAspending is that site, and its programming interface is open: its documentation states that the endpoints need no authorization ^[4]. A study that cites a USAspending figure is therefore citing something a reviewer at the lender can rerun without an account, a password or a fee.

Programs are filed on the site under assistance listing numbers, the numbers assigned to each assistance program in the government's program catalog and on SAM.gov ^[5]. B&I is listing 10.768, which USAspending's reference list titles Business and Industry Guaranteed Loans ^[6]. Within a listing, every award also carries an assistance type, and B&I sits under the guaranteed or insured loan type, which the publisher's dictionary describes as an arrangement in which the government indemnifies a lender against part or all of the defaults on a loan ^[7]. Two other Rural Development programs, Community Facilities and the Rural Energy for America Program, come in the same download with the same structure, and the Loan Analytics tables hold all three. This article stays with B&I.

When the dataset was built, the route had to be USAspending. USDA's own hosts, www.rd.usda.gov and www.usda.gov, turned the dataset's automated fetcher away at the network edge, and the dataset treats such a refusal as final rather than looking for a way around it. Two consequences run through the rest of this article. First, every count and dollar figure comes from the federal spending record, not from USDA's program pages. Second, the budget totals used here as a yardstick come from the site's own transcription of USDA's explanatory notes, which is secondhand material. A reader who has USDA's budget justification in hand can check those totals against the source; the dataset could not, and it does not pretend otherwise.

The transactions download is the working instrument. USAspending's endpoint list describes it as generating a zip file of award data in CSV format ^[8], and that file, not the website's summary pages, is what the dataset reads. The summary pages roll transactions up to the award and lead with an obligation figure that, for a loan, is not the loan, as the section on face value and subsidy cost explains. Readers who want states ranked on these data will find a ranking in the archived index of B&I lending by state. The present article is about the record underneath any such ranking, and about the checks that record has to pass before a number drawn from it can stand in a study.

The Grain: One Record per Award Transaction

The first question to settle about any public loan file is what a row is. In the USAspending transactions download, a row is an action taken on an award, not a loan. The publisher's definition of the action type field says that it records new awards and later changes to a prime award, and that one award can carry several actions ^[9]. The award totals the website shows are sums over those rows: its award subsidy cost, for example, is defined as the sum of the original subsidy costs of every transaction sharing the award's identifier ^[10]. A count of rows is therefore a count of actions, and it overstates the number of loans in any year that saw corrections or cancellations.

Each row is placed in time by its action date, which the dictionary defines as the date the action was signed by the government or a binding agreement was reached ^[11]. It is placed in a fiscal year by that same date, on the federal year that begins on October 1 ^[12]. That is the only time key the dataset uses to assign B&I activity to a year. A study that writes about B&I in a given fiscal year is writing about actions dated in that year, which is not the same thing as loans closed in it, and the section on budget totals below turns on the difference.

For B&I the dataset finds two kinds of action and no others. A new award action carries the loan's face value as a positive amount. A later action, which moves no money, carries a negative face value and takes some or all of the award back out. Netting the two gives the face value that survives today. Counting the distinct awards that have a new award action in a year gives that year's new awards, and that count, never a raw count of rows, is what the dataset prints.

Not every row is a single award. The publisher's record type field separates aggregate records, which pool all awards to individual recipients in a county, state or foreign country so that no individual can be identified, from ordinary non-aggregate records ^[13]. B&I carries few aggregate rows: 301 ^[14] of them against 9,620 ^[14] non-aggregate rows across FY2008 to FY2026, and the aggregate rows sit in the early years of the span. Because an aggregate row hides how many awards it holds, any count of B&I awards in those early years is a floor rather than a census.

The listing also holds more than one program stream. The free-text transaction description separates core B&I, the CARES Act B&I loans of fiscal years 2020 and 2021, and the Food Supply Chain Guaranteed Loans of fiscal years 2022 and 2023. All three share listing 10.768, and nothing but that description tells them apart. The Loan Analytics measures therefore keep a core series and an all-streams series side by side, because USDA's budget line for B&I, as the site transcribed it, does not include the food supply chain loans. Table 1 lists the fields a B&I row carries and how often each is filled across the whole span.

Three readings of that table matter for a study. Face value is filled on almost every row and the subsidy cost on all of them, so the two dollar measures are dependable. The obligation column is never nonzero on a B&I row, which is why a total built on it is not a lending total. And the state and county fields, which look incomplete across the whole span, are in fact full on every non-aggregate row from FY2017 on, the window in which the dataset's state tables are built; the shortfall sits mainly in the first two fiscal years of the span, when many rows were published without a place of performance state.

Loan Data Behind This Article

Every figure in this table comes from the Loan Analytics loan dataset, analytics-loan-data freeze-3 (commit 360c856), and carries its key. USDA figures are USAspending award aggregates; a USDA rate in the dataset is a credit subsidy rate, not a loss rate.

Face Value, Subsidy Cost, and an Obligation of Zero

Three dollar fields sit on a B&I row, and they measure three different things. The face value field holds the face value of the direct loan or loan guarantee ^[15]. The original loan subsidy cost holds the estimated long-term cost of the guarantee to the government, calculated on a net present value basis and excluding administrative costs ^[16]. The federal action obligation holds the amount of the government's obligation, de-obligation or liability for the transaction ^[17]. On a grant, that third field is the money that changes hands. On a B&I loan it is zero on every row the dataset holds, and the subsidy sits in its own field.

The subsidy cost is a budget concept with a statutory definition. The Federal Credit Reform Act defines the cost of a loan guarantee as its estimated long-term cost to the government on a net present value basis ^[25], built from the payments the government expects to make for defaults and delinquencies, net of the fees, penalties and recoveries it expects to collect ^[26]. The estimate is fixed on the assumptions current for the fiscal year in which the funds are obligated ^[27]. That last clause explains a pattern the dataset finds in the file: in each fiscal year from FY2022 to FY2025, every core B&I award carries the same ratio of subsidy cost to face value, and the ratio moves only when the year changes. The dataset's grain note also finds two different ratios inside FY2021 and inside FY2026, so even this regularity has to be checked year by year rather than assumed.

The practical result is that the subsidy cost follows the year's rate far more closely than it follows lending. The site's secondhand transcription of the budget tables puts the B&I subsidy rate at 2.29 percent ^[22] for FY2024 and 0.20 percent ^[22] for FY2025. In the USAspending record, the gross face value of new awards rose from $2,184.4 million ^[21] in FY2024 to $2,382.6 million ^[21] in FY2025, while the subsidy cost on the same listing fell from $35.7 million ^[21] to $3.9 million ^[21]. A reader who added up the obligation shown on the website's award pages, which for a loan is this subsidy cost, would conclude that B&I lending collapsed in a year in which it grew.

The illustration below makes the point with made-up numbers. It takes one loan and one guarantee percentage, both labelled as assumptions, and shows what each field would hold at an assumed subsidy rate and at one tenth of that rate. The face value does not move. The subsidy cost moves tenfold. The guaranteed share, which is the exposure a credit officer usually cares about first, appears in no field at all.

Rounded as shown; lines may not foot by one unit in the last place.

Table 4 also shows why the subsidy cost cannot be read as a loss rate. The ratio in its fifth line is a forecast made when the funds were obligated, under that year's assumptions; it is not the share of loans that later went bad, and the file carries no field from which anyone could compute that share. Its sixth line is a reminder that the rate is stated against the face value of the whole loan, not against the guaranteed portion, so the two percentages part company whenever the guarantee is partial, which under the rule's ceiling on the guarantee it always is ^[28].

Face value is the gross loan, not the guaranteed share, and the data settle the question without USDA's own pages. The rule caps what any single borrower may hold in B&I guaranteed debt at $25 million ^[29], a ceiling the Secretary may lift to $40 million, but only for loans to "rural cooperatives that process value-added agricultural commodities" ^[30], and it caps the guarantee at 90 percent of eligible loss ^[28]. If the field held the guaranteed share, an award recorded at exactly the loan limit would be close to impossible, yet the dataset's grain note finds a cluster of core awards recorded at exactly that limit, which is what loans written to the cap look like and not what a guaranteed slice of them would look like. The face value measure also lines up with the budget's loan principal once timing and streams are aligned, as the section on budget totals shows, and it could not do that if one side counted loans and the other counted guarantees.

Place: County Is as Fine as the File Goes

The transparency statute asks for two locations on every award: where the recipient is, and the primary place where the award is performed, down to city, state, congressional district and country ^[18]. USAspending carries both, and on a B&I row they need not agree. The recipient's county is where the recipient is located; the place of performance is where the funded activity happens. For a market study the second is the one that matters, and it is the one the Loan Analytics state tables use.

How fine is that place? The publisher assigns each award a place of performance scope, and the county-wide scope means that the predominant performance spans several places within one county, up to the entire county ^[19]. B&I rows carry that scope or leave it blank, so the county is the finest level the file supports. Across FY2008 to FY2026 the non-aggregate B&I rows show a place of performance county on 93.16 percent ^[14] of rows, a city on 51.22 percent ^[14] and a ZIP code on 51.43 percent ^[14]. The city and ZIP shares look like a finer grain than the county, but they sit on only part of the rows, under a scope that says the activity covers a county, and a study should not map B&I activity by town or by ZIP code from them.

Where it counts, the county field is complete. From FY2017 on, state and county are filled on every B&I row, which is why the dataset's tables begin in that year. The gap before then is concentrated in fiscal years 2008 and 2009, when many B&I rows were published with no place of performance state; the dataset keeps those dollars under a state code of NONE rather than guess where the projects were. The recipient's county, by contrast, is filled on 100.00 percent ^[14] of the non-aggregate rows, and a study that fell back on it for the early years would be mapping recipients' addresses, not projects.

Congressional districts deserve a caution of their own. The district code as originally reported is a real district on 98.45 percent ^[14] of rows; the rest include a placeholder the publisher uses for places that straddle more than one district. USAspending also derives a current district of its own, and that derived code is filled on 51.30 percent ^[14] of rows across the whole span. A district total is therefore a construction built on another construction, and it belongs in a study only with the code version named beside it.

None of these fields says whether a place is rural. B&I eligibility turns on the rural area definition in 7 CFR 5001.3, whose starting point bars every town or city with more than 50,000 residents, and also the built-up urbanized fringe that adjoins each of them ^[31]. That starting point is not the whole test. The definition also removes certain populations under 7 U.S.C. 1991(a)(13)(H) and is applied on the latest decennial census ^[32]. It then lists exclusions from the urbanized-area bar, among them an area the Under Secretary of Rural Development has determined, for one project at a time, to be rural in character ^[33][34], and an area joined to that kind of urbanized area only by a strip of urbanized census blocks no more than two blocks wide, for which the applicant should ask the Rural Development State Office for a determination ^[35]. The file records where a project is, not whether the project passed that test, and a study must run the test on the site itself rather than infer eligibility from earlier awards nearby.

What the File Leaves Out: Industry, Lender, Decision and Outcome

The fields a B&I row lacks matter as much as the ones it has, and the reasons are structural rather than careless. Start with industry. The statute lets the website identify an award either by its North American Industry Classification System code or by its program catalog number, where applicable ^[36]. Contracts take the first route and assistance awards the second. The publisher's dictionary bears this out: it defines the NAICS element as the industry code of a contract solicitation and award and maps it to the contract download files, not to the assistance files ^[20]. The dataset's column detector found no industry column of any kind in the assistance download, while the same detector, run on the contract columns, finds NAICS at once.

Two consequences follow for B&I. Nothing in USAspending says what share of the program went to hotels, food processing or manufacturing, and industry figures that USDA publishes through its own channels cannot be reproduced from this route. The business type field, which is filled, does not close the gap: the dictionary describes it as a set of indicators of the type of recipient, by socio-economic status and organization ^[37], so it tells a small business from a city or an individual and says nothing about what the business does.

Next, the lender. The program runs entirely through lenders. Under 7 CFR 5001 the lender is the entity the Agency has approved to originate, service and collect payments on the guaranteed loan ^[38], and a guaranteed loan is one that the lender makes and services under a lender's agreement once the Agency has issued its loan note guarantee ^[39]. Even the publisher's definition of the loan type speaks of indemnifying a lender ^[7]. Yet no lender field reaches the B&I row. The SBA loan files behind the rest of this series name the lender on each loan, which is why the SBA side of the Loan Analytics dataset can show lender concentration and the B&I side cannot.

Then the decision. The B&I record begins at the obligation. Nothing in it shows an application, a request that was turned down, or a commitment that was offered and refused before the funds were obligated, so no approval rate or decline rate can be computed from it. A study that wants to say how hard B&I is to obtain in a region needs a different source, because this record cannot supply the answer.

Finally, the outcome. Lenders report to the Agency at least twice a year, as of June 30 and December 31, on the condition of their guaranteed portfolio, including borrower status and loan classification ^[40], and they file a monthly default report for each guaranteed loan in monetary default ^[41]. Those reports exist, but they do not travel to USAspending. The award record has no field for delinquency, repurchase, liquidation or loss. That is the sharpest difference from the SBA files, which carry each loan's status and charge-off, and it is the reason Loan Analytics prints approval-cohort charge-off floors for the SBA programs and prints none for B&I. The limits of the SBA file are set out in the series article on the SBA file's limits; for B&I the limit arrives earlier, because the outcome of a guaranteed loan never reaches this record at all.

The Span, and a Current Year That Is Not Finished

The statute asks for data from fiscal year 2007 onward ^[3], and a message from USAspending's own search endpoint suggests that downloads reach back further still. The download the dataset relied on did not: it refused any request that started before October 1, 2007, and an attempt to reach back to the year 2000 came back as an error. The usable B&I series therefore begins with FY2008 and runs, by action date, to FY2026. A request covering FY2027 to date returned nothing for any of the three listings.

FY2026 is not a finished year in this record. The latest action in the pull is dated August 31, 2026, the latest first report falls in late September, and the pull holds no B&I action dated in the first quarter of the fiscal year, October to December 2025. The year's net face value to date is $1,024.4 million ^[22], on 119 ^[21] new awards, against a budget estimate of $2,091.7 million ^[22] in the site's secondhand transcription, a gap of -51.00 percent ^[22]. That gap cannot be read as a fall in the program: the record holds no action in the first quarter and nothing after August, and it is set against a budget estimate rather than a result, so the file alone cannot say how much of the gap is timing and how much is lending. A study written in late 2026 should print FY2026 as a partial year, or leave it out of any trend line.

Table 5 gives the B&I series from FY2017, the first year of the window in which the state and county fields are complete. The fiscal year measures print the FY2026 net face value as $1,024.4 million ^[21], the same figure the reconciliation table carries for that year.

Read across a row, the table shows the measures pulling apart. Gross face value counts every new award action at its full amount. Net face value today subtracts every cancellation the file has absorbed since, however late it arrived. The December 31 column, explained in the section on the reconstruction, holds the year as it stood three months after it closed. FY2021 shows the widest spread between gross and net, the year in which the CARES Act stream ran beside the core program, and the all-streams and core columns differ only in fiscal years 2020 to 2023, while the two temporary streams were active.

The FY2017 entry in the December 31 column sits far below that year's gross face value, and the reason lies in the record, not in the program. The first report dates the file carries begin in September 2017, so the reconstruction has little to work with for that year and nothing at all before it. In FY2019 and FY2021 the December 31 reading also sits below the net face value today, which can happen only when a large block of the year's new awards was first reported after the cut. The reading therefore describes what had been reported by that day, not what had been lent, and the dataset has tested it against the budget only for FY2022 to FY2025.

Why a Plain Face-Value Sum Misses USDA's Budget Totals

The obvious way to total B&I for a year is to add up the face value on every row dated in it. That sum misses USDA's budget figures by margins too wide to wave away, and the reason lies in how a B&I guarantee comes into being, not in any flaw of the website.

A B&I guarantee does not begin with a closed loan. When an application is selected, the Agency issues a conditional commitment, which the lender and the borrower must both accept, and if they have not accepted it within 60 days the commitment becomes void and the Agency de-obligates the funds ^[42]. A lender that later decides it no longer wants the guarantee must tell the Agency in writing, and the Agency then de-obligates the funds tied to the commitment ^[43]. The commitment itself stays in force for a year, or for as long as the project needs to reach loan closing ^[44], and if it expires the Agency may de-obligate the funds ^[45]. Funds are tied to a commitment from the day it is issued, and a share of them comes back months, or more than a year, later, when a project stalls, a lender withdraws or a commitment lapses.

USAspending records that return in a way that rewrites the past. The dataset's grain note finds that almost every B&I cancellation action carries an action date in the same fiscal year as the award it cancels, most of them on the very same day, even though many of those cancellations were first reported more than three months after the award. USAspending, in other words, dates a cancellation back to the original obligation. As a result, a fiscal year's net face value keeps shrinking for years after the year has closed, and a plain sum read today is a different number from the same sum read a year ago.

USDA's budget tables work the other way. They record obligations as booked when the year ended and do not reopen them. The two routes measure the same loans at different moments, and Table 6 sets them side by side for the years in which the site's transcription carries a B&I program level.

On the plain sum, the gap to the secondhand budget level runs from -14.70 percent ^[22] in FY2024 to 8.90 percent ^[22] in FY2023. Its sign is not even stable: FY2023 lands above the budget because the food supply chain loans, which the budget's B&I line leaves out, sit inside listing 10.768. On the December 31 reading of the core stream, the gaps for the four finished years narrow to between -0.60 percent ^[22] and -2.80 percent ^[22], and every one of them falls on the short side.

The difference can be taken apart exactly. Table 7 starts from the plain net sum and names the three things inside it that the budget's year-end figure does not hold: the other program streams, the cancellations first reported after December 31 and dated back into the year, and the new awards first reported after that date. Taking those three out of the plain sum leaves the December 31 reading of the core stream, and what separates that reading from the budget level is a residual the record does not explain.

Two columns of Table 7 carry most of the weight. The later cancellations are large in every year; in FY2024 they come to -$335.2 million ^[23], and they enter the plain sum with a negative sign precisely because USAspending has dated them back into the year. The food supply chain loans matter only in fiscal years 2022 and 2023, but in those years they are large enough to turn an undercount into an overcount. The later new awards pull the other way and are smaller.

Neither side is wrong. The budget figure answers what USDA obligated in the year as its books stood when the year ended. The plain USAspending figure answers what those same obligations are worth now that later cancellations have been taken back. A study that compares a state's B&I activity with a national total has to take both numbers from the same side, or it will mistake a timing gap for a geographic one.

The December 31 Reconstruction: A Fit, Not a Forecast

USAspending keeps, for each transaction, the date it was first reported. That makes it possible to rebuild a fiscal year as the record looked on any later day, by keeping only the transactions first reported by then. The Loan Analytics measures rebuild each year as it stood on December 31 after it closed, three months past the end of the federal fiscal year ^[12], and they do it for the core stream only, so that the comparison with the budget's B&I line is like for like.

The second illustration shows the mechanics with made-up numbers, all of them assumptions printed in Table 3. One fiscal year's new awards are read three times: at December 31 after the year, a year later, and today. Each later reading absorbs more back-dated cancellations, partly offset by new awards that were reported late.

In the illustration the same year loses a little over a tenth of its December 31 value by the time it is read today, although nothing about the loans the year names has changed in the meantime except what has been reported about them.

Three cautions come with the method. First, the December 31 cut was chosen after the dataset tried six candidate report dates and kept the one that fitted the budget best, so its closeness is fitted after the fact; it is not evidence that USDA closes its books on that day, and it should not be sold as a prediction. Second, the method passes a simple control: measured against the wrong fiscal year's budget figure, the same reconstruction misses by a far wider margin than against the right one, which shows that the fit follows the year rather than the general size of the program. Third, a residual remains in every year, in the last column of Table 7, that the record does not explain, and the reconstruction should be printed with that residual stated, never as a match.

State comparisons inherit all of this and add a problem of their own. In some state-years a large cancellation was first reported before December 31, so the reconstruction removes it, while the budget table, frozen at the year end, kept almost the whole commitment. In those cases even the best reconstruction misses by a wide margin, and the honest course is to print both numbers and the reason for the gap rather than adjust one to fit the other.

Completeness: What the Download Proves, and What It Does Not

A download can arrive whole and still be incomplete, and the two conditions need different tests. The first test asks whether the file the publisher built is the file on disk. For the main pull, a transactions download of the kind the endpoint list describes ^[8], covering the three Rural Development listings from October 2014 to September 2026, the publisher reported 52,299 ^[24] rows. A fresh count of the bytes on disk found 52,223 ^[24] prime transactions and 76 ^[24] subaward rows, which together equal the publisher's figure, and the hash recorded when the file was fetched equals the hash of the file on disk. That proves the zip came through intact. It says nothing about whether every award the government made is inside it.

The second test needs a witness that did not produce the file. USAspending's search interface has a separate endpoint that returns the number of awards in each award type ^[46]. Asked for B&I loans over FY2015 to FY2026, it counts 4,037 ^[24] awards; the download, deduplicated, holds 4,002 ^[24] distinct awards over the same span. The shortfall is small, and the dataset records plainly that its cause was not established. The search endpoint assigns awards to dates by its own award-level logic, which differs from the transaction action date, but that difference has not been shown to account for the whole gap. The same probe finds shortfalls for the other two Rural Development programs as well.

The honest reading puts these facts together. The download is intact; a small number of awards may be missing from it; and the budget reconciliation leaves a residual of a similar order. A study can rely on the B&I totals for scale and for trend. It should not present them as a census accurate to the last award, and where a county or a single year rests on a handful of awards, the chance that one of them is missing is material and worth a sentence.

What a Study Can Say About B&I in a State or County, and What It Cannot

Taken together, the record supports a limited but useful set of statements. A study can say how many new B&I awards were dated in a state in each fiscal year from FY2017, what their face value was when they were made, and how much of it survives today after cancellations. It can show whether the program has been active in a market at all, and whether that activity was steady or concentrated in one or two years. It can separate the core program from the temporary streams. It can place a county against its state and the nation on the same measure, provided the measure is named. The dataset's committed tables are by state and fiscal year; the rows beneath them carry the county, so a county figure is built from those rows under the same rules.

Those rules include protection for small cells. Wherever a cell of the state table touches fewer than 3 awards, its dollar values are withheld, and the withheld amounts are pooled so that the states still add up to the national total. A state with one or two awards in a year will therefore show activity without dollars, and a study should print it that way rather than go looking for the dollars elsewhere. The same discipline applies with more force to a county, where one award can be one identifiable business.

What a study cannot say is just as clear. It cannot give B&I's industry mix from this route, because the field does not exist ^[20]. It cannot name the active lenders or their market share. It cannot give an approval rate, a decline rate, a delinquency rate or a loss rate, because the file holds none of the events those rates count. It cannot state the guaranteed exposure, because the guaranteed share is not recorded. It cannot treat an award as a closed loan, because the rule ties obligated funds to a conditional commitment that can still be withdrawn or lapse ^[43]. And it cannot treat the subsidy cost as a measure of lending, or of losses that actually happened, because it is a forecast fixed by the year's assumptions ^[27].

For a borrower's feasibility study, then, the B&I record is context, not evidence of demand. It can show a lender that the program reaches the county, that projects have been committed there, and at what scale wherever enough awards stand behind the figure. It cannot show that those projects succeeded, that the market has room for another, or that the proposed project will earn enough to carry its debt. Those questions are answered from the project's own market, cost and cash flow analysis, and the program tests that analysis has to meet are walked through in the B&I program method article. Worked examples, such as the North Dakota RV park study, show those tests applied to a single project, and state figures for both SBA and USDA lending sit on the state data pages.

A Checklist for Reading a B&I Figure

Before a B&I figure goes into a study, or a lender relies on one, eight questions settle most of what can go wrong.

• Which field? Face value, subsidy cost or obligation. Only face value measures lending; the obligation on a loan row is zero, and the subsidy cost is a budget estimate of long-term cost ^[16].
• Which measure of face value? Gross new awards, net face value as read today, or the year as reported by December 31 after it closed. Each answers a different question, and one comparison must use one measure throughout.
• Which streams? Core B&I alone, or all of listing 10.768 with the CARES Act and food supply chain loans included. A comparison with USDA's budget line needs the core stream.
• Which years, and are they finished? FY2008 is the earliest year the download gives, the state and county fields are complete from FY2017, and FY2026 is a partial year.
• Read when? A net figure for a past year falls as later cancellations are dated back into it, so the date the data were pulled belongs beside the figure.
• Which place? The place of performance, not the recipient's address, and the county at the finest, never the ZIP code or the town ^[19].
• Whose budget figure? In this dataset, any USDA budget total is the site's secondhand transcription, not a primary copy, and it should be labelled that way wherever it is printed.
• What is missing? No industry, no lender, no decisions and no outcomes. A figure that seems to report one of these did not come from this record ^[36].

Each item on that list traces to a field, a rule or a check described above, and each can be repeated by anyone. The interface requires no authorization ^[4], the download is a plain CSV file inside a zip, and the dataset's manifests name the row and column behind every B&I figure printed here. A reader who doubts a number can rerun it, and a study that cites one should give the reader enough to do exactly that.

Related Pages on Loan Analytics

• The Loan Analytics methodology series (all articles)
• How a feasibility study is done: the Loan Analytics methodology
• USDA feasibility studies
• Loan data by state
• Loan data by asset class
• Worked examples (case studies)
• Worked Example: USDA Business and Industry Loan Glamping Resort Feasibility Study in Ohio
• Worked Example: USDA Business and Industry Hotel Feasibility Study in Connecticut
• Worked Example: USDA Business and Industry Loan RV Park Feasibility Study in North Dakota
• USDA Business & Industry Loans by State: A Ranked Index, and Why the Obvious Number Is the Wrong One
• USDA B&I Guarantees Under 7 CFR 5001: What the Agency Tests and How a Study Answers
• What the SBA Loan File Cannot Tell You: The Limits of FOIA Data in Underwriting

Commission a Feasibility Study

Loan Analytics prepares lender-grade feasibility studies for SBA 7(a), SBA 504, USDA Business and Industry, conventional and EB-5 financing, built on the methods this series sets out. A study is not an appraisal: it tests whether a project's market, costs and cash flow can carry the debt a lender is asked to approve. To discuss a study for your project, contact Loan Analytics.

Daniel Smith, MAI, Loan Analytics