Analytics for CDCs: Data in the 504 Workflow

Type "cdc sba 504 analytics" into a search box and the results describe a program, not a practice. SBA's own pages explain what a 504 loan is and who a certified development company is. A congressional research summary explains the policy history. A login screen offers quarterly performance data to anyone who already has credentials. None of it answers the question a credit officer at a CDC is actually asking, which is narrower and much less discussed: across the life of a 504 project, from the first call to the last annual review, where does data enter the file, what kind of data is it, and what does a supporting analytics stack have to cover to keep that file defensible.

The answer is unusual, and it is why 504 deserves its own treatment rather than a footnote in a general lending-analytics piece. A 504 file carries an evidentiary burden that no conventional commercial mortgage carries. It has to prove a fixed-asset purpose, an occupancy percentage, an economic development outcome measured in job opportunities, a property-character judgment that moves the borrower's cash contribution by ten points, a collateral value on a trigger written against property value rather than loan size, and a repayment analysis that survives a second reader at a processing center the CDC never meets. Then, after funding, the same organization that originated the loan services it, reports on it quarterly when it goes past due, and is graded on the aggregate by a federal risk rating it does not control.

Every one of those obligations is a data obligation. Some of them are satisfiable from public records at a quality no purchased dataset improves on. Some of them are not satisfiable from public records at all, and the honest version of a buying guide says so out loud.

Here is the finding that reorders the rest. A CDC's market-analytics problem is the inverse of a national lender's, and the inversion is written into the regulation rather than into anyone's business strategy. Because 13 CFR 120.802 fixes a CDC's Area of Operations with the state of incorporation as its floor, and because expanding beyond it is conditioned and slow, the entire 504 program was delivered across fiscal 2021 through fiscal 2025 by 180 distinct CDC names against 2,040 distinct 7(a) lender names, and over seventeen fiscal years the 504 program has never touched 734 of the 3,144 counties in the 50 states and the District of Columbia while 7(a) has missed only 49 (MMCG tabulation from the SBA 7(a) and 504 FOIA loan-level files as of June 30, 2026). A national lender buys breadth because its footprint is unbounded. A CDC cannot use breadth, and in exchange it gets something a bank almost never has inside its own territory: with one organization writing 97.2 percent of Nebraska's 504 approvals and 90.6 percent of Maine's, a CDC's own portfolio is close to a census of the program in its market rather than a sample of it. The analytics job is not to buy coverage. It is to join public layers onto a book of business that is already the population.

What follows walks the 504 workflow stage by stage as a sequence of data problems, names the provision that creates each one, shows what the public record can and cannot answer, and ends with a stack specification written as requirements rather than as a shopping list. It is the CDC chapter of a broader argument about choosing a commercial real estate analytics stack, and it sits beside the pieces on what market data the SBA loan file actually asks for, on the structure and caveats of the SBA FOIA files, and on analytics for small-balance lenders, whose problem set overlaps the CDC's without matching it.

The bounded map, and why it inverts the buying decision

Start with geography, because it governs everything downstream. 13 CFR 120.802 defines the Area of Operations as "the geographic area where SBA has approved a CDC's request to provide 504 program services to small businesses on a permanent basis", and adds that "the minimum Area of Operations is the State in which the CDC is incorporated" (Electronic Code of Federal Regulations, 13 CFR 120.802, text as of August 1, 2026). Expansion is not a business decision taken alone. Under 13 CFR 120.835 a CDC must meet all requirements to be an Accredited Lender Program CDC and show it can competently fulfil its responsibilities in the proposed area; a multi-state expansion must be into a state contiguous to the state of incorporation, and the CDC must either seat a loan committee in that state whose members live or work there or bring at least two directors who live or work there into the vote on projects located there. Lending outside the Area of Operations at all runs through 13 CFR 120.839, which allows the processing center to approve a case only where the CDC has previously assisted the business or its affiliates, or the CDCs already serving the area consent, or there is no CDC within the Area of Operations, and which directs SBA to weigh the CDC's Risk Rating and review history in the decision.

No comparable geographic gate applies to a 7(a) lender. The consequence shows up in the loan-level record with unusual clarity, and it is the reason the two programs draw different maps of the same country. Across fiscal 2010 to fiscal 2026 the 504 program reached 2,410 of 3,144 counties, or 76.7 percent, against 3,095 for 7(a); 689 counties carry 7(a) activity and no 504 activity, four carry the reverse, and 522 of the 734 counties with no 504 loan in seventeen years fall in the two nonmetro classes with an urban population under 5,000 (MMCG tabulation from the SBA FOIA files as of June 30, 2026, joined to USDA Economic Research Service Rural-Urban Continuum Codes, 2023 vintage). The map is drawn by regulation before it is drawn by demand, an argument developed at length in the piece on SBA lending mapped from the public tape.

For a buying decision this changes the arithmetic completely. National coverage is the headline feature of most analytics products and the main thing their pricing tiers are organized around. To a CDC operating in one state and two contiguous counties of another, national coverage is a rounding error in value. What matters instead is resolution: parcel-level records in the counties actually served, current assessor attributes, local traffic counts, flood and wetlands layers at the site rather than at the county centroid, and demographic estimates at block group rather than at metro. A CDC that pays for reach and receives coarse local data has bought the wrong axis.

The second half of the inversion is the one CDCs consistently undersell. In a state where one CDC writes nine of every ten 504 loans, that CDC's servicing file is not a sample of the 504 market, it is very nearly the market. Its own history of project types, project costs, third party lender partners, appraised values, occupancy structures and job outcomes is a proprietary dataset that no vendor can assemble, because the underlying records are not published at that grain. The analytics question is therefore about joining, not about buying: can the stack take an internal portfolio table and put it on the same map as public parcels, flood zones, traffic counts and demographics, with the vintage of each layer carried on the displayed value. That is a systems requirement, and it is the requirement most often left out of an evaluation checklist.

Stage one: eligibility screening is a records problem

The first data pass on a 504 inquiry answers questions that have nothing to do with credit. 13 CFR 120.802 defines a Project as "the purchase or lease, and/or improvement or renovation of long-term fixed assets by a small business, with 504 financing, for use in its business operations", and Project Property as "one or more long-term fixed assets, such as land, buildings, machinery, and equipment". Working capital, inventory and goodwill are outside the program by definition, so the first screen is a use-of-proceeds screen against a fixed-asset test.

The second screen is occupancy, and it is quantitative. Under 13 CFR 120.131 a borrower acquiring an existing building may permanently lease up to 49 percent of the rentable property provided it permanently occupies and uses no less than 51 percent. For new construction the borrower may permanently lease up to 20 percent while occupying no less than 60 percent, with the balance planned for occupancy on the timetable the rule sets. Both tests are computed on rentable area, which means the very first analytical artifact in a 504 file is a measured area figure and a lease schedule that reconciles to it. Assessor records give a building area and a year built; a parcel layer gives lot size and coverage; neither is a substitute for a rent roll, but both are the check that catches a borrower's stated square footage before it becomes a certification. The techniques are the ordinary ones described in the work on parcel-derived land metrics, applied to a compliance test rather than to a valuation.

The third screen is cost eligibility. 13 CFR 120.882 makes eligible those costs directly attributable to the Project, a contingency reserve for cost overruns not exceeding 10 percent of construction cost, professional fees directly attributable and essential to the Project including title insurance, architectural and engineering costs, appraisals and environmental studies, and repayment of interim financing including points, fees and interest, with separate paragraphs governing expansion refinancing and the refinancing of qualified debt. Two consequences follow for the analyst. Diligence spend is inside the project rather than outside it, which changes the economics of ordering evidence early. And the 10 percent contingency ceiling makes the construction budget a number that has to be defensible on the day it is set, which is why public cost indices and permit records earn their place in the screening file rather than only in the appraisal.

Stage two: the economic development test almost nobody automates

Here is the obligation with no analogue anywhere in conventional commercial real estate lending, and the one most likely to be handled in a spreadsheet that lives on one person's desktop. A 504 project has to justify itself in economic development terms, and the justification is numeric.

13 CFR 120.861 states that a project "must create or retain one Job Opportunity per an amount of 504 loan funding that will be specified by SBA from time to time in a Federal Register notice". The operative notice sets that amount: a project must create or retain one job opportunity per $75,000 guaranteed by SBA, and for a project of a small manufacturer, one job opportunity per $120,000 guaranteed by SBA (U.S. Small Business Administration, notice on Development Company Loan Program job creation and retention requirements, 83 FR 55224, November 2, 2018). The same notice sets the portfolio test that sits behind 13 CFR 120.829: a CDC's portfolio must average one job opportunity for every $75,000 guaranteed by SBA, rising to an allowance of no more than $85,000 per job created or retained for projects in Alaska, Hawaii, designated zones and other areas SBA names, including opportunity zones. A newly certified CDC has two years from its certification date to reach the average, and 13 CFR 120.829(b) requires the CDC to indicate in its annual reports the job opportunities actually or estimated to be provided by each project.

Where a project cannot meet the job test on its own, 13 CFR 120.862 supplies the alternative: community development goals covering improving, diversifying or stabilizing the economy of the locality, stimulating other business development, bringing new income into the community, assisting manufacturing firms in NAICS sectors 31 to 33 and assisting businesses in labor surplus areas; and public policy goals covering revitalization of a business district with a written revitalization or redevelopment plan, expansion of exports, expansion of small businesses owned and controlled by women, veterans and minorities, aiding rural development, increasing productivity and competitiveness, modernization to meet health, safety and environmental requirements, reduction of unemployment rates in labor surplus areas, reduction of energy consumption by at least 10 percent, increased use of sustainable design, and renewable energy or renewable fuel projects.

Read that list as a data specification and it resolves into named public series. A labor surplus area is a defined federal classification, not an adjective: under 20 CFR part 654, subpart A, the Assistant Secretary classifies a civil jurisdiction as a labor surplus area when its average unemployment rate for the reference period reaches 120 percent of the national average or 10 percent or higher, no jurisdiction qualifies where the rate is below 6.0 percent, the reference period is the two-year period ending December 31 of the year before the October 1 eligibility date, and the list is published annually (Electronic Code of Federal Regulations, 20 CFR part 654, subpart A, text as of August 1, 2026). Rural development has an equally concrete definition in USDA Economic Research Service county classifications. Manufacturing is a NAICS sector test. Unemployment reduction runs on Bureau of Labor Statistics series. Local economic stabilization is an establishment-count and wage argument that the Quarterly Census of Employment and Wages and County Business Patterns answer directly, using the methods set out in the work on QCEW and CES as demand drivers and on counting competitors with County Business Patterns.

None of that is exotic data. All of it is time-stamped, jurisdiction-specific and revised on a schedule, which is exactly the class of input that decays silently when it is pasted into a memorandum once and never refreshed. A labor surplus area designation that was true in the prior cycle may not be true in the current one, and the eligibility date is October 1. A stack that cannot tell a credit officer which vintage of which list a determination rested on is not supporting the file; it is decorating it.

Stage three: four percentages that decide the deal

The structuring stage is where 504 becomes arithmetic. 13 CFR 120.801(c) describes permanent 504 financing as a contribution by the small business of at least 10 percent of project costs, a loan funded by a CDC debenture for up to 40 percent of project costs plus certain administrative costs secured by a second lien on the project property, and a third party loan for the balance secured by a first lien; paragraph (d) states the debenture is guaranteed 100 percent by SBA with the full faith and credit of the United States.

13 CFR 120.910(a) then sets the borrower's minimum contribution at four levels: at least 15 percent of project cost if the borrower or operating company has operated for two years or less; at least 15 percent if the project involves a limited or single purpose building or structure; at least 20 percent if both conditions apply; and at least 10 percent in all other circumstances. 13 CFR 120.920(a) moves the third party loan in parallel, requiring one or more third party loans totalling at least as much as the 504 loan and rising to at least 50 percent of total project cost when either condition is present. 13 CFR 120.930 caps the 504 loan at 40 percent of total project cost plus 100 percent of eligible administrative costs, which SBA may raise to as much as 50 percent for good cause, holds federal sources to no more than 50 percent of eligible project costs, and sets a floor of $25,000 on a 504 loan. 13 CFR 120.931 sets the ceilings: $5,000,000 for each borrower and its affiliates, and $5,500,000 for each project where the borrower is a small manufacturer in NAICS 31 to 33 with all production facilities in the United States, where the project reduces the borrower's energy consumption by at least 10 percent, or where the project involves renewable energy or renewable fuel production.

SOP 50 10 8 tabulates the same structure as three typical cases: third party lender 50 percent, CDC and SBA 40 percent, borrower 10 percent for standard financing; 50, 35 and 15 where the borrower is a new business or the property is limited or special purpose; and 50, 30 and 20 where both apply (U.S. Small Business Administration, SOP 50 10 8, Section C, Chapter 1, Third Party Lender Participation, effective June 1, 2025). The same paragraph requires the third party loan to be at least the net debenture proceeds, with a term of at least 7 years for a 10-year debenture and at least 10 years for a 20-year or 25-year debenture.

Everything in that paragraph is deterministic except one input: whether the property is limited or special purpose. That single judgment is worth five points of borrower equity and five points of debenture, and it is the CDC's to make and to document.

The same project, three structures, in dollars

Percentages hide the size of the decision, so it is worth putting a number on it. On a $1,000,000 project cost, the standard structure asks the borrower for $100,000, the debenture for $400,000 and the third party lender for $500,000. Where either the new-business condition or the limited or special purpose condition applies, the borrower puts in $150,000 and the debenture falls to $350,000. Where both apply, the borrower puts in $200,000 and the debenture falls to $300,000. The third party lender's $500,000 does not move. The arithmetic is the regulation's percentages applied to a round project cost, not a quoted term sheet, and the point of stating it is that a property-character conclusion reached in a paragraph of the credit memorandum has doubled the borrower's cash requirement.

The caps behave the same way at the top of the range. Because 13 CFR 120.930 sets the debenture at 40 percent of project cost and 13 CFR 120.931 caps it at $5,000,000, the percentage stops binding above a project cost of $12,500,000, after which the dollar ceiling governs and the borrower and the third party lender absorb the remainder. At the 35 percent structure the crossover is $14,285,714 and at 30 percent it is $16,666,667. On the $5,500,000 ceiling available to small manufacturers and qualifying energy projects the same crossovers move to $13,750,000, $15,714,286 and $18,333,333. Those are arithmetic consequences of two published limits, not observed pricing, and they are worth precomputing because they tell an originator at intake which structures are still available at a given project size.

Limited or special purpose: a classification with a price in equity

13 CFR 120.910 and 13 CFR 120.920 use the phrases "limited or single purpose building or structure" and "limited or single purpose asset" without defining them, and neither 13 CFR 120.10 nor 13 CFR 120.802 supplies a definition. SOP 50 10 8 fills the gap. Appendix 3 defines a limited or special purpose property for 504 as a limited-market property whose unique physical design, special construction materials or layout restricts its utility to the use it was built for, and Section C, Chapter 1, Paragraph E.1.c carries a 25-item example list that the SOP itself states is not all-inclusive, running from amusement parks, bowling alleys and car wash businesses through cold storage facilities with more than half their square footage refrigerated, gas stations, golf courses, hotels and motels, marinas, nursing homes including assisted living, theaters and wineries. The same paragraph requires the CDC to state in its credit memorandum whether the project property is limited or special purpose and to explain its conclusion.

Because the list is explicitly not exhaustive and property type alone does not settle the question, this is an evidenced judgment rather than a lookup, and it is one of the few places in the file where a market argument changes a regulatory outcome directly. The evidence that supports it is market evidence: how many comparable buildings in the trade area, how many alternative users, what conversion has actually happened locally, what the improvements cost relative to the shell. The related question of how that risk shows up in performance is treated separately in the work on what the loan data shows about special-purpose property risk.

The public tape lets a CDC test the intuition behind the rule against outcomes, and the answer is more nuanced than the regulation implies. Across 504 approvals from fiscal 2010 through fiscal 2019, a window seasoned enough to compare, 58,887 disbursed loans produced 887 charge-offs, a count rate of 1.51 percent on an average approval of $698,498 (MMCG tabulation from the SBA 504 FOIA loan-level file as of June 30, 2026). Within that window, several property types the SOP names by example sit well below the program average: car washes at 0.89 percent across 787 loans and golf courses and country clubs at 0.98 percent across 102. Others sit well above it: hotels and motels at 2.55 percent across 2,743 loans, gasoline stations with convenience stores at 2.56 percent across 821, and nursing care facilities at 2.07 percent across 193. Meanwhile several property types nobody would call special purpose sit high as well, with supermarkets and other grocery retailers at 4.20 percent across 429 loans. The equity tier is a regulatory instrument, not a risk model, and a CDC that treats it as a risk ranking will misprice its own portfolio commentary. No rate here is published for any cohort below ten loans, consistent with the suppression standard applied throughout this library and described in the piece on small-balance loan performance by property type.

Collateral, and an appraisal trigger written on value rather than loan size

The 504 appraisal rule is frequently misremembered, including by people who work with both SBA programs, so it is worth stating precisely. SOP 50 10 8 Section C, Chapter 1 requires a real estate appraisal when the estimated value of the 504 project property is greater than $500,000, and in other listed circumstances including related-party transactions, a seller carry-back forming part of the borrower's contribution, and changes of ownership. The appraiser must be independent and state-licensed or state-certified, state-certified where the estimated value exceeds $1,000,000; the report must comply with the Uniform Standards of Professional Appraisal Practice, be dated no more than twelve months before application, and identify SBA as client or intended user, with the third party lender permitted to be the client. New construction or substantial renovation must be appraised at market value on completion. A going concern appraisal must allocate separate values to land, building, equipment and business including intangibles. Where the collateral is a special purpose property, the appraiser must be experienced in the particular industry. And if the appraisal comes in below 90 percent of the estimated value, the debenture must be reduced or additional collateral or borrower investment added.

Two features of that rule matter for analytics. First, the trigger reads on estimated project property value, not on loan amount, which is a different test in kind from the thresholds a bank applies on its own paper and different again from the 7(a) requirements, which vary by product chapter rather than running on a single program-wide dollar line. The practical effect is that the estimate of value has to exist, and be defensible, before the appraisal is ordered. Somebody at the CDC is making a value estimate early, from comparable evidence, and that estimate determines whether an appraisal is required at all and whether it needs a state-certified appraiser.

Second, the 90 percent rule turns a valuation shortfall into an immediate structural change. A CDC that can see, before ordering, that local sale evidence is thin or that the improvements are heavily specialized has time to reset the borrower's expectations rather than restructure at the last moment. That early value work is not appraisal, and it should never be presented as appraisal; the distinction between the three documents is set out in the piece on how a market study differs from an appraisal. It is screening, and it runs on assessor sale records, parcel attributes and public transfer data.

The credit memorandum is where the market argument gets written down

SOP 50 10 8 Section C, Chapter 1, Paragraph E.1 sets out what the CDC credit memorandum must address, and read as a data specification it is unusually explicit. The memorandum must carry a pro-forma balance sheet adjusted for the third party loan, the 504 loan, other new debt, any required equity injection and loan costs, with a complete debt schedule. It must carry a repayment ability analysis using two or three years of statements and federal tax returns depending on the size standard applied, plus statements dated within 120 days, showing a debt service coverage ratio of operating cash flow to debt service equal to or greater than 1:1 on calculations acceptable to the processing center. For projection-based projects it must carry at least two years of projections with assumptions justified by comparison to current industry trends. It must carry a ratio analysis against industry averages, including the current ratio and debt to tangible net worth among others. It must state whether the project property is limited or special purpose and why. Rental income from the project property may enter the global cash flow analysis but not the repayment ability analysis.

Three of those requirements cannot be met from inside the borrower's own records. "Comparison to current industry trends" needs an external series with a vintage. "Ratio analysis against industry averages" needs a comparator population. A defensible statement about the level of competition in the market area needs an establishment count with a geography attached. The same SOP paragraph lists reports prepared independently of the small business that may help mitigate identified weaknesses, and notes that SBA has regulatory authority under 13 CFR 120.160(b) to request an independent study where it is needed to understand the business type and market conditions at the project location, a request that sits with the processing center rather than with the CDC.

The public series that answer these are unglamorous and completely citable: County Business Patterns for establishment counts by industry and county, the Quarterly Census of Employment and Wages for establishment, employment and wage levels by NAICS, the American Community Survey for trade-area income and household composition using the ring and block-group methods described in the work on ACS trade-area demographics, and state department of transportation counts for any project whose demand argument depends on passing traffic. What separates a memorandum that survives a second reader from one that does not is rarely the sophistication of the analysis. It is whether every figure carries a publisher, a series name, a vintage and a geography, the discipline argued for in the piece on the provenance standard.

Environmental screening starts as a NAICS lookup

The environmental requirement is a data step before it is a consulting engagement, and its first move is a classification decision. SOP 50 10 8 Section A, Chapter 5, Paragraph E requires an environmental investigation of all commercial property offered as security for a loan or debenture, beginning with a good-faith determination of the NAICS codes for the property's current and known prior uses compared against the SOP's Appendix 6 list of environmentally sensitive industries. A match means the investigation must begin with a Phase I environmental site assessment regardless of loan amount. With no match, a loan up to and including $250,000 may begin with an environmental questionnaire, and a loan over $250,000 must begin at minimum with a questionnaire plus a records search with risk assessment. Any result other than low risk escalates. All transaction screens and Phase I and Phase II assessments must be performed by an environmental professional and accompanied by the SOP's reliance letter, and reports must be dated within one year of issuance of the SBA loan number.

Two things follow. The prior-use question is a historical records question, which means aerial photography, city directories and fire insurance maps are part of the file rather than part of a consultant's discretion, and a CDC that can pull the current and prior use signal early avoids ordering the wrong first step. And the loan-amount threshold means the screening decision has to be made while the structure is still moving. The broader treatment of the layers that answer these questions before any report is commissioned is in the piece on environmental and hazard screens before the Phase I, and the flood side, which is a separate obligation running on the National Flood Hazard Layer, is covered in the work on reading FEMA flood zones.

Program scale, and what the channel actually moves

Context matters for a stack decision, because the volume a CDC processes determines whether an analytical step can be manual. The program is smaller and lumpier than its public profile suggests. SBA approved 5,874 504 loans for $4,753,644,000 in fiscal 2018, rose to 9,676 loans for $8,218,105,540 in fiscal 2021 and 9,254 loans for $9,207,996,290 in fiscal 2022, fell to 5,924 loans for $6,419,378,000 in fiscal 2023, and reached 6,762 loans for $7,804,172,000 in fiscal 2025 (U.S. Small Business Administration, 7(a) and 504 Monthly and Yearly Activity Report, data as of September 30, 2025). Over those eight fiscal years the count rose 15.1 percent while approved dollars rose 64.2 percent, because the average approval moved from $809,269 to $1,154,122, a rise of 42.6 percent, all three figures derived by arithmetic from the published rows rather than published as such.

Spread across the 174 certified development companies with at least one approval in fiscal 2025 (U.S. Small Business Administration, 504 CDC Activity Report, data as of September 30, 2025), that is an average of 38.9 approvals per CDC for the year, arithmetic on the two published figures. The distribution is heavily skewed toward the largest originators, so most CDCs sit below that average rather than at it. That is the number that should govern the build-or-buy conversation, and it argues against heavy internal engineering: at that volume the marginal value of a bespoke data pipeline is low and the marginal value of a shared, provenance-carrying layer set is high. The same conclusion, reached from the bank side and at a different loan size, appears in the work on what matters for lenders under $10 million.

After funding, the file is read again, on a schedule

Origination analytics get the attention; servicing analytics carry the regulatory exposure. Under 13 CFR 120.970 the CDC is responsible for routine servicing, including receipt and review of the borrower's financial statements annually or more frequently, assuring the borrower makes all required insurance premium payments, verifying tax payments, filing security interest renewals and extensions, responding to modification requests and cooperating with SBA on cures and workouts. The same section provides that for any 504 loan more than three months past due the CDC must promptly request that SBA purchase the debenture, absent an approved deferment or catch-up plan.

The reporting obligations are separate and additive. 13 CFR 120.830 requires a CDC to submit an annual report within 180 days after the end of its fiscal year, containing audited or reviewed financial statements per 13 CFR 120.826, detailed compensation information for officers, directors, employees and contractors above a stated threshold, a written annual certification by each board member, and a written report on investments in economic development in each state where the CDC has an outstanding 504 loan. It also requires quarterly service reports on each loan in the portfolio that is 60 days or more past due. 13 CFR 120.826 sets the audit line at a 504 loan portfolio balance of $30,000,000, above which financial statements must be audited annually by an independent certified public accountant, and requires each CDC board to adopt an internal control policy covering credit review, collateral management and a loan review program with standardized classification.

Sitting above all of that, 13 CFR 120.1015 provides that SBA may assign a risk rating to all SBA lenders and intermediaries on a periodic basis, based on risk-related portfolio performance factors set out in notices or SBA's standard operating procedures, and classified as acceptable or less than acceptable. SBA's own lender guidance states that CDCs can use the lender portal to view their own quarterly performance data, including their most current composite risk rating and the lender risk rating, and can also access data on peer group and portfolio averages (U.S. Small Business Administration, SBA lenders resource page, read August 2026). A CDC therefore receives a graded external view of its portfolio four times a year and has no ability to reconstruct how the grade was computed.

That asymmetry is the strongest argument for an independent benchmark, and the public tape provides one. Charge-off rates on 504 approvals fell steadily across the seasoned window, from 3.36 percent for fiscal 2010 approvals to 2.06 percent for fiscal 2011, 1.42 percent for fiscal 2013, 1.04 percent for fiscal 2016, 0.44 percent for fiscal 2018 and 0.28 percent for fiscal 2019 (MMCG tabulation from the SBA 504 FOIA loan-level file as of June 30, 2026). The trend is real, but so is the caveat that governs how far it can be read: the open share of each cohort climbs from 21.6 percent for fiscal 2010 to 77.1 percent for fiscal 2019, so the later years in the series are measuring a partially resolved population and will drift upward as those cohorts age. Any CDC that benchmarks itself against a recent approval year without stating the open share is comparing an incomplete outcome to a complete one. The practice of reading these files correctly, including the status codes and the quarterly update cadence, is set out in the piece on the SBA FOIA loan datasets, and the wider surveillance routine in the work on annual reviews and portfolio surveillance with public data layers.

What the public record reaches, and what it does not

A category piece owes the reader both halves of the ledger. On the reachable side, most of what a 504 file needs is federal, free and citable. The 7(a) and 504 FOIA loan-level files carry every approval since fiscal 1991 with project state and project county, updated quarterly, which supports program benchmarking, industry comparison and geographic context. Census County Business Patterns and the American Community Survey supply establishment counts and trade-area demographics. Bureau of Labor Statistics QCEW and CES supply employment and wage series for the job and industry arguments. USDA county classifications supply the rural test. FEMA's National Flood Hazard Layer, the National Wetlands Inventory, USGS terrain data and state department of transportation counts supply the site layers. Assessor and recorder records supply parcel geometry, building attributes and sale evidence, with the coverage caveats that vary state by state. Those layers are exactly the ones an analytics platform built on federal, state and public data carries with source and vintage provenance on displayed values, and the SBA layer specifically is derived from the same public 7(a) and 504 FOIA datasets, applying a minimum-cohort suppression standard that shows no performance rate for a cohort under 10 loans.

On the unreachable side, four things a CDC would very much like are not in the public record and cannot be honestly manufactured from it. There is no live pipeline: the FOIA files publish approvals after the fact on a quarterly cadence, so nothing in them tells a CDC what a competing lender is quoting this week. There are no competitor terms: rate, amortization, covenant and fee structure on the third party loan are not published anywhere, and any figure presented as a market rate for a first-lien 504 companion loan is an estimate wearing a citation. There is no property-level occupancy or rent series: the tape carries no rent roll, and public sources do not produce one. And there is no forward job outcome: job opportunities are reported by the CDC to SBA, not published back at project level in a form that supports a predictive model. A stack that implies otherwise is selling a number that the underlying record does not contain. The general version of this audit, applied across the categories a lending team buys from, is the subject of the pillar on choosing a CRE analytics stack.

A stack specification for a CDC, written as requirements

Category guidance is more useful as a specification than as a shopping list, so here is what the 504 workflow implies, requirement by requirement, with the stage that generates each one.

It must resolve at parcel grain inside a bounded territory rather than at coarse grain nationally, because 13 CFR 120.802 bounds the territory and the occupancy and collateral tests are computed on specific buildings. It must carry vintage and publisher on every displayed value, because the labor surplus area list changes annually on an October 1 cycle, the ACS refreshes on its own schedule, the FOIA files update quarterly, and a determination is only as good as the vintage it rested on. It must join an internal portfolio table to public layers on the same map, because in a concentrated state the CDC's own book is the closest thing to a census of the program that exists. It must support cohort-level benchmarking with a minimum-cohort rule, because a comparison drawn from four loans is not a benchmark. It must reproduce a number on demand, because the second reader is a processing center, a purchase reviewer or an examiner who was not in the room, and reproducibility is the only property of an analysis that survives that reading. And it must be honest about absence: where the public record has no answer, the correct behavior is a blank cell and a note, not an interpolation.

Two non-requirements are worth naming as well, because they absorb budget. National coverage is close to worthless to a single-state CDC, and paying for it is paying for the wrong axis. And a predictive default model built on a CDC's own portfolio is arithmetically unavailable at typical volumes: across the full seasoned window the entire national 504 program produced 887 charge-offs from 58,887 disbursed loans, so a CDC with a few hundred loans in the same window is working with a handful of adverse outcomes at most, and any model trained on that is fitting noise. Benchmarking against the program is the defensible move; modelling one's own tail is not. MMCG Analytics is a data and analytics provider, never a lender, an appraiser or a credit decision-maker of record; the eligibility determination, the credit conclusion and the value opinion rest with the CDC, the third party lender, SBA and the licensed professionals each engages.

The larger point is that the 504 workflow rewards a different analytical posture than a bank's commercial real estate book does. A bank's analyst is asked to find opportunity in an open field. A CDC's analyst is asked to prove things about a closed one: that this borrower will occupy this percentage of this building, that this project will create this many job opportunities or satisfy this named public policy goal, that this property is or is not limited in market, that this value estimate justified this appraisal decision. Proof beats discovery in that setting, and proof is a provenance problem before it is an analytics problem. The tools that help are the ones that show their work, and a lending team evaluating them should test that property first, using the same questions any buyer should ask, as set out in the companion piece on what map-first analytics means for a lending team.

Frequently asked questions

What data does a CDC actually need for an SBA 504 loan?

Five distinct classes, generated by five different provisions. Rentable area and lease data to test the occupancy rule in 13 CFR 120.131, which requires the borrower to occupy at least 51 percent of an existing building or at least 60 percent of new construction. Employment and industry data to support the job opportunity test, which under the operative Federal Register notice requires one job opportunity per $75,000 guaranteed by SBA, or one per $120,000 for a small manufacturer. Market and competitive data to support the credit memorandum's comparison to current industry trends and its ratio analysis against industry averages. Comparable sale and property evidence to support the value estimate that decides whether an appraisal is required. And property-history and NAICS data to determine the correct first environmental step.

How much does a borrower have to put into a 504 project?

At least 10 percent of project cost in the ordinary case under 13 CFR 120.910(a), rising to at least 15 percent where the borrower or operating company has operated for two years or less, at least 15 percent where the project involves a limited or single purpose building or structure, and at least 20 percent where both conditions apply. The third party loan moves with it, reaching at least 50 percent of total project cost when either condition is present under 13 CFR 120.920(a). On a $1,000,000 project that is the difference between a $100,000 contribution and a $200,000 contribution.

When does an SBA 504 loan require a real estate appraisal?

SOP 50 10 8 Section C, Chapter 1 requires an appraisal when the estimated value of the 504 project property is greater than $500,000, and in other listed circumstances including related-party transactions, a seller carry-back forming part of the borrower's contribution, and changes of ownership. The trigger reads on estimated property value rather than on loan amount, which is why an internal value estimate has to exist first. A state-certified appraiser is required where the estimated value exceeds $1,000,000, the report must comply with USPAP and be dated within twelve months of application, and if the appraisal comes in below 90 percent of the estimated value the debenture must be reduced or additional collateral or borrower investment added. The 7(a) appraisal requirements are written per product chapter and do not run on a single program-wide dollar line.

How many jobs does an SBA 504 project have to create?

One job opportunity per $75,000 guaranteed by SBA, or one per $120,000 where the borrower is a small manufacturer, under the Federal Register notice issued November 2, 2018 that implements 13 CFR 120.861. Separately, a CDC's whole portfolio must average one job opportunity per $75,000 guaranteed, with an allowance of no more than $85,000 per job for projects in Alaska, Hawaii, opportunity zones and other designated areas, and a newly certified CDC has two years from certification to reach the average. Where a project cannot meet the test on its own, 13 CFR 120.862 allows it to qualify on community development or public policy goals instead, including assisting manufacturing firms, aiding rural development, business district revitalization under a written plan, or reducing energy consumption by at least 10 percent.

Why does the 504 program reach fewer counties than 7(a)?

Because 504 origination geography is regulated and 7(a) origination geography is not. A CDC works inside an Area of Operations approved by SBA whose minimum is the state of incorporation under 13 CFR 120.802; expanding requires Accredited Lender Program status and, for a multi-state expansion, a contiguous state plus a local loan committee or two resident directors under 13 CFR 120.835; lending outside it needs a case-by-case approval under 13 CFR 120.839. Across fiscal 2010 to fiscal 2026 the 504 program reached 2,410 of 3,144 counties against 3,095 for 7(a), and 180 distinct CDC names delivered the entire program across fiscal 2021 through fiscal 2025 against 2,040 distinct 7(a) lender names.

Can a CDC benchmark its own portfolio against public data?

Yes, and it is the only external benchmark available to it that is reproducible. The SBA 7(a) and 504 FOIA loan-level files publish every approval since fiscal 1991 with status, project state, project county, industry code and approval amount, updated quarterly. Across fiscal 2010 to fiscal 2019 approvals, 58,887 disbursed 504 loans produced 887 charge-offs, a count rate of 1.51 percent. Two disciplines make the comparison honest: state the open share of every cohort, which runs from 21.6 percent for fiscal 2010 to 77.1 percent for fiscal 2019 and makes recent years look better than they will finally settle, and publish no rate for any cohort under 10 loans.

What can public data not tell a CDC?

Four things, and no amount of processing recovers them. It cannot show a live pipeline, because the loan-level files publish approvals after the fact on a quarterly cadence. It cannot show competitor terms, because rate, amortization, covenant and fee structure on the third party loan are not published anywhere. It cannot show property-level occupancy or rent, because no public source produces a rent roll. And it cannot show forward job outcomes at project level, because job opportunities are reported by the CDC to SBA rather than published back in modellable form. Anything presented as one of those four is an estimate, and it should be labelled as one.

Sources

1. 13 CFR 120.802, Definitions, including Area of Operations, Project, Project Property, Third Party Loan, Debenture and Net Debenture Proceeds, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.802
2. 13 CFR 120.835, Application to expand an Area of Operations, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.835
3. 13 CFR 120.839, Case-by-case application to make a 504 loan outside of a CDC's Area of Operations, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.839
4. 13 CFR 120.131, Leasing part of new construction or existing building, occupancy percentages for 504 borrowers, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.131
5. 13 CFR 120.882, Eligible Project costs for 504 loans, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.882
6. 13 CFR 120.861, Job Creation or Retention, and 13 CFR 120.829, Job Opportunity average a CDC must maintain, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.861
7. U.S. Small Business Administration, Development Company Loan Program, Job Creation and Retention Requirements; Additional Areas for Higher Portfolio Average, notice, 83 FR 55224, November 2, 2018. https://www.govinfo.gov/content/pkg/FR-2018-11-02/html/2018-24033.htm
8. 13 CFR 120.862, Other economic development objectives, community development and public policy goals, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.862
9. 20 CFR part 654, subpart A, labor surplus area classification criteria and annual list, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-20/part-654/subpart-A
10. 13 CFR 120.801, How a 504 Project is financed, and 13 CFR 120.900, Sources of a Project's financing, Electronic Code of Federal Regulations, current 2026. https://www.ecfr.gov/current/title-13/section-120.801
11. 13 CFR 120.910, Borrower contributions, and 13 CFR 120.920, Required participation by the Third Party Lender, Electronic Code of Federal Regulations, current 2026. https://www.ecfr.gov/current/title-13/section-120.910
12. 13 CFR 120.930, Amount, and 13 CFR 120.931, 504 lending limits, Electronic Code of Federal Regulations, current 2026. https://www.ecfr.gov/current/title-13/section-120.931
13. 13 CFR 120.970, Servicing of 504 loans and Debentures, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.970
14. 13 CFR 120.830, Reports a CDC must submit to SBA, and 13 CFR 120.826, Basic requirements a CDC must maintain, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.830
15. 13 CFR 120.1015, Risk Rating, Electronic Code of Federal Regulations, text as of August 1, 2026. https://www.ecfr.gov/current/title-13/section-120.1015
16. 13 CFR 120.160, Loan conditions, including the authority to require additional evidence, Electronic Code of Federal Regulations, current 2026. https://www.ecfr.gov/current/title-13/section-120.160
17. U.S. Small Business Administration, SOP 50 10 8, Lender and Development Company Loan Programs with Technical Updates, effective June 1, 2025, Section A Chapter 5 Paragraph E, Section C Chapter 1 Paragraphs B and E, Appendix 3 and Appendix 6. https://www.sba.gov/document/sop-50-10-lender-development-company-loan-programs
18. U.S. Small Business Administration, 7(a) and 504 Monthly and Yearly Activity Report, fiscal 1991 to fiscal 2025, data as of September 30, 2025. https://data.sba.gov/dataset/7a-504-activity-reports-fy2025-year-end
19. U.S. Small Business Administration, 504 CDC Activity Report, fiscal 2025 year end, data as of September 30, 2025, dataset created November 13, 2025. https://data.sba.gov/dataset/7a-504-activity-reports-fy2025-year-end
20. U.S. Small Business Administration, Office of Capital Access, 7(a) and 504 FOIA loan-level datasets and data dictionary, files as of June 30, 2026. https://data.sba.gov/dataset/7a-504-foia
21. U.S. Small Business Administration, SBA lenders resource page, including lender portal quarterly performance data, composite risk rating and peer group averages, read August 2026. https://www.sba.gov/sba-lenders/
22. U.S. Small Business Administration, 504 loans program page, eligible uses and the $5.5 million maximum, 2026. https://www.sba.gov/funding-programs/loans/504-loans
23. U.S. Department of Agriculture, Economic Research Service, Rural-Urban Continuum Codes, 2023 vintage, county classes and 2020 Census population. https://www.ers.usda.gov/data-products/rural-urban-continuum-codes
24. U.S. Census Bureau, County Business Patterns, establishment counts by county and industry, 2023 data year. https://www.census.gov/programs-surveys/cbp.html
25. U.S. Bureau of Labor Statistics, Quarterly Census of Employment and Wages, establishment, employment and wage series by NAICS and county, 2026. https://www.bls.gov/cew/
26. MMCG Research, SBA 504 Workflow Series: 504 cohort resolution rates by approval fiscal year and by property-backed industry, county coverage and originator concentration by program, computed from the SBA 7(a) and 504 FOIA loan-level files as of June 30, 2026 under a 10-loan minimum-cohort rule; MMCG database, 2026. https://mmcganalytics.com/methodology/