Research Methodology
How Franchise Evidence sources, extracts, validates, scores and publishes franchise data, with every formula, threshold and limitation stated in full.
The process at a glance
- Find the FDD. Identify the primary filing and record its year, registry and source pages.
- Extract structured fields with AI. Keep each figure’s units, population and citation.
- Check against the source independently. A citation check and blind second reading test the extraction.
- Flag conflicts. Re-inspect disagreements; unresolved material figures stay out of calculations.
- Show the arithmetic. Derived figures carry formulas and their inputs.
- Keep estimates separate. Model assumptions remain labeled and editable.
Profiles have not been reviewed line by line by a person. The detailed method below explains both the checks and their limits.
Why we publish this page
Any score is an argument about what matters. Ours are arithmetic applied to disclosed figures, and the arithmetic is set out here in full so that you can check it, disagree with it, or reproduce it from the same documents. Where our method has limits — and it has several — this page states them rather than leaving them to be discovered.
Sources
Our primary source is the Franchise Disclosure Document (FDD) that a franchisor files with a state franchise regulator. Franchisors that sell franchises in registration states must file an FDD annually, and several states publish those filings. We obtain documents from public state registries, chiefly the Wisconsin Department of Financial Institutions franchise search and the Minnesota Department of Commerce CARDS system, and we record for each document the registry it came from, the registry’s public URL, the document year, the issuance date on its cover page, and the date we retrieved it.
Where an FDD is silent and a reliable government or public dataset covers the same question, we may use that dataset and label the figure accordingly. We occasionally cite a franchisor’s own public materials for descriptive, non-financial detail. We do not use franchise-portal listings, broker marketing, press releases, or aggregator sites as sources for financial figures.
We do not host or redistribute FDD PDFs. We cite them precisely and link to the public registry where they can be searched and obtained.
Which FDD we use
Each profile is built from one primary document, normally the most recent FDD available in the registries at the time of extraction. All figures on a profile come from that document unless a fact is individually cited to another source. The profile names the document, its year, and its issuance date.
What “current” means here
A profile’s document is treated as current only when it is the newest filing located in the public registries AND its FDD year matches the most recent annual filing cycle (FDDs are reissued within 120 days of the franchisor’s fiscal year end, so a document more than roughly a year behind the current cycle is stale even if nothing newer is public). A registration that has expired or been cancelled in a registry is a registration-status event — it does not make the disclosures wrong, but the profile then says, with the check date, exactly what the latest located filing is rather than calling it current. Example: Take 5 Oil Change’s Wisconsin registration expired 6/26/2026 and Minnesota issued an order of cancellation on 7/27/2026 (both checked 2026-09-03); its December 2025 amended FDD remains the latest filing we can locate, and the profile is flagged accordingly.
Older documents
Registry publication lags, and for some brands the newest available filing is a year or more old. When we knowingly build a profile from a document that is not the current year’s, the profile carries an explicit flag saying so, and the document year is shown next to the figures. Franchise economics change annually: an older filing can be materially out of date on fees, investment ranges, and unit counts. Treat a flagged profile as a starting point and verify against the current FDD from the franchisor.
How a profile is built
Acquisition
We search the state registries for the brand, identify the franchisor’s legal entity, and retrieve the filing. We record the document metadata described above, including the file number and page count where available, and note whether the document is current.
Extraction
Figures are extracted from the document into a structured record. Extraction is AI-assisted: a language model reads the relevant items and proposes values, each with a citation to the FDD Item number and the PDF page on which the value appears, together with a short verbatim excerpt where exact wording matters. Extraction is never a summary of the whole document into prose; each field is captured separately so that it can be checked against its own page.
Every extracted value is stored as a fact with four parts: the value, an evidence tier, a source citation, and an optional note explaining basis or qualification. A value that is not in the document is stored as null with the tier “not disclosed”. The extraction record also carries the method used, the date of extraction, an overall confidence level of high, medium or low, and an explicit list of fields the extractor was unsure about.
Validation
Every record is checked against the publication schema and against a set of automated rules before it can be published. Structural validation rejects unknown fields, out-of-range values, missing citations on disclosed facts, and any figure carrying a tier that its source does not support. Consistency rules check the record against itself: that Item 20 opening and closing unit counts reconcile with the openings, terminations, non-renewals, reacquisitions and other cessations reported for the year; that unit totals equal franchised plus company-owned outlets; that the low end of an investment range does not exceed the high end; that a royalty expressed as a percentage falls in a plausible band; that an Item 19 figure carries a stated population and measurement period; and that fiscal years are contiguous. Rule failures block publication until the underlying figure is re-read against the document or the discrepancy is documented as a genuine feature of the filing — franchisors do sometimes publish tables that do not reconcile, and where that is the case we say so in a note rather than silently adjusting the numbers.
Records are not reviewed line-by-line by a person. Instead, every profile goes through an independent machine-verification process that is stricter than a single read-through: two separate AI passes read the source document independently — one checking every material field against its cited page, the other re-extracting the same fields blind, without access to our record — and a script compares the two readings. Every disagreement is re-inspected against the source page (or a rendered image of it) in a third tie-break pass. Fields the passes cannot agree on are marked unresolved and excluded from derived metrics, scores and rankings rather than guessed. Each profile displays its verification status and counts, so you always know how much of what you are reading has been machine-verified against the cited source document. The passes and their limits are described under AI use.
Derived metrics
Only after validation do we calculate anything. Derived figures are computed from validated disclosed values using the published formulas below, and each carries the “derived” tier and shows its formula. If an input is missing, the derived figure is not produced; it is not approximated.
Publication
The validated record, its derived metrics and its scores are rendered into the profile, the screener, the comparison tool and the ranking pages from the same underlying data. Nothing shown on a ranking page is calculated differently from the profile it links to.
AI use
AI assists with source extraction, independent verification, explanatory drafting, and code development. It does not fill missing disclosed values from memory or another brand. Human direction sets the research scope and editorial rules; it is not a claim of human review of every profile.
Independent verification
Pass A — citation check: checks the record’s material values, units, context, fiscal year, population and page against the archived FDD.
Pass B — blind re-extraction: receives field definitions and the source document, without our record, and extracts the values independently. It also builds a census of recurring fees and re-foots Item 20 tables.
Reconciliation and Pass C: a script compares the readings with numeric tolerances. Agreed corrections are logged; remaining disagreements are re-inspected on the exact source page, including rendered page images where needed. Unresolved material figures are flagged and excluded from derived metrics, scores and rankings.
Profile records distinguish source-verified values (value and exact page re-confirmed), AI-verified values (independent readings agree, page not separately re-confirmed), corrections, unresolved fields and confirmed non-disclosures. Counts and extraction dates appear under Sources on each profile. Verification and correction reports are retained in the project record.
Limits of machine checks
Independent readings can make the same mistake, especially with scanned tables or ambiguous definitions. Verification tests our record against the document; it cannot make the document current, complete or accurate. Source inconsistencies are recorded, not silently repaired. Obtain the current FDD and investigate important figures yourself. Report a possible error with its primary source.
The evidence tiers
Every financially meaningful number on the site carries one of six tiers, shown as a colored marker and explained where it appears.
| Tier | Meaning |
|---|---|
| Disclosed | Stated by the franchisor in the cited FDD or another primary document. The citation names the document, its year, the Item number and the PDF page. |
| Public data | Taken from a reliable government or public dataset, cited to that dataset. |
| Derived | Arithmetic we performed on disclosed or public figures. The formula is shown with the number. |
| Model estimate | Our analytical estimate, produced from stated and editable assumptions. Never a franchisor figure and never presented as one. |
| Not disclosed | No reliable figure was found. We write that the item is not disclosed in the reviewed source. We do not fill gaps with guesses, category averages, or estimates borrowed from other brands. |
| Unresolved | Independent verification found the source itself ambiguous or self-contradictory on this figure. The flag stays on the page, and the figure is excluded from derived metrics, scores and rankings. |
Three rules govern how these are presented. Revenue is never presented as profit. An Item 19 figure is always shown with the population it represents. An estimate never looks like a disclosed number.
How we define key fields
The initial franchise fee vs. other Item 5 payments
“Initial franchise fee” on this site means the fee the FDD itself names the initial franchise fee (or its closest named equivalent), for a new single-unit franchisee on standard terms — never a sum of several Item 5 payments. Many FDDs also require other pre-opening payments to the franchisor or its affiliates in Item 5 — a mandatory POS or software purchase, an initial marketing contribution, a territory fee, an equipment package. Those are listed separately on each profile under “Other required initial payments to the franchisor”, and where a defensible total exists — stated by the FDD, or computable because every mandatory component is quantified — the profile shows a clearly labeled “Total Item 5 payments to franchisor/affiliates”. We do not call that combined amount the franchise fee.
AUV (average unit volume). The mean annual gross sales of the population of outlets stated in the franchisor’s Item 19, for the measurement period stated there. It is a revenue figure, not earnings. The population statement travels with the number everywhere it appears on the site. Where a franchisor discloses average sales per week or per month but no annual figure, we may annualize it arithmetically; that result is tiered as derived, the formula is shown, and it is not treated as a disclosed AUV.
Investment. The total estimated initial investment range disclosed in Item 7, low to high, for the format the profile describes. Item 7 ranges are franchisor estimates and their contents differ between brands: some include several months of working capital, some exclude real estate purchase, some assume a leased site. The profile records those assumptions and, where a franchisor publishes several Item 7 tables, names which format the headline range refers to and lists the alternatives.
Units. Outlet counts as reported at fiscal year end in Item 20 Table No. 1, split into franchised and company-owned. Where the tables cover United States outlets only, the profile says so.
Attrition. The share of franchised outlets that left the system during a fiscal year, calculated as terminations plus non-renewals plus outlets reacquired by the franchisor plus outlets that ceased operations for other reasons, divided by the number of franchised outlets at the start of that year, using the totals row of Item 20 Table No. 3. Transfers between franchisees are not attrition and are excluded. Note that reacquisitions include units the franchisor bought back for strategic reasons as well as units it took over from a failing operator; the FDD rarely distinguishes them.
Derived metrics and their formulas
AUV-to-investment ratio = disclosed annual AUV ÷ midpoint of the Item 7 total investment range. A rough measure of how much annual revenue a system’s typical unit produces per dollar of capital committed. It says nothing about margin.
Total fee load = royalty percentage + advertising fund percentage. The recurring share of gross sales owed to the franchisor and its marketing fund under the standard fee structure. It excludes technology fees, required local marketing spend, cooperative contributions and other charges that are not expressed as a percentage of sales; those are listed separately on the profile because they are not comparable across brands as a single percentage.
Three-year net franchised growth = (franchised units at the end of the most recent year − franchised units at the start of the three-year period) ÷ franchised units at the start of the period. Net of both openings and closures, and calculated on franchised outlets only, so that a franchisor converting company stores to franchises is not shown as organic system growth.
The scoring dimensions
We publish five computed dimensions, each rated 1 to 5, and each derived from disclosed data by the rules below. They are presented in two groups that must never be conflated. System performance — System Growth, Unit Stability and Investment Efficiency — measures how the system has performed, from the disclosed Items 7, 19 and 20. Evidence and disclosure quality — Financial Disclosure Quality and Evidence Confidence — measures how much the FDD discloses and how well-supported our record is. Missing data never produces a low performance score: where the inputs are absent, the performance dimension reads “Not enough evidence to rate” and the absence itself is visible, while the disclosure score reflects the thin disclosure. A brand is never marked a poor performer for disclosing little.
Two further exclusion rules apply. Where independent verification found a material unresolved inconsistency in the source’s own tables (see How we use AI) affecting a dimension’s inputs, that dimension reads “Not rated — unresolved source inconsistency” and the documented issue is shown instead of a number; the affected figures are likewise excluded from rankings. And a figure whose verification status is unresolved is excluded from every derived metric. Ties in rankings are broken by the underlying unrounded metric, then alphabetically by brand name; brands whose metric is excluded or missing do not appear in that ranking at all rather than ranking last.
Financial Disclosure Quality
This dimension rates how much the franchisor tells prospective franchisees about unit performance — not how good that performance is. Points are awarded on a scale of 0 to 5:
- An Item 19 financial performance representation is present: +1
- The representation gives an average and either a median or a distribution: +1
- The population covers at least half of franchised outlets, or is clearly described: +1
- Cost or profit data is included, not only sales: +1
- Multi-year or cohort data is provided: +1
Points convert directly to the rating: 5 points scores 5; 4 points scores 4; 3 points scores 3; 2 points scores 2; 0 or 1 point scores 1. A franchisor that makes no Item 19 representation at all scores 1, with a note explaining that omitting Item 19 is permitted by the FTC Franchise Rule and is not itself evidence of weak performance.
System Growth
Based on three-year net franchised growth as defined above.
| Three-year net franchised growth | Rating |
|---|---|
| 15% or more | 5 |
| 5% to 15% | 4 |
| 0% to 5% | 3 |
| −5% to 0% | 2 |
| Below −5% | 1 |
| Not derivable from the filing | Not rated |
Unit Stability
Based on average annual franchised attrition across the years reported in Item 20 Table No. 3.
| Average annual attrition | Rating |
|---|---|
| Below 2% | 5 |
| 2% to 4% | 4 |
| 4% to 6% | 3 |
| 6% to 10% | 2 |
| Above 10% | 1 |
| Not derivable from the filing | Not rated |
Investment Efficiency
Based on the AUV-to-investment ratio.
| AUV ÷ midpoint investment | Rating |
|---|---|
| 2.0 or more | 5 |
| 1.5 to 2.0 | 4 |
| 1.0 to 1.5 | 3 |
| 0.7 to 1.0 | 2 |
| Below 0.7 | 1 |
This dimension requires a disclosed annual AUV. Where none exists, it is not rated, and a franchisor that discloses nothing is never penalized with a low rating for doing so. Where the AUV population is a subset of the system — the top quartile, only units open several years, only one format, or a minority of outlets — the rating is flagged, because a subset average inflates the ratio relative to a system-wide one.
Evidence Confidence
A summary of how much of the record we could actually establish, combining three inputs: the share of twelve key fields that are disclosed, the recency of the document, and whether the record’s material fields have been independently machine-verified against the cited source document.
The twelve key fields are the Item 7 total investment low and high, the initial franchise fee, the royalty rate, the advertising fund contribution, franchised and company-owned unit counts at fiscal year end, three years of Item 20 franchised outlet status data, whether an Item 19 representation is present, the headline AUV, the initial term of the franchise agreement, and the Item 15 owner-involvement requirement.
The rating is assigned as follows. High: at least 80% of the key fields are disclosed, the primary document is the current FDD, and the record has been machine-verified. Medium: at least 50% of the key fields are disclosed. Low: anything below that. Evidence Confidence is a statement about our record, not about the franchisor: a brand can have a low Evidence Confidence rating simply because the newest registry filing is old.
Indicators we publish but do not score
Three further characteristics matter to a decision but do not reduce sensibly to a 1-to-5 scale, so we publish them as labeled indicators without a rating.
Franchisor Track Record — years franchising the concept, current system size, and what Items 3 and 4 disclose about litigation and bankruptcy. Litigation counts are not comparable across systems of different sizes and ages, and a large mature system will almost always disclose more matters than a small young one.
Multi-Unit Scalability — whether area development or multi-unit programs are disclosed, and whether Item 15 permits a trained manager to run day-to-day operations.
Operational Intensity — what Item 15 requires of the owner personally, from full-time owner-operation to manager-permitted or semi-absentee arrangements.
Why there is no composite score
We do not publish a single overall franchise score, and this is deliberate.
Combining these dimensions into one number would require weighting them against each other, and no defensible weighting exists: the relative importance of growth, stability, capital efficiency and disclosure depends entirely on what an individual buyer is trying to do and what they can absorb if it goes wrong. A composite would also quietly mix dimensions that measure the franchisor’s business with dimensions that measure the quality of its paperwork, and it would let a strong rating on one axis conceal a weak one on another. Worst of all, it would invite exactly the ranking-by-single-number treatment that makes franchise coverage misleading in the first place.
Our ranking pages therefore rank by one stated, published metric at a time, and always show the underlying figure next to the rank.
The illustrative unit-economics model
On profiles where an annual AUV and an Item 7 investment range are disclosed — and every material fee has passed verification — we publish an illustrative unit-economics model. It runs from revenue through operating-cost assumptions and the brand’s disclosed fee schedule to an estimated store-level EBITDA; then subtracts manager compensation to give a manager-run EBITDA; then subtracts illustrative debt service to give the headline the table actually prints: illustrative pre-tax cash flow — before taxes, capital expenditures and unmodeled fees. The model is disabled entirely for brands whose dominant fee a revenue slider cannot compute (for example a royalty defined as a share of split profits or gross margin), rather than approximated.
What goes into that arithmetic falls into five distinct categories, and the page labels each:
Disclosed inputs — taken from the FDD and computed exactly as disclosed, each line showing its arithmetic. The revenue anchor (Item 19 AUV). The verified fee schedule: percent-of-sales fees; tiered fees using the method the FDD itself establishes — marginal bands, whole-base repricing, or threshold-triggered rate changes, never a guessed treatment — with thresholds compared at their disclosed period (a monthly threshold is never compared to annual revenue); where a threshold-triggered rate depends on the prior year’s performance, the page offers both year contexts (threshold-crossing year vs. previously-qualified year) with the conservative one preselected; disclosed minimums; fixed recurring fees at their disclosed amounts; and mandatory variable fees at their disclosed floors, labeled as floors. The Item 7 midpoint anchors the debt-service illustration.
Model assumptions — numbers the FDD does not supply and we chose: category-level operating-cost ratios (cost of goods, payroll excluding the owner, occupancy, other operating costs), manager compensation, and the financing assumptions (loan share, interest rate, term). They are drawn from typical operating structures for that kind of business, not from the franchisor, and each is labeled assumption on the page.
Unmodeled mandatory fees — mandatory fees whose amounts the FDD does not state. These are never modeled at $0 and never silently dropped: they are listed beside the results as known exclusions, with an input for your own estimate, and the results row states that real outflows are higher by these amounts.
User-editable assumptions — every model assumption above, the revenue slider, the tier-year context where one applies, and your estimates for the unquantified fees. Changing any of them recomputes every output and every fee line’s arithmetic trace immediately; nothing is hidden inside the calculation.
Exclusions — categories the model deliberately does not attempt: income taxes; capital expenditures and equipment-replacement reserves; working-capital needs; ramp-up losses; owner draw and owner-specific costs; one-time and per-event fees (transfer, renewal, audit, late charges); and the unmodeled mandatory fees listed above. The result is not owner income, and the page says so directly under the table.
The model runs three cases. The base case uses the disclosed AUV. The downside case uses 80% of the disclosed AUV, or the disclosed bottom-quartile sales figure where the franchisor publishes one. The upside case uses 115% of the disclosed AUV.
Every output of the model is labeled Model Estimate. It is an arithmetic illustration of how a cost structure behaves at a given sales level, not a forecast, not a projection of what any outlet will earn, and not a statement of expected returns. Real outcomes are driven by site, rent, local wages, competition, management, and the operator, and the spread between outlets within a single system is routinely wider than the spread between the three cases here.
Limitations
The FDD is a disclosure document, not an audited operating report. Item 19 is optional under the FTC Franchise Rule, and franchisors that do publish one choose their own population, measurement period and statistic; two brands’ averages are frequently not comparable, and our tables cannot make them so.
Item 19 populations are often selective. Averages are pulled up by strong outlets and by the exclusion of weak or recently closed ones. An average is not a typical unit, and no figure on this site tells you what a specific site in a specific market would do.
Item 20 tables are compiled by franchisors and occasionally do not reconcile. Definitions of an outlet vary, restatements happen, and reacquisitions can reflect either strength or distress.
Item 7 ranges are estimates that exclude different things at different brands, so investment comparisons across brands are approximate even when both figures are disclosed accurately.
Our own process introduces error. AI-assisted extraction can misread a table, attach a figure to the wrong population, or miss a footnote that qualifies it. Automated validation catches contradictions, not every misreading. Registry lag means some profiles rest on older filings. Coverage is incomplete, and a brand’s absence from the site means only that we have not yet built its profile.
Finally, none of this measures the things that most often decide whether a franchise works: the franchisee’s own capability, the specific site, the local labor market, and the relationship with the franchisor’s field support.
Fee verification
Every recurring, conditional and one-time fee found in a brand’s Item 6 table (plus mandatory recurring costs disclosed in Items 7 and 11) is captured as a structured entry with its page citation and a per-fee verification status: verified (2-pass) — both independent readings captured the fee and agreed; verified (tie-break) — the readings disagreed or only one captured it, and a re-inspection of the source page decided it; single-pass — captured by one reading and not independently confirmed; unresolved — the readings disagree and nothing has decided it.
A single-pass status is permitted only for fees that cannot move modeled economics. Our materiality rule: a fee is material when it is not optional and either belongs to a core recurring category (royalty, minimum royalty, ad funds, local or cooperative advertising, technology, software/POS, call center) whatever its size; or it is modeled by the calculator and is at least 0.25% of sales or $1,000 a year; or it would be modeled but its amount is unstated — an unknown quantity is treated as material, never as zero. Every material fee must be independently verified or tie-break verified before a brand’s calculator is enabled; a material fee in single-pass or unresolved state disables that calculator. Tiered fees additionally require the FDD to establish how the bands compute (marginal vs. whole-base vs. threshold-triggered) and what period the thresholds measure; where the FDD does not say, the fee — and the calculator — stays off rather than guessing.
Verification status
Each profile shows whether — and when — its material fields were independently machine-verified against the cited source document, with counts: how many fields were confirmed with their exact page citation re-checked (“source-verified”), how many were confirmed by two independent readings without the page citation being separately re-confirmed (“AI-verified”), how many the source itself does not disclose, how many were corrected during verification, and how many remain unresolved because the source is ambiguous or self-contradictory. Unresolved figures are excluded from derived metrics, scores and rankings. No profile is labeled human-reviewed, because none has been; the verification process, its limits, and what it does and does not catch are described in How we use AI.
Corrections and franchisor submissions
If a figure on this site is wrong, tell us. Write to [email protected] with the page address, the field, the value you believe is correct, and the primary document that supports it, including the Item number and page. We welcome submissions from franchisors on the same terms as from anyone else: cite the primary document.
We correct factual errors and note material corrections on the affected page. We do not remove accurate, lawfully obtained information because a franchisor would prefer it were not published. The full process is set out in the Corrections Policy.