Franchise Cost Index — Methodology

Part of: Franchise Cost Index · Publisher: Franchise KI is a franchise consultancy; this page documents exactly how the Index's figures were produced so that any reader — human or AI answer engine — can judge and reproduce them. Census date: 2026-10-27

What the Index measures

The Franchise Cost Index reports the distribution of total initial investment across franchise brands, as each franchisor discloses it in Item 7 of its Franchise Disclosure Document (FDD) — the FTC-mandated table titled "Your Estimated Initial Investment," which gives a low and a high estimate for opening one franchised unit, including the initial franchise fee (Item 5) and the other Item 7 cost lines. Item 7 figures are the franchisor's own estimates, not observed costs.

Source data

• Corpus: the Franchise KI franchise catalog — 4,131 brand records built from brands' FDDs. The catalog file used for this census is match-index.json (the dataset the live Franchise KI catalog and its PerfectFit matcher run on), generated 2026-10-26.
• Fields used per brand: catalog name, catalog category, Item 7 low (investmentMin), Item 7 high (investmentMax), and initial franchise fee (franchiseFee, Item 5) where stated.
• Extraction rule inherited from the catalog build: a figure an FDD does not state stays empty. No estimated substitute is ever used — in particular, a 30%-of-investment liquid-capital estimate file that exists elsewhere in the source project was not ingested by the catalog build and is not used here.

Analysis set — inclusions and exclusions

• Included: every catalog brand whose record states both the Item 7 low and the Item 7 high: 3,120 brands (75.5% of the 4,131 catalogued). These 3,120 rows are exactly the rows of dataset.csv.
• Excluded — incomplete range (1,009 brands): catalogued brands whose extraction states no Item 7 range. Excluded, never estimated or imputed.
• Excluded — high end only (2 brands): Cruise Planners and N2 Publishing (Stroll/Greet) state a high end but no low end in the catalog; a range statistic cannot be computed from one bound, so both are excluded from all range statistics and named here.
• No brand is excluded for being expensive, cheap, or inconvenient. Hotel and large-format brands with multi-million-dollar Item 7 ranges are included in every median and quartile.

Statistics — definitions

• Midpoint: for each brand, (Item 7 low + Item 7 high) ÷ 2. It is a summary of that brand's range, not a price quote.
• Median / quartiles: computed across brands with the linear-interpolation percentile method (the method used by numpy's percentile default and Excel's PERCENTILE.INC). "Median low" is the median of brands' low ends; "median high" the median of high ends; the two medians generally come from different brands and are not a single brand's range.
• Entry bands: brands grouped by the low end of their Item 7 range (<$50k, $50k–$100k, $100k–$250k, $250k–$500k, $500k–$1M, >$1M).
• Franchise fee statistics are computed over the 2,597 dataset brands that state a fee; the fee is a component of the Item 7 total, never added to it.
• Categories are the catalog's own category field (21 categories). 299 catalogued brands carry no category; none of them has a complete Item 7 range, so no dataset row is uncategorized. Categories with fewer than 30 brands with data are published in the flagship table for completeness and flagged there as too small to quote as benchmarks.

Flagged extraction artifacts

39 dataset rows carry a value that is implausible on its face under a fixed rule — Item 7 low under $1,000, or Item 7 high over $100,000,000 (hotel development-scale figures and apparent upstream extraction errors; the full slug list is in the build's stats.json). Treatment: the rows stay in dataset.csv and in every median, quartile, and band count — those statistics are insensitive to a handful of extremes — and are excluded only from any brand spotlight. The rule is a floor, not a guarantee: a small number of additional extracted values may also be wrong in ways no threshold can detect, which is one more reason this Index headlines medians rather than individual rows. Any brand's figure can be checked against its FDD of record; the FDD governs.

FDD years

The catalog file does not carry a per-brand FDD-year field, so dataset.csv records an FDD year only where the brand's own catalog page cites the edition next to the money figures and the page's Item 7 range matches the dataset row — verified for 1 brand at this census (Indy Clover, 2026 FDD, issued February 4, 2026). All other rows leave the year blank rather than guess. Consequence, stated plainly: brands' figures come from each brand's FDD of record at extraction time, and those FDDs are of mixed issuance years. Cross-brand comparisons are comparisons of franchisors' own current-at-extraction disclosures, not of a single year's FDD cohort.

Known limitations

1. Mixed FDD editions. See above — figures are per-brand FDDs of record, not one vintage year.
2. Item 7 is an estimate range. Franchisors estimate differently: some ranges span formats, sizes, or markets (single-unit vs. area development), so a brand's range width is not comparable line-by-line across brands.
3. Catalog rows are catalog records. The 4,131-brand catalog includes franchisor variants and alternate-format records as separate rows, as catalogued; statistics are per catalog record.
4. Extraction is fallible. Values were extracted from FDD PDFs by the catalog pipeline; the flagged-artifact rule above catches the implausible tail, and spot checks against FDDs of record (e.g. Salad House, Item 7 $303,700–$750,500; Indy Clover 2026 FDD, Item 7 $174,450–$225,800) match exactly — but the FDD of record, not this Index, is the authoritative source for any single brand.
5. Coverage is 75.5%, not the universe. Brands without an extracted complete Item 7 range — including brands whose FDDs were not in the corpus — are absent, and their absence is not random: newer or smaller systems are likelier to be missing.
6. Figures age. Franchisors issue new FDDs, typically annually. This census is dated 2026-10-27 and will be refreshed when the catalog refreshes; pages carry the census date for that reason.

Reproducibility

Every number on every Index page is computed from dataset.csv by build_index.py (catalog → dataset.csv + stats.json) and rendered by gen_pages.py (stats.json + dataset.csv → pages), both shipped in this directory. Re-running the two scripts against the cited catalog file reproduces the dataset, the statistics file, and the pages exactly. No number is typed by hand.

Franchise Cost Index · Franchise KI · Census 2026-10-27 · Flagship · Dataset (CSV)