How we build our M&A multiples data

A transaction enters the database only with a specific, fetchable page behind it: an SEC filing, a company press release, or trade-press coverage. There is no modelled data, no index-derived estimate, and no licensed third-party feed. A record without a source URL is rejected at import.

Only what a source page literally says is recorded. In particular, none of the following is treated as a disclosed figure:

• A back-calculated number. Price divided by a multiple the source quoted is arithmetic, not disclosure, and it carries a false precision.
• Forward contribution guidance.“Expected to add approximately $11 million of adjusted EBITDA annually” describes what a buyer thinks it is getting, often synergy-inclusive. It is not the target’s trailing earnings.
• Equity value presented as enterprise value.Consideration “net of debt and transaction expenses” is not EV, and summing disclosed components into a total the filing never states is not either.
• The price of a partial stake.A figure that buys 80% of a company, with an option on the rest, is not that company’s enterprise value, and no multiple built on it is valid.
• A range midpoint, or a qualitative phrase.“$20 million to $25 million” is not $22.5M, and “in the low 2x’s” is not 2.9x.

Records are checked against their own sources by a separate verification pass that re-fetches the URL and looks for the quoted sentence. Every example above is a real record that pass struck.

When two sources describe the same deal, the more authoritative one governs: SEC filing, then company press release, then trade press. A more authoritative source can supersede a figure already recorded; a weaker one can only fill a gap it left. Either way the disagreement is written into the record rather than quietly resolved.

An accurate number attached to the wrong kind of event is still a bad row. A transaction is excluded when there is no named acquirer (an ongoing sale process, non-binding bids, “exploring strategic alternatives”), when it is a rumour or a terminated deal, when it is a financing, minority investment or joint venture rather than a change of control, or when the “target” is a product line or contract rather than a business with its own earnings.

Many sources disclose an enterprise value and a multiple but never the earnings. Where that happens the implied EBITDA or revenue is derived arithmetically and flagged as derived — never overwriting a disclosed figure. The derivation is refused outright when it produces something implausible: an EV/EBITDA outside 1×–30×, an EV/Revenue outside 0.1×–15×, or an implied EBITDA margin above 60%. Deals whose size cannot be established either way sit outside every size band.

The same deal is routinely announced by the buyer, filed with the SEC, and written up in trade press, each spelling the parties differently. Counted twice, a single deal moves a median — and in a sector whose whole sample is eight transactions, it moves it a lot. Worse, two records for one deal can carry two different multiples, in which case the median is computed over a deal counted twice at two different prices.

So every transaction has a stable identifier derived from the acquirer, the target and the year, and a record that matches an existing deal is merged into it rather than inserted. Merging fills gaps and never overwrites a disclosed value.

Name matching alone is not enough in either direction. Matching only exact strings misses “Hansen & Adkins” against “Hansen & Adkins Auto Transport, Inc.”; matching loosely fuses genuinely different companies. Both have happened here.

Names are therefore compared on whole words rather than raw character runs — otherwise a company called IES matches a dry cleaner called Sudsies — and a match resting on a single shared word is treated as weak. A weak match needs the acquirer to agree before it counts. Where the acquirer and target both match, that is conclusive on its own regardless of the year, since a company can only be bought once; where only the target matches, the year has to corroborate, because a business genuinely can change hands twice.

Published multiples cover targets at or below $25Mof EBITDA. This matters more than it sounds: disclosure skews hard toward large deals, because large public acquirers are the ones obliged to disclose. Across this same database, deals above $50M of EBITDA run several turns higher than deals below $5M. An unbanded “M&A multiple” quoted to the owner of a small business is not a slightly optimistic number; it is roughly double the right one.

The valuation estimator narrows further, benchmarking a business against transactions near its own size before falling back to the wider band, and falling back to conservative defaults rather than a mega-deal median when the size-comparable evidence is thin.

A sector is published only once at least eight qualifying transactions stand behind it, and the sample size is printed beside every figure. Below the floor the sector shows a conservative default range, labelled as a default rather than dressed up as data. Sectors cross the floor on their own as the database grows — and can fall back below it when a record is withdrawn, which is why the counts move.

Every figure is reported as 25th percentile, median and 75th percentile. A single point estimate hides the dispersion that actually determines what one particular business is worth, and an average lets one outlier speak for a whole sector.

The live tables recompute as the database grows, which is right for a working tool and useless for citation: a figure that moves cannot responsibly be quoted. So each quarter is frozen into a permanent snapshot at its own URL and is never recomputed.

When a published snapshot turns out to be wrong, it is restated rather than edited. The original date stands, the correction date and reason are recorded, the superseded per-sector figures stay in the file, and the page shows what moved — including any sector withdrawn because its corrected sample fell below the publishing floor.

Published snapshots: 2026-Q3.