The Acquisition Index

Why this exists

Canadian small-business transaction data is not published anywhere. Buyers and sellers negotiate against anecdote, and the only benchmarks available are American, drawn from a different tax system, a different lending market and a different set of buyers. The index is an attempt to replace that with a measured record.

It works by observing public listings every day and recording what changed. A listing that disappears is the most valuable observation in the dataset — it has sold or been withdrawn — and it is also the easiest thing to record wrongly, because a firewall block and a withdrawn listing look identical if you only check whether a page loaded. Everything about how this is handled is set out in the methodology.

What the index can report today

The value of a longitudinal series is a function of how long it has run, and that is the one thing that cannot be accelerated. Continuous observation began on September 1, 2026, across 10 sources.

Publishing a duration or an outcome rate before the series can support it would be the fastest way to lose the right to be cited, so those figures wait. Where a figure does ship, it carries its sample size, and any cut too small to be honest is suppressed and marked as suppressed.

What the listings actually state

Coverage is uneven and stating it plainly is part of the method. Sellers advertise a price far more often than they advertise earnings, which is why this index leads with asking prices and treats earnings multiples as a longer-term goal.

What a city cut actually reaches

Province is recorded for every listing and a provincial cut carries no caveat. City is different, and by more than one mechanism: 18% of listings carry no location string at all, and another 6% name a region rather than a municipality. A city cut therefore reaches 75% of the corpus, and that number belongs next to any city figure.

Quebec is the sharpest case because its listings are geocoded mostly by administrative region — Montérégie, the Laurentians, Lanaudière, Chaudière-Appalaches — where most provinces record municipalities. Only 46% of Quebec listings sit at a named city.

The rest of the country is 77%, not “nearly all”: Ontario is 80%, British Columbia 75%, Alberta 64% and Manitoba 57%. Ontario is not purely municipal either — “GTA” is its third most common location at 380 listings. So Quebec is the extreme of a pattern the whole dataset has, not an exception to an otherwise clean rule, and a city ranking understates it for reasons unrelated to how many businesses are for sale there. None of this is corrected by inference: a region is not a city, and mapping Montérégie or “GTA” onto a plausible municipality would move a listing to the wrong place rather than leave it honestly unplaced.

How sector is decided, and why it was rebuilt

Sector is derived from the listing’s own title against a single fixed taxonomy. It used to be copied from each source site’s category instead, which meant the corpus carried thirteen incompatible taxonomies at once — a gas station was “Retail” on one site and “Automotive” on another.

Measured against eleven unambiguous business types named in the title, the stored sector matched the obvious answer 63% of the time. Deriving it from the title instead took that to 98%, and 2,197 listings were reclassified. Salons had been 1% correct, fast food 8%, gas stations 30%.

The old failure is worth stating because of its shape rather than its size. The deriver emitted labels that were not in the taxonomy at all, and those fell through to a generic “Business services” default — so a catch-all wore the name of a real sector, consistently, and nothing downstream could tell the difference. Where the title names no recognisable business type, roughly a third of listings, the sector is now left as the source gave it rather than guessed at.

The clock, and why two kinds of listing are not the same

Days on market needs a start date. For 9,640 listings that were already in the corpus when observation began, we know when we first captured them but never when the source first published them — the true duration is longer than anything we can see. For 5,240 listings we watched appear, the clock is exact.

The two are recorded separately and are never averaged together. Mixing them would bias every duration downward and produce a median that is simply wrong.