Skip to content
PoolBuilderIndex

How PoolBuilderIndex ranks companies

Last updated:By The Pool Trade Desk

Every company is ranked by one published formula, and no company can pay to move up. This page shows the exact math: one score out of 100, built from three pillars, with every point of it listed and the wording of every claim held to what was actually checked. It also sets out how the index itself is built, including where a language model is used and where it is not. If one listing sits above another, this is precisely why.

How does PoolBuilderIndex rank companies?

Each company carries one number out of 100, and every roster on this site is sorted by it. It is built from three pillars, worth 45, 30 and 25 points, and every point of it is listed further down this page. It is deterministic: the same data always produces the same order. Across the 442 companies listed here the scores run from 12 to 84, with the middle company on 56.

score = reviews (45) + own_statements (30) + record (25)

out of 100, highest first

What changed, and when. Until 30 August 2026 the order came from the star rating alone, adjusted for review count, plus four small boosts for having a website, a phone number, photographs and posted hours. That formula had stopped separating anybody: every company here posts hours, most have a phone and photographs, and the ratings cluster at 4.59★. Meanwhile 442 companies' own websites had been fetched on 19 August 2026 and no page on this site read a word of it. The pillars below spend that work.

What is every point worth?

All of it, in three tables. These are generated from the code that does the ranking, so they cannot drift away from it.

Reviews and reputation45/100

The only signal here that comes from customers rather than from the company.

What earns itPoints
Adjusted Google rating, from 3.5★ up to 5.0★30
Review volume, log-scaled and capped at 20015

What it states on its own site30/100

Sentences read off the company's own website. Quoted, never checked with any registry.

What earns itPoints
Says it is licensed5
Says it is insured4
Says it carries WSIB coverage3
Says it is bonded2
…those four together cannot exceed12
Each trade body it says it belongs to (capped at 10)5
Says it has been going 20 years or more8
…10 to 19 years6
…3 to 9 years4
…under 3 years2

Website and record25/100

What this site holds on file, and what happened when its website was fetched.

What earns itPoints
Website answered when we fetched it12
Website is there and turned our crawler away9
Website did not answer at all2
No website on file0
3 or more photographs on file5
Phone number on file3
Online booking on its own site2
3 or more pool-trade phrases in its own material3

What is a Bayesian-adjusted rating?

A raw star rating is misleading when the review count is tiny. The rating half of the first pillar blends each company's rating with the dataset average, treating every company as if it started with 10 imaginary reviews at the mean of 4.59 stars. A company with thousands of reviews barely moves, while a 5.0 built on four reviews is pulled toward the average until it earns more history. That adjusted figure is then mapped from 3.5★ up to 5.0★ onto 30 points, and the count itself is worth a further 15, log-scaled so the two-hundredth review is worth far less than the tenth.

This is the same shrinkage approach behind IMDb's Top 250. A company with no rating at all is scored on the market baseline rather than on zero, because having no reviews yet is not the same as having bad ones.

“Says it is licensed”, and why not just “licensed”?

Because this site has not checked. On 19 August 2026 a crawler fetched every company website on file and read the pages. Where a company wrote that it is licensed, insured, bonded or WSIB-covered, that sentence is recorded with the company, quoted word for word, and worth the points in the table above. Nobody rang TSSA. Nobody called an insurer. So the wording on every chip and every profile row stays at its ceiling: says it is licensed, never “licensed”, and never “verified”.

Two kinds of match are thrown away rather than scored. A sentence that tells the reader to check licensing themselves is advice, not a claim, and a credential named in a quoted customer review or in a list of somebody else's awards belongs to somebody else. Where a company's page carried no such sentence, the company simply earns nothing here: no sentence, no claim, no points, and nothing invented to fill the gap. The full set of quotes is public, in data/enrichment/enrichment.json, keyed by Google place id.

What does it mean when a company's website did not answer?

Less than it sounds. A fetch from a data centre can fail for reasons that have nothing to do with whether a company is trading, so the third pillar reads each outcome the way this table says, and no listing is ever removed, hidden or dropped down the page for being unreachable.

Every address on file was fetched on 19 August 2026. Out of 12 points for the check.
What happenedHow we read itPoints
Answered normallyIts site is up12
403, or a robots.txt refusalThere is a site and it turned our crawler away. A person with a browser is very likely served it.9
Timed outOne slow answer is not evidence a host is gone9
404 or 410The page is genuinely gone2
No DNS, refused connection, or the server's own 5xxNothing answers at that address2
Answers as somebody else's businessA lapsed domain gets rebought. Answering is not being who you say you are.2
No address on fileNothing to fetch, and nothing claimed0

A closed business and a company whose web host went down look identical from here, and this site does not pretend to tell them apart. Nothing on it has been checked for closure.

What does the score look like on a real company?

Not an invented example. This is the top-ranked company on the site right now, taken apart by the same function that ranked it. The same breakdown sits on every profile page.

  • Reviews41/45
  • Its own statements20/30
  • Its record23/25
  • Total84/100

Which companies are excluded or ranked last?

We do not screen out closed businesses, and this page used to say we did. The records behind this site were collected in August 2026and the export carries no closed-or-open field, so nothing has ever been dropped for being closed; check the company's own Google record before you rely on it. We do drop records that turn out to be facilities rather than companies: a municipal swimming pool or a City by-law page is not somebody you can hire. A company with very few reviews or a low rating keeps its place on every roster it belongs on, but ranks last and stays out of the top-rated modules until it builds a track record. It gets a profile page of its own only when the public record carries review text to put on it. 57 of the 442 companies here have none, so they appear in the lists without a page behind the name.

Every web address on this site was fetched on August 26, 2026. 5 of them are not linked from their profile: the domain no longer answers, or it answers with a business that has nothing to do with the listing, because a lapsed domain gets rebought. A bot block, a slow response or a “coming soon” page is not counted, because none of those is evidence a site is gone. The company stays listed either way: a lapsed domain is not evidence anyone stopped trading.

Why is a company from the next town over on my area's page?

A company is filed under an area when its registered address sits within 30 km of that area's centre. Pool companies travel: a builder in Vaughan works across half the north GTA, and a service route runs through several municipalities in a morning. Filing on the postal address alone would hide most of the companies that actually turn up. The distance is measured from the coordinates on the Google record, so it is arithmetic, and every page prints the addresses the companies publish so you can judge the travel for yourself.

How is this index built, and where is AI used?

The index is built in four steps, and we would rather you knew all four.

  1. 1. Scraped from public data

    Company records start as public Google Maps listings: name, address, phone, website, category, rating, review count, hours and photos.

  2. 2. Filtered

    Records sharing a Google place ID are collapsed into one, and a relevance filter removes companies that are not in the pool trades, reading the text of their own website as evidence. This is a filter on the public record, not a check of the business: we do not verify licences, insurance or workmanship.

  3. 3. AI-assisted editorial

    Category assignment, listing summaries, page copy and cost-guide drafts are AI-assisted. A language model writes the first pass, working only from the scraped record and named sources. It never invents a rating, a review, a price or a credential.

  4. 4. Human-reviewed

    The Pool Trade Desk reviews the dataset, the rankings and every price guide before publication, and reviews corrections as they arrive.

Two things are never model-written: the numbers and the ranking. Ratings, review counts, phone numbers and addresses are copied from the source record, and the position of every listing comes from the score set out above, computed in code.

Where does the data come from?

Ratings, review counts and business details come from public Google Maps profiles, then each record is enriched from the company's own website for services, photos and contact details. Market pricing across 8 categories is researched separately, and every figure cites a named GTA company or authority, so you can trace it. We do not scrape private data and we do not buy reviews.

How often is the data refreshed?

The current dataset is stamped August 2026, and that is the whole of it: every rating, review count and position on this site comes from that one scrape. There is no live feed and nothing re-ranks itself overnight. Positions move when the dataset is rebuilt by hand and the site is redeployed, and the footer prints the date of the scrape the page was built from so you can see how old the numbers are. If a business is sold, closed or corrects its details, email us and we change the record without waiting for a rebuild.

What are the limits of these rankings?

A ranking is a strong starting point, not a guarantee. The reviews are Google's, not ours, so they carry Google's own biases. We do not verify a company's WSIB coverage, liability insurance or licences, and a high rank does not vouch for any single job. A pool is a large, permanent build: get written quotes from more than one company, check insurance directly, confirm who pulls the municipal permits and the required pool fencing, and read recent reviews before you sign. We help you build the shortlist. The due diligence is still yours.

Read our editorial standards

Now go and read the ranked list this method produced

No obligation. Requests reach our desk, never a company.