Methodology
How The Production Index ranks vendors
Every vendor is scored 0 to 100 across five dimensions, weighted for the category being ranked. Weights are published on each ranking page above the ranking, are set before the scoring runs, and apply to every vendor in that category.
The five dimensions
Each dimension is derived from public, checkable evidence. The arithmetic is given in full so a reader can recompute any score on any ranking page rather than take the total on trust.
Rating quality
The mean star rating, discounted by the share of ratings below four stars. A clean record counts for more than a high average with a bad tail hidden inside it.
How it is scored. score = (mean / 5) x 100 x (1 - share of ratings below 4 stars). A vendor with a published mean but no rating breakdown is scored on the mean alone and flagged.
Review depth
How much evidence sits behind the rating, measured against the other vendors in the same category rather than against the whole market.
How it is scored. score = (this vendor's review count / the highest review count in this category) x 100. Normalised within the category on purpose: a specialist with 20 reviews in a narrow category is not behind a generalist with 100 in a broad one.
Service depth
How much itemised capability the vendor actually publishes, rather than how much it claims in prose.
How it is scored. score = (this vendor's count of published, itemised services / the highest count in this category) x 100, taken from the services block of the Google Business Profile.
Price transparency
Whether a buyer can see a number before making contact. Almost nobody in this market publishes one, which is exactly why it separates vendors.
How it is scored. score = 100 where the vendor publishes a rate or a starting price a buyer can read without contacting them, 0 where a fully audited profile publishes none. Unscored where we have not audited the profile.
Operational proof
Checkable evidence that there is a real operation behind the listing: an owner-claimed profile, a street address, work performed at the client site, a photographed inventory, and breadth of registered category.
How it is scored. Five checks worth 20 points each: profile claimed by the owner; a street address published; the on-site services attribute set; 25 or more photos; three or more Google categories.
Why review count is not the main criterion
Counting reviews produces an ordering that Google itself disagrees with, and the San Francisco data says so plainly. On 2026-08-01 a vendor with no reviews at all held a top-three local-pack position for hybrid event production in this city, and a vendor with 21 reviews held first position for conference AV ahead of one with 52. Raw volume is not what Google is weighing, and it is not what a buyer should weigh either.
Review data still matters, in two separate forms. Rating quality asks whether anyone left unhappy, which a mean alone hides: a 4.4 average across 272 ratings can conceal 31 one-star reviews. Review depth asks how much evidence sits behind the rating, normalised inside the category so a specialist with 20 reviews in a narrow category is not ranked behind a generalist with 100 in a broad one.
Why weights change by category
A buyer renting an LED wall for one evening and a buyer commissioning a three-day conference are not weighing the same things. Fixed weights across every category would flatter whichever category the weights happened to suit. Each ranking sets its own, and states why:
| Category | Rating quality | Review depth | Service depth | Price transparency | Operational proof |
|---|---|---|---|---|---|
| Conference AV, San Francisco | 25 | 20 | 25 | 15 | 15 |
| Trade Show AV, San Francisco | 25 | 20 | 25 | 15 | 15 |
| LED Wall Rental, San Francisco | 20 | 20 | 25 | 20 | 15 |
| Teleprompter Rental, San Francisco | 25 | 15 | 25 | 20 | 15 |
| Video Production, San Francisco | 25 | 25 | 25 | 10 | 15 |
The constraint that keeps this honest is that the weights go up before the ranking is computed and are applied to every vendor in the category. They are not adjusted after seeing who wins.
What happens when we cannot verify something
An unverified dimension scores null, not zero. Null dimensions are dropped and the remaining weights are renormalised, so a vendor is never punished for a gap in our research. Every ranking page prints how many of the five dimensions each vendor was scored on, so a total built from two dimensions is never presented as if it were built from five.
A vendor we cannot score at all is listed without a score in the “also serving” section rather than dropped from the page. Being absent from Google is a gap in what a company publishes, not evidence about its work.
When our order disagrees with Google
It sometimes does, and we leave it. Each ranking page shows Google’s local pack for that query on the date we checked, in the order Google showed it, directly beside our own. Where the two differ, the page says so and gives the reason. Reordering to match Google would make the rubric decorative.
Sourcing standard
Every figure carries a source and the date it was checked. Ratings, review counts, rating breakdowns, published service counts, categories, prices and profile attributes come from Google Business Profiles pulled through the Google Business Profile. Local-pack positions come from the Google search results, localised to the city being ranked. Scope and years-in-business claims taken from a vendor’s own website are recorded as that vendor’s claim and labelled as such rather than stated as fact.
Nothing is estimated to fill a gap. If a figure is not in the sources above, it is not on the page.
How often this is refreshed
Vendor data is re-pulled monthly and the full rubric is re-scored quarterly. The verified date on a ranking page changes only when the underlying figures have actually been re-checked. A date that moved without a re-check would be a fabricated data point, so the dates on this site are not bumped for freshness.
Paid placement
There is none. No vendor can buy a position, and there is no tier that changes an order. If that ever changes, any paid element will be labelled on the page it appears on and kept out of the scoring.
Corrections
If a figure here is wrong or out of date, we want to fix it. Vendors can ask for a re-check of any figure attributed to them, and a re-check that changes a score changes the ranking, in whichever direction the data goes.