How Caddence Works

Where the data comes from

Every day we scrape product listings, prices, and stock indicators from 32+ UK golf retailers. The list spans national chains, regional pros, and used-club specialists. Each scrape captures the current price, available variants (shaft, flex, hand, loft), and whether the retailer is showing the product as in stock.

We don't pull from auction sites, listings without a fixed price, or retailers selling outside the UK.

Pipeline cadence is 24 hours. A price displayed on Caddence is at most one day old. Stock indicators are reliable but can lag a retailer's checkout by up to a day, so treat them as a guide, not a guarantee.

How we link the same product across stores

This is the hard part. The same driver might be listed as "TaylorMade Stealth 2 9° Stiff" on one site, "Stealth2 Driver 9 Stiff Ventus" on another, and "TM Stealth 2 9 degree" on a third. Out of the box, those look like three different products.

We run every listing through a three-tier matching engine:

1. Barcode (GTIN). When the retailer publishes a real barcode, we match on it. Exact, fast, no ambiguity.
2. Canonical product key. We maintain a curated graph of canonical golf products (brand, model, year, type, variant) and match each new listing against it. Listings are normalised before matching: retailer-specific language is stripped, abbreviations are expanded, variant attributes are extracted.
3. LLM-assisted linking. For the unmatched remainder (roughly 20% of listings), a language model proposes a match against the existing catalogue. Confident matches are auto-accepted; uncertain ones enter a human review queue before being promoted.

Match decisions are logged, auditable, and reversible. Roughly 78% of all listings are linked deterministically by the first two tiers; the LLM and review queue handle the rest.

What sits on top of the price data

Editorial coverage describes what the club is, who it suits, and where it fits in the brand's range. It's overlaid on the price data for products where we have it: specs (head size, MOI, adjustability for drivers; construction and cavity for irons; insert and head shape for putters), pros and cons, target player profile, key technologies, and competitor comparisons.

Editorial is clearly distinguished from raw price data. We don't claim to have personally tested every product. Specs are sourced from manufacturer data and reputable reviewers (MyGolfSpy, Today's Golfer, manufacturer spec sheets), and editorial reasoning is reviewed before it goes live.

Price history

Every product carries a full price history across every retailer we track. That history is what makes "is this actually a deal?" answerable. If the £80-off banner on a driver is the lowest it's been all year, the chart shows it. If it's the same price the same retailer sold it for three weeks ago, the chart shows that too.

We currently hold hundreds of thousands of price observations and are adding more every day.

How rankings work

On every product page, retailers are ranked by current price, cheapest first. That's the whole algorithm.

Some retailers have affiliate relationships with us; most don't. Affiliate status does not affect ranking, placement, or display priority. The cheapest price wins, even when it costs us the commission. If that ever changes, we'll say so on this page first.

What we don't claim

• We don't verify stock in real time. The "in stock" indicator reflects the retailer's last published state at scrape time.
• We don't aggregate used or refurbished products as direct equivalents to new ones. They're shown as separate listings where they exist.
• We don't aggregate prices across countries. Caddence is UK-only.
• We don't claim to have hands-on tested every product. Where we have, it's stated.
• We don't accept payment for higher placement, sponsored slots, or removal of competitor listings.

How we fix errors

Mislabeled products, missing retailers, broken matches, pricing bugs: email hello@caddence.golf. We read everything and most fixes ship within a day or two. Confirmed errors are corrected on the next pipeline run.