🌿 Cannabis🍷 Wine & Spirits
fuzzy.
Entity Resolution for Wine & Spirits Data
fuzzy.socal.com/fuzzy-tim/30min

Same bottle. A hundred ways to spell it.
One canonical record.

A single SKU shows up as "Caymus Cab Sauv 2019 750ml" in a distributor feed, "CAYMUS VINEYARDS NAPA CABERNET '19" in a retailer POS, and "Caymus 2019 Cabernet Sauvignon" in a wholesaler catalog. Multiply that across thousands of producers and dozens of vintages. Fuzzy resolves all of them automatically — and gets sharper with every allocation.

Caymus Cab Sauv 2019 750mlCAYMUS VINEYARDS NAPA CABERNET '19Caymus 2019 Cabernet Sauvignon
➜
Caymus VineyardsCabernet SauvignonNapa Valley2019 · 750ml
The Problem
  • Producer names never match. “Château Margaux”, “Ch. Margaux”, and “Margaux 1er Cru” — same estate, three rows, zero joins.
  • Vintages hide inside the string. ‘19, 2019, ‘19 Vintage, NV — every distributor, importer, and retailer encodes the year differently.
  • Bottle sizes are a free-for-all. 750ml, .75L, Standard, 1.5L, Magnum, 375ml Half — same juice, no consistent column.
  • Allocations and depletions break. Without a clean catalog, depletion reporting, allocation forecasting, and on-premise tracking are guesswork.
Why Fuzzy
Wine-native normalization. Understands producers, AVAs, varietals, vintages, and bottle-size aliases — 750ml = .75L = Standard.
Self-improving. Every cellar manager correction trains the system. The same allocation row is never re-decided.
Fully explainable. See exactly why every match was made — confidence scores, tier breakdown, reasoning. No black boxes.
Veto fields prevent false positives. Different vintage, different SKU. A 2018 Cabernet never collapses into a 2019 Cabernet.
Deeply configurable. Tune matching rules, vintage tolerance, bottle-size aliasing, and confidence thresholds — every distributor’s data is different.

Four-Tier Matching Pipeline

T1
Exact Match — Normalized
Free
T2
Fuzzy Consensus — 3/5 vote
Free
T3
LLM Reasoning — Ambiguous pairs
~$0.001
T4
Human Review — Feedback loop
You
80% free15% LLM5% human

LLM + Human Review

The LLM understands that “Dom Pérignon Brut Vintage” and “Dom Pérignon 2013” are the same listing when context fits. It resolves “Silver Oak Alexander Valley” vs “Silver Oak AV” by reasoning about wine geography. Cellar managers handle the final 5% — and every correction is cached permanently so the same pair is never re-decided.

“Ch. Lafite ‘19” → “Château Lafite Rothschild 2019” — LLM resolves abbreviation + vintage with 0.96 confidence

Intelligent Industry Warm-Up

Fuzzy learns your domain by asking targeted questions — not generic setup wizards. One answer applies across the entire dataset.

“I found 31 variations of ‘Opus One’. Should ‘Overture by Opus One’ be grouped with Opus One, or tracked as its own SKU?”
Your team's expertise stays with you. Fuzzy just learns from it.

Design Partner Pilot

  • Run Fuzzy against your actual distributor or retailer catalog — real results, not synthetic benchmarks
  • Wine-specific field mapping tuned to your inventory and allocation schemas
  • Shape the product roadmap alongside your cellar and operations team’s needs
  • Early access pricing — locked in before general availability
Let’s see if Fuzzy fits your wine & spirits catalog.
30-minute call — we’ll look at your producer/SKU matching pain together.
Book a Call