Every fine wine auction starts the same way: a collector decides to sell, and a consignment list lands on someone’s desk. The problem is that “consignment list” is a generous term. At a company like Acker Wines—the world’s largest fine and rare wine auction house and oldest wine store in the United States—these lists arrive in every format imaginable. Excel spreadsheets with comma-delimited wine names. Word documents where the consignor bolded the bottles they want appraised and left personal cellar notes mixed in. PDFs that are actually screenshots of Excel spreadsheets. Emails with free-form text descriptions. Google Drive links. Pictures taken of a screen. Documents in French, German, Mandarin. Nothing is ever formatted correctly, and every single one needs to end up in the same structured SQL Server template before an auction can proceed.
The scale of the problem is staggering. A major auction house processes 10,000 to 20,000 individual wines per month across auctions in New York, Hong Kong, Singapore, and Switzerland. Individual consignments range from 20 bottles to more than 1,200 line items. Each wine record needs to be broken into the same canonical fields—vintage, producer, wine name, designation, vineyard, bottle size, quantity, region, appellation, and supplier—before it can enter the system. When your CEO and auction directors call this a “top, top priority,” and one person on staff is building Excel formulas to handle it manually, something has to change.
This is not a hypothetical scenario. It is the daily reality at auction houses worldwide, and the gap between how consignment data arrives and how it needs to be structured is where enormous amounts of human time disappear.
Lido is the best option for wine auction houses that need to digitize consignment lists from collectors into structured, standardized catalog data. It extracts producer names, vintages, bottle formats, and quantities from any consignment format — handwritten lists, spreadsheets, photographed cellar inventories, and PDFs — without templates or per-format configuration. Auction teams using Lido process thousands of wines per month with the accuracy that fine wine valuation demands.
Wine nomenclature is inherently unstructured. A single wine can be described in dozens of valid ways. “2005 Domaine de la Romanée-Conti Romanee-Conti Grand Cru” contains a vintage, a producer, a wine name, and a designation—but the boundaries between those fields depend entirely on your knowledge of Burgundy. A machine that doesn’t understand wine classification will mangle the parsing. A machine that only understands wine classification won’t survive the formatting chaos of real-world consignment lists.
Consignors use whatever format is easiest for them, not for you. One collector sends an Excel file where commas inside a single cell separate the data: one comma means producer only, two commas mean designation plus producer, three commas mean designation plus vineyard plus producer. Another sends a Word document with personal tasting notes interleaved with the bottles they actually want to sell, distinguished only by bold formatting. A third sends a PDF that was printed from Excel and scanned back in, destroying any underlying cell structure. A fourth sends an email that reads like a letter: “I have about forty cases of Bordeaux from the 2009 and 2010 vintages, mostly first growths, plus some Burgundy from Rousseau and Roumier.”
The international dimension compounds every challenge. When you run auctions in Hong Kong, Singapore, Switzerland, and the United States, consignment lists arrive in English, French, German, and Mandarin. Wine regions have names in their native languages. A consignor in Geneva might list “Chambertin Clos de Bèze” while one in Hong Kong lists the same wine in traditional Chinese characters. Both need to resolve to the same canonical record in your database.
Volume is accelerating faster than headcount. International expansion means the pipeline of consignment lists is growing month over month. You cannot hire your way out of a problem where each new market doubles the variety of formats and languages hitting your intake process.
Manual Excel formulas are the most common first attempt. A technically inclined staff member builds a library of VLOOKUP, SUBSTITUTE, and text-parsing formulas to handle the most common formats. This works until it doesn’t. The formulas break on edge cases. They cannot handle PDFs or images. They require the operator to pre-sort each incoming file by format type and apply the right formula set. As one auction house data specialist put it: “It saves them a ton of time but costs me a ton of time.” The bottleneck doesn’t disappear—it just moves to one person.
Traditional OCR fails on wine-specific content. Off-the-shelf OCR tools can extract text from scanned PDFs and images, but they have no understanding of wine data structures. They will faithfully transcribe “05 DRC RC GC” without any ability to expand that into vintage 2005, producer Domaine de la Romanée-Conti, wine name Romanee-Conti, designation Grand Cru. The text extraction is only the first step. The hard part is the semantic parsing that comes after.
Rigid templates and intake forms do not survive contact with consignors. Some auction houses have tried mandating a standard submission template. Consignors ignore it, fill it out incorrectly, or simply attach their existing cellar inventory in whatever format they already have. When you are courting a collector with a million-dollar cellar, you do not send their list back and ask them to reformat it. You figure it out on your end.
General-purpose AI chatbots produce inconsistent results. Pasting a consignment list into a chatbot and asking it to parse the wines might work for a single list, but it is not a repeatable, auditable workflow. The output format varies between sessions. There is no structured mapping to your database schema. There is no way to process thousands of wines per day through a chat interface. And there is no quality assurance layer to catch the inevitable hallucinated vintages or misattributed producers.
Multi-format ingestion without manual pre-sorting. The system needs to accept Excel files, PDFs, Word documents, email body text, images, and Google Drive links. It needs to identify what kind of document it is looking at and apply the right extraction logic without a human telling it which parser to use. A PDF of a scanned Excel spreadsheet needs different treatment than a native Excel file, and both need different treatment than an email with free-form text.
Domain-aware parsing that understands wine taxonomy. Extracting text is not enough. The system must understand that “Lafite” is a producer, “Pauillac” is an appellation, “Premier Grand Cru Classé” is a designation, and “1982” is a vintage. It needs to handle the comma-delimited encoding schemes that individual consignors invent. It needs to distinguish between the wine data and the personal notes, formatting artifacts, and extraneous text that surround it.
Configurable output mapping to your exact database schema. Every auction house has its own system of record with its own field names, validation rules, and data types. The parsing output needs to map directly to your template—vintage, producer, wine name, designation, vineyard, bottle size, quantity, region, appellation, supplier—without requiring a developer to build a custom integration for each new field.
Multilingual support that resolves to canonical records. A wine listed in French, German, English, or Mandarin should all resolve to the same structured output. The system needs to normalize across languages and regional naming conventions, not just transliterate text.
Throughput that matches auction-house volume. Processing 10,000 to 20,000 wines per month is a baseline, not a ceiling. Individual consignments of 1,200 rows need to complete in minutes, not hours. And as international markets expand, the system needs to scale without proportional increases in staff time or compute cost.
A feedback loop for continuous improvement. No system will parse every wine list perfectly on day one. The auction house staff who correct errors need a way to feed those corrections back into the system so the same mistake is not repeated. As one auction executive noted, “We don’t expect out of the box to do this. But if we can train it, that changes everything.”
The core shift is from format-specific parsing to semantic understanding. Instead of building a separate parser for every file type and formatting convention, modern AI document processing reads the content the way a human would—understanding that a four-digit number near a wine name is probably a vintage, that “mag” means magnum, that bold text in a Word document signals emphasis. This is what makes it possible to handle the infinite variety of real-world consignment lists without requiring a dedicated formula or template for each one.
Lido approaches this problem by combining document extraction with configurable field mapping. You define your target schema once—the exact fields your SQL Server system expects—and the platform handles the translation from whatever the consignor sent to whatever your database needs. When a comma-delimited Excel cell contains “Romanee-Conti, Grand Cru, DRC,” the system parses that into wine name, designation, and producer based on learned patterns, not brittle string-splitting rules.
The multilingual challenge becomes manageable with AI models trained on global wine data. A consignment list from a Geneva-based collector writing in French gets the same structured output as one from a New York collector writing in English. The system recognizes that “Chambertin Clos de Bèze” and its English equivalent refer to the same vineyard and applies the correct canonical mapping regardless of source language.
Volume scales without proportional labor. Once the extraction rules are configured and validated, processing 20,000 wines per month requires the same staff effort as processing 2,000. The human role shifts from data entry to quality review—spot-checking parsed output, correcting edge cases, and refining the system’s accuracy over time. This is the difference between a process that breaks under growth and one that absorbs it.
Each correction improves future accuracy. When a staff member fixes a parsing error—say, the system misidentified a vineyard as a designation—that correction feeds back into the extraction logic. Over weeks and months, the system learns the specific patterns and conventions that your consignors use. The auction house staff member who was spending all their time on Excel formulas can redirect that expertise toward training a system that gets better with every consignment processed.
Faster consignment-to-catalog timelines compress your auction cycle. When consignment lists take days to process manually, your auction schedule is constrained by your data entry capacity. When they process in minutes, you can accept more consignments closer to auction date, accommodate late additions, and run more frequent sales without expanding your operations team.
Consignor experience improves when you stop asking them to change. The auction houses that win the best consignments are the ones that make selling easiest. When you can accept any format—a photo of a handwritten list, a Google Drive link to a messy spreadsheet, an email with free-form descriptions—you remove friction from the consignor relationship. The collector with 1,200 bottles does not need to spend a weekend reformatting their cellar inventory. They send what they have, and you handle the rest.
International expansion becomes operationally feasible. Opening an auction market in a new country used to mean hiring local staff who could read consignment lists in the local language and manually enter them into your system. Automated multilingual processing means your existing team can handle consignments from Hong Kong, Singapore, and Switzerland without language-specific bottlenecks.
Data quality becomes consistent and auditable. Manual data entry introduces variation. One person abbreviates “Cabernet Sauvignon” as “Cab Sauv” while another spells it out. One person enters bottle sizes as “750ml” while another uses “standard.” Automated extraction with a defined schema eliminates this inconsistency, which matters when your catalog data feeds pricing algorithms, collector search interfaces, and regulatory reporting.
If your team is spending hours reformatting wine lists that should take minutes, Lido can help you turn any consignment format into structured, database-ready records. Start with a free trial—no credit card required.
Yes. Modern AI-powered document processing goes beyond traditional OCR by handling native Excel files, Word documents, PDFs (including scanned images of spreadsheets), email body text, and even photos. The system identifies the document type automatically and applies the appropriate extraction logic without requiring manual pre-sorting. This means a PDF of a scanned Excel spreadsheet, a Word document with mixed formatting, and a plain-text email can all be processed through the same pipeline and produce identical structured output with fields like vintage, producer, wine name, designation, and quantity.
AI document processing uses semantic understanding rather than rigid string-splitting rules. Instead of relying on a specific delimiter pattern, the system recognizes wine terminology contextually. It understands that a four-digit number is likely a vintage, that certain words are known producers or appellations, and that formatting cues like commas, bold text, or cell boundaries carry meaning that varies by consignor. When one collector uses commas to separate producer from designation from vineyard and another uses line breaks, the AI interprets each convention based on learned patterns and maps both to the same structured output fields.
The system scales to handle tens of thousands of individual wine records per month. Major auction houses processing 10,000 to 20,000 wines monthly across multiple international markets can run all consignments through the same automated pipeline. Individual consignment lists ranging from 20 bottles to over 1,200 line items process in minutes rather than hours. Because the extraction is automated, increasing volume does not require proportional increases in staff time, making it feasible to expand into new auction markets without hiring additional data entry personnel for each region.
Yes. AI models trained on global wine data can process consignment lists written in English, French, German, Mandarin, and other languages. The system normalizes wine names, appellations, and designations across languages so that a bottle listed in French by a Geneva-based consignor and the same bottle listed in English by a New York collector both resolve to the same canonical database record. This multilingual capability is essential for auction houses operating across markets in the United States, Europe, and Asia where consignment lists arrive in the local language of each region.
%20(1).svg)
