AI for wine importers: five jobs to hand over, three to keep.

Every tool below follows one rule. A person approves every send. The tool can research, calculate, rank and draft. In what I build for you, nothing emails a buyer, a distributor or a regulator on its own.

The other rules are plain too. The tools connect read-only by default. They run in your own accounts, so your data stays with you. And I keep no copy after handover. If a tool breaks one of those rules, it isn’t ready.

An importer quotes the same wine into many states. In each one, the price from cellar to shelf is different, because excise, the margins and the rules on discounts are different. Do that by hand and you have a spreadsheet per state, an exchange rate somebody typed in last month, and a guess about whether a volume discount is even legal.

This is the best first job, because the answer is arithmetic. The price walk runs as plain code, not as an AI guess. I wrote about the steps of that walk. The AI part is the reading and the writing around it.

I run this myself. The diagnostic on my site has been live since July 2026, and on a verified run it took 297 seconds from an inquiry to a finished three-page PDF. For a client, I build the version that takes your whole range, your importer’s real margins and the states you sell in, then reprices one wine in all 50 states and DC in about seven seconds on a timed run, as a branded PDF your team can send. You can read how the pricing engine works.

What it costs to get wrong: a stale price in a quote. That’s why the walk carries a date.

Producers write to importers all year. So do buyers. Most of those messages wait in an inbox until somebody has an hour.

An inquiry desk reads each one as it comes in, looks the producer or buyer up, scores how well they fit your portfolio, and drafts a reply. A person reads the draft, edits it and sends it. Nothing waits unread, and nothing goes out unread either.

For an importer, the useful part is the scoring. You decide what a good fit looks like: the regions you buy, the price band, the volumes, the states you cover. The desk applies that every time, the same way, at 2 a.m. or on a Friday afternoon. It will also say “no” for you, politely and quickly, to the producers who were never going to fit.

What it costs to get wrong: a bad draft that a tired person approves. The fix is a rule-based check in front of the person, with prices and commitments held back unless somebody has approved them. I run that kind of check on the AI drafts I send myself, and I’d build it in front of yours: how the approve-before-send check works.

Ask any importer where last year’s price list is, and you’ll hear three answers. Past quotes sit in sent mail, tech sheets sit in a shared folder, and call notes sit in somebody’s notebook.

Trade memory puts all of that in one place. Price lists, tech sheets, past quotes and call notes go in, each one dated and searchable. When you ask “what did we quote that distributor for the 2023 rosé?” the answer comes back with the document it came from. Every answer is cited, so you can check it.

The dates are the point. The most expensive mistake in this job is quoting last year’s price, and a good trade memory retires superseded lists so it can’t happen. It runs in your own accounts, so the documents never leave your control.

What it costs to get wrong: an answer with no source. So the tool shows its source or says it doesn’t know.

Every importer needs the next 25 or 100 accounts. The usual method is a rep, a few directories and a lot of memory, one city at a time, and nobody can say why one bar ranks above another.

Account Finder does this from public records: federal permits, state wholesale and retail licenses and city licenses, refreshed every day. At the time of writing it holds 1.47 million rows across all 50 states. A fixed score in plain code ranks each venue for a wine and a place. A language model writes one line on why each top account fits, and it never invents a venue: every row comes from a record.

For a client, I build a private version. You put in your best accounts, and it gives back the look-alike accounts you should call next. Your reps decide who to call. The tool ranks and explains. It never contacts anyone. See how Account Finder works.

What it costs to get wrong: calls to the wrong places. Because every row has a record behind it, a rep can check in seconds.

After a buyer call or a tasting, the follow-ups sit in someone’s head until Friday. Half of them never happen.

A follow-up drafter takes the notes (typed, dictated or from a transcript), pulls out who promised what, and writes the thank-you and the next-step email as drafts. You read them, change what you like and send them yourself.

One caution. Transcribing a live call depends on recording-consent rules, which differ by state and by country. I’m piloting a meeting-notes tool that starts only with everyone’s agreement, and so far it’s run on one real call and one simulated call. Any version I build for you starts with a written consent check. Drafting from your own typed or dictated notes doesn’t record anyone else. See the meeting-notes pilot.

What it costs to get wrong: a draft that promises something nobody agreed to. That’s the same approve-before-send rule again.

Now the other side. These three stay with you, and no tool of mine will take them.

1. The relationship. An importer’s real asset is trust: with producers who give you their best lots, with distributors who push your brands, with buyers who call you first. AI can draft the email. It can’t be the person a buyer trusts after a bad vintage or a late shipment. Pick up the phone for the important calls. Use the tools for the time they give back.

2. Pricing judgment. The tools calculate. They don’t decide. Whether to hold a price in New York and give way in Texas, whether to take a thin margin to win a chain, which producer to back when two wines compete for the same slot: that’s judgment, and it comes from your knowledge of your market. The tool shows you the walk and the range. You make the call.

3. Compliance filings. Your permits, label approvals, import filings, state registrations and price postings are legal records. A tool can help you find a notice or draft a checklist. It should never be the record itself, and nothing it produces should go to a regulator without a person checking it against the source. A wrong number in a filing costs far more than the time it saves, and the person who signs is the person who answers for it. I don’t sell anything that becomes the record for a TTB or state filing.

Don’t start with the tool. Start with the week. Write down the hours that go on pricing, inquiries, finding old quotes, building account lists and follow-ups. Then pick the job where the answer is clearest and a mistake is cheapest. For most importers that’s job 1 or job 3.

That’s what the AI Opportunity Review does. It’s half a day with you and your team, then a written ranking of the three builds worth doing, each with a price band and what it saves or sells, and a sample list of 25 accounts. If nothing’s worth building, the report says so and I don’t propose a build. The price is fixed and quoted in writing after our first call.

Every proposal starts with a written check of your distributors and accounts against my day job. If it finds an overlap, you hear about it in writing before any proposal, and you decide whether we go on. I never sell for you. I build what your own people use to sell.

If that sounds useful, Let’s talk. The first call is free, and I’ll run one of your wines through the shelf-price walk while we’re on it.