Short answer: “AI” is an umbrella word, not a product. For a distillery it covers three very different things — dashboards and rules (often mislabelled AI), machine learning that predicts numbers from your own data, and generative AI that drafts and explains in plain language. They need different data, carry different risks and fail in different ways. The single most useful skill is not buying AI — it’s knowing which of the three a vendor is actually selling you.
This is the first post in a five-part series for distillers who keep hearing “AI”, “machine learning” and “generative AI” and want a straight, working definition of each — not a sales deck. No hype, as always. Where these tools earn their place in a distillery, I’ll show you. Where they confidently lie, I’ll show you that too.
The word does too much work
“AI” has become a word that means whatever the person saying it wants to sell. A spreadsheet macro that flags casks over a certain age gets called AI. A genuine machine-learning model that forecasts the angel’s share gets called AI. A chatbot that drafts your monthly report gets called AI. All three wear the same badge, and that is exactly why the word is dangerous on its own — it hides the question that actually matters.
The useful definition is simple. AI is software that performs a task we used to assume needed a human — recognising a pattern, predicting a number, writing a sentence. That is broad on purpose, because the field is broad. The work for you is not memorising a textbook definition; it’s learning to split that umbrella into the parts that behave differently.
The three forms you will actually meet
Rules and dashboards. “Highlight every cask past its planned bottling window.” “Total this month’s regauges by warehouse.” This is logic you specify — fast, transparent, and completely predictable. It is often the most valuable thing a distillery can build, and it is frequently sold to you as “AI”. It isn’t, and that’s fine — but you should know you’re paying AI prices for a report if that’s what’s happening.
Machine learning. Instead of you writing the rule, the software learns the pattern from your historical records — then applies it to a new case. “Given this cask’s fill strength, wood, position and warehouse, what loss should I expect by year ten?” It only works if your past data is real and measured, and it is covered in the next post. This is the branch most people mean by “AI” in a serious operational sense.
Generative AI. The newest form, and the one behind the chatbots everyone has tried. It produces new output — text, a table, a summary — in plain language. A Claude copilot that reads your cask ledger and drafts the maturing-stock narrative is generative AI. It’s brilliant at drafting and explaining and genuinely unreliable at remembering specific numbers, which is its own post later in this series.
Why keeping them separate protects you
Each form needs different things and fails in different ways. Rules need someone who knows the business logic. Machine learning needs clean historical data and breaks quietly when the world shifts away from what it learned. Generative AI needs grounding in your real documents and will state a wrong cask figure in the same confident tone as a right one. Lump them together as “AI” and you can’t reason about cost, risk or readiness — you can only react to a brochure.
So when a vendor, a board member or a LinkedIn post says “AI”, the move is always the same: which of the three do you mean, what data does it need, and how will I know when it’s wrong?
Where this breaks
Honesty is the house rule here, so: the three-way split is a working model, not a hard wall. Modern systems blend them — a generative copilot might call a machine-learning model under the hood, which reads from a rules-based dashboard. The categories blur at the edges. They are still worth holding in your head, because the questions they force — what data, what risk, who’s accountable — stay sharp even when the technology mixes. And one caution that outlasts every definition: none of this matters if your records are estimates. AI of any kind sits on top of measured data. A distillery that doesn’t record every regauge has nothing for any of these three to stand on.
The bottom line
“AI” is not one thing you buy — it’s an umbrella over three tools that behave nothing alike. Rules and dashboards run on logic you write. Machine learning predicts from your own measured history. Generative AI drafts and explains in words, confidently, sometimes wrongly. Learn to ask which one is on the table, and you’ve already outgrown most of the hype. The rest of this series takes each in turn — starting with machine learning, the one most likely to be quietly doing real work in a distillery.
Frequently asked questions
What does AI actually mean for a distillery? AI is an umbrella word for software that does tasks we used to think needed a person — spotting patterns, predicting numbers, writing text. For a distillery it shows up in three practical forms: dashboards and rules (not really AI but often sold as it), machine learning that predicts from your own data, and generative AI that drafts and explains in plain language. They are not the same thing and they fail in different ways.
Is AI the same as machine learning? No. Machine learning is one branch of AI — the part that learns patterns from data. “AI” is the broad umbrella; under it sit machine learning and generative AI. When a vendor says “AI”, ask which one they mean, because the data you need and the risks you carry are completely different.
Do I need AI to run a distillery? No — you need good records first. AI sits on top of measured data. A distillery with a real cask ledger and honest regauge history can get value from AI; one running on estimates and memory cannot, no matter what it buys.