Features · AI Bartender
An AI bartender, built for craft beer.
Barley is a recommendation engine, not a chatbot. It learns each customer's taste from real Square POS data, speaks the language of craft beer (ABV, IBU, dry hop, lambic, barrel-aged), and recommends what to pour next — grounded in your actual menu, never made up.
34%
Reported lift in repeat visits at The Keg Stand
Real-time
Each transaction sharpens the taste profile within seconds
Menu-grounded
Only recommends beers you're actually pouring — no hallucinations
Opt-in
Customers control their data; one-tap reversible
What an AI bartender actually does
The job of a great bartender is to remember what each regular likes, recommend something they'll love, and tell them when a new beer they'll go for hits the tap. Barley does that for every customer, at scale, every day.
Remembers every customer
Each guest's purchase history becomes a taste graph: hop preferences, ABV range, can vs. draft, time-of-day patterns, repeat favorites.
Recommends what to pour next
Customer asks 'what should I drink?' or browses the menu — Barley suggests beers from your live tap list ordered by predicted taste match.
Speaks craft-beer fluently
Knows that Citra is tropical-dank, that lambics need explanation, that imperial stouts pair with cold weather. Generic LLMs guess; Barley knows.
Pings the right people on release
A new beer hits — Barley alerts the customers whose history says they'll love it. Segmented messaging, not blast.
Grounded in your menu
Recommendations only come from beers you're currently pouring or selling. Cannot invent beers, prices, awards, or specs.
Privacy-first by design
Customers opt in explicitly, opt out one-tap, and never have their data shared across breweries.
Generic AI chatbot vs. purpose-built AI bartender
Most "AI for hospitality" tools are a chat UI bolted to a general-purpose LLM. The differences show up the first time a customer asks something specific.
| Capability | Generic chatbot (ChatGPT-on-rails) | Barley |
|---|---|---|
| Knows what's actually on tap | No — invents beers | Yes — live menu only |
| Speaks craft taxonomy (IBU, dry hop, hop bill) | Surface-level | Native |
| Personalizes from real POS data | No | Yes — Square-driven |
| Updates in real time as a customer orders | No | Yes — within seconds |
| Sends segmented release alerts | No | Yes — taste-profile based |
| Privacy guarantees per brewery | Unclear | Per-customer opt-in, scoped to brewery |
| Grounded — cannot invent specs or prices | No — common | Yes — menu-grounded |
How Barley learns
Four data sources combine into one taste graph per customer. The graph powers every recommendation, alert, and dashboard insight.
- 1
Square POS — the foundation
Every transaction adds signal: which beer, what time, what tab size, what location. The model learns repeat behavior, time-of-day patterns, and product affinities — all from data you already have.
- 2
Menu metadata — the taxonomy
When a beer hits your menu (in Square or in Barley), you describe it in craft-beer terms: style, ABV, IBU, hops, malt, ferment style. Barley uses these features for cold-start recommendations.
- 3
Customer interactions — the chat layer
When a customer chats with Barley (web or in-taproom), every preference they share — 'I'm not into sours', 'I want something hoppy under 7%' — becomes part of their taste graph.
- 4
Aggregate, anonymized patterns
At the model level, Barley learns universal patterns ('people who like Hazy IPAs often like New England Pales'). Per-brewery and per-customer data is never shared cross-brewery; only the abstract patterns travel.
The craft-beer taxonomy Barley speaks
A non-exhaustive look at the dimensions Barley reasons over when it makes a recommendation. None of this is hand-coded prompt engineering — it's structured menu data that the model uses as features.
Style
- Hazy IPA
- West Coast IPA
- Imperial / Double IPA
- Pilsner
- Lager
- Stout
- Imperial Stout
- Saison
- Sour
- Lambic
- Porter
- Wheat / Hefeweizen
Hop profile
- Citra
- Mosaic
- Galaxy
- Simcoe
- Centennial
- Cascade
- Amarillo
- Nelson Sauvin
- Idaho 7
- Sabro
Strength & character
- ABV bands (sub-5%, 5–7%, 7–9%, 9%+)
- IBU range
- Dry-hopped vs. not
- Barrel-aged
- Fruited
- Brett character
- Coffee / vanilla / chocolate adjuncts
- Sessionable vs. sipper
Where customers meet Barley
One AI, three surfaces. Each surface is the right channel for a different moment.
In-taproom QR
A code on the menu opens a chat with Barley — instant recommendations based on the customer's prior visits. No app install.
On the brewery's website
An embedded chat widget that recommends beers, captures loyalty signups, and points first-time visitors at the right release.
SMS & email
When a new release matches a customer's taste profile, Barley pings them on the channel they opted into. Segmented, not blasted.
In production at The Keg Stand
“Barley has completely changed how we connect with our customers. Our repeat visits are up 34% and our community is stronger than ever.”
Frequently asked questions
- Is Barley powered by ChatGPT?
- Barley uses large language models under the hood — but the product is not a thin wrapper around ChatGPT. The recommendation engine, the taste-profile model, the brewery-specific taxonomy, and the data layer that connects to Square POS are all proprietary. The LLM is one component, not the whole product. Generic chatbots don't know that Citra is dank or that 9% imperials are usually winter releases; Barley does.
- Does Barley learn across breweries, or per-brewery?
- Per customer, per brewery. Each brewery sees only their own customer data, and each customer's taste profile is scoped to their interactions with that brewery's products. We do learn aggregate, anonymized signals at the model level (e.g. 'people who like Hazy IPAs also tend to like New England Pales'), but we never expose one brewery's data to another.
- What happens when you tap a brand-new beer Barley has never seen?
- When you add a new beer in Square (or in Barley directly), you describe it in the brewery's taxonomy: style, ABV, IBU range, hop bill, malt bill, special notes. Barley uses those features to recommend it to customers whose taste profile suggests they'll love it — even on day one, before anyone has ordered it. As real orders come in, the model sharpens.
- Can Barley hallucinate or make up beers that don't exist?
- Recommendations are always grounded in your actual menu — Barley can only suggest beers that are currently pouring (or available in your bottle shop). It cannot invent beers. For descriptive copy in the chat experience, Barley draws from the taxonomy you provide and from the brewery's own descriptions; it doesn't fabricate awards, prices, or technical specs.
- Does Barley work in real time?
- Yes. Every Square transaction updates the relevant customer's taste profile within seconds. New menu items propagate immediately. Release alerts fire on a schedule you control (immediately, batched, or scheduled), but the underlying segmentation is always live.
- Can customers turn Barley off?
- Yes — opt-in is explicit and one-click reversible. A customer who opts out keeps their loyalty rewards but stops appearing in segmented messaging campaigns and stops contributing to the taste model. Privacy isn't a settings panel feature; it's the default.
- Where does Barley show up — is it just a chat widget?
- Three surfaces: the AI bartender chat (embedded on your website or accessed via QR in the taproom), automated SMS / email release alerts segmented by taste profile, and the brewer-facing dashboard where you see what's working. Customers most often interact through the chat; brewers most often interact through the dashboard. Both are part of the same product.
See Barley reason about your beer list.
30-minute demo. We'll plug in a sandbox Square account, walk through the taste graph, and show you what Barley would recommend to a real customer profile from your brewery.
Book a Demo