The Great-Grandfather Test — How a San Diego Liquor Store Built an AI Concierge

I run Shop With Me from San Diego, and before any of this was software, my family ran a liquor store. The store taught me almost everything I know about retail — and the test I now apply to every feature we ship is simple. I call it the great-grandfather test: would my great-grandfather have understood it, and would he have trusted it standing behind his own counter?

That sounds like a marketing line. It isn't. It's the gate every product decision has to pass before it ships into a real store.

What the store actually taught me

Independent liquor stores are not small versions of big-box stores. They are a different business. The customer who walks in at 4:45 on a Friday is not browsing — they're picking up something for a dinner, a fight, a celebration, or a regular Friday. They have one or two minutes of patience and no interest in being upsold. They want a recommendation from someone who knows the shelf.

When my family ran the store, that "someone" was a person. Sometimes my dad. Sometimes a clerk who'd worked the register for fifteen years and could tell you which Cabernet under twenty bucks was actually drinking well that week. The recommendation engine was a human with shelf knowledge, taste, and a relationship.

The thing nobody outside indie retail seems to understand: that human recommendation is the whole product. The store is not a vending machine. It's a curated shelf plus a guide. The guide is what people pay the markup for.

So when I started thinking about an AI kiosk, the question was never "can we automate the clerk." The question was: can we extend the clerk so that on a Friday night when there are eight people in line and one register, the customer who walks past the bourbon shelf still gets a real answer to "what's this like?"

Why I named it Remi

We named the kiosk Remi because we wanted a name a customer would say out loud. Not a model number. Not "the AI." A name. Same reason a good restaurant has servers introduce themselves — names create trust.

Remi is animated, talks in English or Spanish, and lives on a tablet on the counter or by an endcap. The customer asks a question — "I'm bringing a bottle to a dinner where they're cooking salmon, what do I want under thirty bucks" — and Remi answers from that store's actual shelf, not a generic catalog.

That last part matters. Remi is not Google. Remi only knows the inventory of the store it's deployed in. If the store doesn't carry it, Remi won't recommend it. That constraint is on purpose.

The great-grandfather test, in practice

Here's how the test plays out in real product decisions.

Voice in, voice out. My great-grandfather wouldn't have typed on a keyboard. He'd have talked to a customer. So Remi listens and responds in voice by default. Touch is the fallback, not the primary mode. When we tested a text-first version, the over-50 customers in our pilot store walked past it. The voice version gets used.

No fake reviews. Big-box e-commerce has trained customers to expect star ratings on everything. We won't generate them. If a wine doesn't have an honest review from someone the store trusts, Remi will say "I don't have a tasting note on this one — let me tell you what's similar that I do know." My great-grandfather would not have told a customer something he didn't know. We won't either.

Recommendations explain themselves. Every Remi recommendation comes with a reason — "this is a smoother bourbon than the one you mentioned, and it's eight dollars less." If we can't articulate the reason, we don't make the recommendation. Black-box answers fail the test.

Cash and card both work. A surprising amount of indie retail still runs on cash. Remi has to function in a store where the POS is whatever the owner has run for fifteen years. We don't replace the register. We add a layer above it.

The store owner is in charge. Every prompt, every recommendation rule, every store-specific persona is editable in the dashboard. If a store owner wants Remi to never recommend anything over a certain price, that's one toggle. If a store wants Remi to always mention the local distillery first, that's a setting. The owner is still the owner.

Why this matters for AI in retail

The current pitch around AI in retail tends to come from people who have never run a store. The pitch is: replace the clerk, scale up, cut labor cost. That's the wrong frame. In an indie store, the clerk is the moat. You don't replace the moat — you reinforce it.

What an AI concierge actually does, when it's built right, is shift the clerk's time. Instead of the clerk being interrupted twelve times an hour by "do you have any of that one wine," the clerk does inventory, talks to the regulars, and handles the conversations that need a human. Remi handles the conversations that didn't need a human and were never getting one anyway because the clerk was at the register.

That's the value. Not replacement. Coverage.

What we won't build

A few things we've turned down because they fail the test:

• Personalized pricing. No surge, no dynamic pricing per customer. Same price on the shelf for everyone. My great-grandfather would have walked you out of the store for that.
• Dark patterns. Remi will not nag, will not create false urgency, will not push the highest-margin item. The recommendation algorithm optimizes for "the customer comes back," not "this transaction is bigger."
• Selling customer data. Conversations are a store asset, not ours. Aggregated trends help the store decide what to stock. We don't license the data to brands.

If you want to see how this actually plays in a store, we have a public demo and a writeup of who we are on the about page. Pricing is on the pricing page — flat per kiosk, no per-conversation fees, no usage games.

A note on family

My dad still asks me, every couple of weeks, whether the kiosk is really helping the stores or whether it's the kind of thing that ends up unplugged in a corner because the staff hated it. That's a fair question. The answer I want to be able to give him, in a year, is: the staff treats it like another teammate, and the customers — especially the older ones, the ones who would have been my great-grandfather's customers — don't think of it as a robot. They think of it as the friendly tablet that knows the shelf.

That's the bar. Everything else is implementation.

Frequently asked

Who is the founder of Shop With Me?

Mike Yadago, based in San Diego. My family operated an independent liquor store, which is the operational background that informs how Remi is built. I run the company day-to-day and personally talk to every pilot store.

What is the "great-grandfather test"?

A simple gating question I apply to every feature: would a customer who has never used technology like this still understand it and trust it on the first interaction? If the answer is no, we redesign or cut the feature.

Why focus on independent retail instead of big-box?

Independent stores are the businesses where a knowledgeable clerk is the differentiator, and that's exactly what an AI concierge can extend. Big-box already has data infrastructure and dedicated tech teams; indie stores have shelves, regulars, and not enough hours. Remi is built for the latter.

Is Remi trying to replace store employees?

No. Remi handles the recommendation conversations that staff often miss because they're at the register. It frees up clerks for the conversations and tasks that need a human. The framing of "AI employee" is something I push back on directly.

How can a store owner try Remi?

Book a demo on the demo page or contact us through the site. We do a short scoping call, then a setup that takes under an hour with the existing tablet at your counter.