One Year of Shop With Me — What Worked, What Didn't, What's Next

A year ago Shop With Me was two stores, a tablet on a counter, and a thesis. The thesis was that independent retail — liquor, convenience, small markets — needed an AI concierge that worked for one-store operators, not for chains. Twelve months later we have a clearer picture of what that thesis got right, what it got wrong, and what year two has to fix.

This is the honest version. Not the deck.

Hardware: off-the-shelf was the right call, but barely

We made a deliberate choice early on to run Remi on commodity Android tablets instead of building or buying a custom kiosk enclosure. The argument was speed: we could ship a working product to a pilot store in a month, not a year. We could fix bugs by pushing software, not by RMA-ing devices.

That choice was right, but it was closer than I expected.

What worked: store owners understood a tablet. They didn't need a sales pitch about hardware. If a tablet broke, the owner could replace it from any electronics store and we'd reactivate it in five minutes via the device provisioning flow. Cost per kiosk stayed under $400 including the stand. We never had a stuck-in-customs SKU.

What didn't work: tablets in a 4 PM Friday rush get fingerprints, smudges, and the occasional spilled mixer. Battery life on the cheaper SKUs degraded inside six months. The audio on consumer tablets is bad enough that voice recognition in a noisy store sometimes failed at exactly the moment the customer needed it most. We patched around it with external speakers and microphone arrays in a couple of stores, but that's a hack.

The honest read: off-the-shelf got us to product-market signal. To go from signal to scale, year two needs a designed enclosure — same tablet inside, but a frame that handles glare, audio, and the inevitable beverage incident. We're working with a contract manufacturer on a $250 enclosure that does this without inflating COGS. We are not going to build our own silicon. That's a fight for someone with more capital and less to prove.

Pricing: we started wrong

The first pricing page had three tiers and a free trial. The trial converted at a rate I'm not going to brag about. The tiers confused everyone.

What we missed: independent retailers don't price-shop SaaS the way a SaaS buyer does. They want one number, they want it predictable, and they want it to come out of the same revenue Remi generates. The Starter / Pro / Enterprise model — which works fine for software sold to a software buyer — felt arbitrary to a liquor-store owner who already pays a flat monthly to his POS vendor and a flat monthly to his alarm company.

What we changed: simpler pricing, keyed to whether the store wants a single kiosk or a multi-kiosk setup, with usage transparency baked in. We dropped the free trial in favor of a 30-day money-back guarantee. The trial was attracting tire-kickers who didn't have the operational maturity to onboard a kiosk in 14 days; the guarantee attracts owners who already know they want it and want to be protected if it doesn't deliver.

What we still don't know: the right price for a multi-store operator with five-plus locations. We have one customer in that segment and we're learning from them in real time. Year two will have a real multi-store tier. Right now it's bespoke.

Customer acquisition: trade events worked, content didn't (yet)

I tracked every dollar of acquisition spend this year. The numbers are blunt.

What worked:

• Trade shows for independent retail. One mid-size beverage trade show generated more pipeline than three months of paid search. Owners want to see a kiosk in person, ask whether it integrates with their POS, and watch another owner from a similar store nod.
• Direct outreach by zip code. I drove to twenty-six stores in San Diego and Riverside counties personally. Twelve agreed to a demo. Five became customers. That's not scalable, but it's a 19% close rate from cold, and it taught me what objections actually sound like.
• Existing customers. Two of our current stores came from a referral by an early adopter. That ratio matters more than any other number on the dashboard.

What didn't:

• Paid search for "AI retail kiosk" and adjacent terms. Search intent for those terms is overwhelmingly enterprise. We were paying to compete with vendors selling to Walmart. Wrong audience, wrong CPC.
• Generic content marketing. We published. People read. Nobody bought. Independent retailers don't typically read SaaS blogs at 9 PM the way a CTO does. The content that did work was concrete — a solution page for liquor stores outranked our marketing essays for every search term that actually closed.
• LinkedIn ads. Mostly noise. Mostly other vendors clicking on us.

What we're committing to in year two: trade-show presence in three regions, a referral program with real economics, and content that's specific to a vertical or a problem instead of generic AI-retail thinking.

Product: where Remi is good, where she isn't yet

Remi is genuinely good at what we built her for: an English-or-Spanish voice conversation about the products on a specific store's shelf, with the store owner's persona and rules layered on top. The recommendation quality on liquor and wine — where the catalog is finite and the language is rich — is the part of the product I'm proudest of. Customers ask "what's a good cab under 30 bucks for someone who likes Caymus" and Remi answers from that store's actual inventory, with reasoning. That works.

Where Remi isn't yet:

• Long tail SKUs in convenience. Convenience inventory is messier than liquor — same product, six SKUs, three vendors, frequent re-skinning. The embedding layer needs better dedup. We're rebuilding it for year two.
• Multi-turn memory across sessions. Today Remi remembers a conversation but not the customer over weeks. The kiosk-shopper identity layer (Wallet, NFC) is wired but underused. Once mobile is in the App Store, this gets fixed.
• Surveillance integration. We have a V2 service stub. It is not running. I'm not in a hurry to ship it without privacy thinking that matches our great-grandfather test. When it ships, it ships opt-in.
• Reporting depth. Store owners want to know what Remi recommended, what closed, and what got rejected. The analytics rollup gives them daily totals; they want SKU-level attribution. That's a year-two build.

People: I didn't hire fast enough, and that was right

I had advisors push me to hire by month four. I didn't. We're still very small. I now think both views were partially right.

The hire I did make — an engineering contractor who knows infrastructure cold — paid back inside two months. The hires I didn't make — sales rep, marketing lead — would have been wrong at the time. We didn't have product-market fit established firmly enough that a salesperson could repeat what was working without me in the room. Hiring sales before that point burns money and burns the rep.

What I learned: the rule "hire when the founder becomes the bottleneck" is correct, but founder-as-bottleneck has a delayed signal. By the time you feel it, you're already three months late. Year two I plan to hire one operator-type person who can run pilot onboarding without me, and one sales-led person once we have a rep-able playbook from the trade-show channel.

What we're committing to in year two

A short list, on purpose. We will:

1. Ship a designed kiosk enclosure with better audio and screen treatment, at the same per-unit cost.
2. Launch the consumer mobile companion app on iOS and Android (already built, sitting in submission).
3. Deepen POS integrations beyond the current two — the goal is four mainstream POS systems with bidirectional inventory sync.
4. Open the multi-store tier with real multi-store admin features, not bespoke configuration.
5. Hold the line on privacy. Surveillance and facial auth stay opt-in. We will not ship dark-pattern features to chase a chain customer.

Things we are explicitly not doing in year two: enterprise sales motion, white-label, hardware vending. Each of those is a fine business; none of them is the business we're building.

If you run an indie store and want to see what year two looks like before we ship it, book a demo or read about us. Year one taught me to listen more than I talk to potential customers. That part isn't changing.

Frequently asked

How many stores are live on Shop With Me after one year? We're being deliberately small. The exact count is in the single digits, by design — we wanted depth of feedback per store, not breadth. Year two we open the gate.

Did you raise money in year one? No outside capital yet. Bootstrapped from founder savings plus revenue. Year-two roadmap doesn't require a raise, but a small seed would accelerate the hardware and POS-integration work.

What's the single biggest thing you'd do differently? Drop the three-tier pricing on day one. We knew it was wrong by month two and kept it for another two months out of inertia. Don't keep things that aren't working.

Are you profitable? Operating profitable on the unit economics of an active store; not yet covering full overhead at current store count. The path to overhead breakeven is in the year-two roadmap and it's a function of new-store adds, not of price increases.

What worries you most about year two? Distribution. The product is in good shape. The thing I haven't yet cracked is repeatable acquisition outside trade shows and personal driving. That's the year-two job.