A cashier,
not a register.

Everything fast growth needs is in place. What it needs now is capital.

The counter is transcription. Nobody automated it.

A cashier's core job is to hear a sentence and type it into a register. It stayed manual because until about eighteen months ago voice AI was too slow, too expensive and too uncanny to put in front of a paying guest. That changed, and the counter is still staffed the old way.

One counter, line by line.

Model defaults: 75 orders a day, the middle of a working lunch counter (the owner's numbers use 60, the low end); 45 seconds of AI per order; lease tier; a $2,460 kit with three years of cover. Every line below is a slider on the model page.

What the owner gets

$120,000 a year of counter payroll not paid, plus about $52,980 of add-on sales, net of what we charge: a $172,980 swing on one counter. The add-on line assumes 60 orders a day, the low end of a working lunch counter, and $2.70 of add-ons per order, because the screen asks on every ticket and a tired person doesn't. The owner can check both halves against his own payroll and tickets.

What runs today, and what ships next.

The model never computes a price. It proposes what the guest wants; a deterministic state machine checks every line against the merchant's real menu and does all the arithmetic. An item that isn't on the menu can't reach the receipt.

The counter decides which rail the card takes.

Of a ~2.7% card-present fee, about 1.7–1.9% is interchange and network assessments that nobody negotiates. The contestable slice is the processor's 0.8–1.0%. Through our processor partners the owner pays 2.3% all-in, under Square’s 2.6% + 15¢, and we keep 0.4%.

The model assumes $52,500 a month of card volume per counter. That is the restaurant's volume through the counter, not only the orders the AI takes, so earning it means becoming the merchant's processor. Set to zero, the model still clears its bar.

Bottom-up, California first.

By hand to a hundred. Then by download.

The first fifty are installed by the founder. The next fifty, and every one after, are sold by commission-only reps: $100 at signing, then 10% of the counter’s collected revenue for ten months, paid only from money received. That is about $1,000 per counter, 10% of its first-year revenue.

Then growth becomes a download. The Series A builds self-install: the owner orders the terminal, stands it up and approves a menu the AI built from a photo. That takes installs from what a field force can carry (about 2,000 a month) to 5,000, and passes 125,000 counters in year five. It is how Square reached millions of sellers without knocking on their doors.

The idle screen is also a channel. A face guests already know on a hundred thousand idle screens is inventory we own, and we intend to bring recognised talent on equity rather than fees. None of it is in any figure here.

A founder who ships, and engineers who scale it.

Engineering: a development agreement with Fastshot and its three engineers, drafted and not signed yet, led by Dmitry Fatkhi: no fixed fee and no salary. 20% of gross margin from acceptance of a working build, capped at $1.5M, until the Series A, 2,000 paying counters or 24 months; then the prior twelve months' fee in cash, or three times that in shares, never more than 2% of the company. All work product is assigned to Walletta. Dima Sukharev leads engineering as CTO and builds the in-house team from the seed.

Counters from sales now. A Series A on live counters.

The plan, year by year

100 installs in month three, growing 12% a month to 5,000 once self-install ships, 1.5% monthly churn, ~$836 per counter with payments. 125,000 counters × ~$836 × 12 is ~$1.25B a year, and the aisle, the idle screen and lending are not in it. The downside keeps a field force alone at 2,000 a month. Without the payments layer revenue is about 25% lower. Illustrative arithmetic on stated assumptions, not a forecast — move any of it on the model page.

What could go wrong.