Live model · updated September 2026

Don't take my numbers. Move them.

Every figure on the investor page and in the memorandum comes from this page. The assumptions are inputs, not claims — change them and watch the business change. If you can break it, I would rather find out from you now than from a board later.

01 · One counter

The whole company is this table, multiplied.

If a single counter works, the rest is an execution problem. Set the assumptions on the left.

Orders per day 50 A working lunch counter runs 60–90. We model 50 deliberately — every figure on this page is computed below the volume a real counter does. Check this first. Minutes per order 1.5 Voice order, modifiers, upsell, payment. Our AI cost / minute $0.15 Speech, model, voice, avatar. Computed at full list cost even though Tavus currently carries us on their $210K tier at effectively nothing — the concession is headroom, not something the plan leans on. We sell every minute at $0.25 and keep the difference — that retail rate is ours and it does not move, so this row is the whole margin story. Hardware tier The 13" Pro carries a four-microphone array, and for a product that lives or dies on hearing an order across a noisy counter that is an engineering decision, not a taste one. Merchant pays $1,500 to own the kit and the install outright, or $200 + $99/mo to lease. Amortised over 24 months on the lease tier only. Merchant tier Card volume / month $52,500 What the counter already routes today, through someone else. Our take on card volume 0.40% 0% today. 0.3–0.5% under a sponsor platform. ~1% at our own registration, past $500M routed.

Line	Per month	Per year

02 · Payback

What one counter costs to win, and how fast it returns.

Sales are commission-only and paid out of money the merchant has already sent us, so the acquisition cost is real cash but it is never paid before revenue.

Commission per activation $950 Paid to whoever closes the merchant, out of money that merchant already sent us. Install fee, per counter $200 Standing the terminal up and handing it over — often the same rep who closed it, paid for the trip. On the purchase tier the merchant's $1,500 covers this along with the kit, so it costs us nothing; on the lease tier we carry it. Either way it disappears when self-install ships. Marketing per activation $0 Zero, and not by accident: brand partnerships are structured on equity, not fees. Move this to see what paid acquisition would do to the ratio below. Monthly churn 2.5% Independents close at 15–25% a year on their own. 2.5% is closures plus room for everything else. Roughly 80% of counters have no self-checkout, so we capture demand rather than win switches — churn is mostly not ours to cause.

03 · The cohort build

A hundred thousand counters, month by month.

Counters are added, then they churn. Five years, so the $20B design floor of 100,000 counters shows up as a date rather than an adjective. The same engine at a slower rate is the $250K case.

Case Changing this resets the operating cost and install rate to what that amount actually buys. Both stay editable. Installs in month 3 100 Months 1 and 2 run at a third and two thirds of this while the product is finished and the first merchants are installed by hand. Growth compounds from month 3. Install capacity ceiling 2,000 / mo Nothing compounds forever, and a model that says otherwise is lying to you. A field force of a given size can only stand up so many counters a month. Past this ceiling growth stops being exponential and becomes linear until the self-install app raises the ceiling. Monthly growth in install rate 12% Field reps recruit reps. Growth compounds until the self-install app removes us from the loop. Monthly operating cost $100,000 Engineering, founder floor, legal, compute. Not sales — commission is separate. On $100K nobody draws $100K a month: contractors, founder takes nothing. The plan slows because that is what the money buys, not because we are careless.

04 · The cost curve

This number has only ever gone down. Here is what that is worth.

How to read this: we buy a minute of AI at one price and sell it at $0.25. Today we buy at $0.15. That buying price is set by OpenAI, Deepgram and Tavus, not by us — and every year so far it has fallen for the same capability. Each row is a price we might be buying at, and what one counter is then worth to us. We hold the retail rate at $0.25 and keep the difference, so the business gets better while we sleep. The rows below today are the protection: they are what happens if the trend reverses.

05 · Hardware, stated plainly

The box pays for itself, and nothing more.

$1,500 buys the kit and the install. At our sourcing that is roughly $1,300 of hardware and $200 of labour — it lands on the line, deliberately. We are not a box reseller and we are not selling below cost either.

Every figure on this page is computed live in your browser from the inputs above; nothing is hard-coded except the hardware bill of materials, which is a procurement estimate and is labelled as such. Layers marked plan are not live today and are not funded by the current round. Cohort projections are illustrative arithmetic under stated assumptions, not forecasts, and depend on execution and on raising every round between here and the outcome. Market and comparable-company figures are cited on the memorandum. Nothing here is an offer to sell securities.