The problem
Tiny Fish has built one of Switzerland's most thoughtful fast-casual sushi brands — fresh bio salmon at CHF 20, fed by a central kitchen in Altstetten that reaches Geneva in under three hours, a Swiss Economic Forum High Potential Label, 14 stores already open, four to five new openings per year. The hard part is no longer how to operate a store. The hard part is choosing which next stores to open.
Each store decision is a multi-year lease commitment in a market where rent variance between two ostensibly comparable locations can be 3x, and foot-traffic variance can be 10x. The wrong store doesn't bankrupt anyone — Tobias's small-footprint, secondary-market-liquid design is forgiving — but it locks up management attention, slows the cadence, and chips at the brand discipline that makes Tiny Fish credible. The right store, by contrast, becomes a self-funding asset that pays its own opening capex inside a year.
Today that decision is made the way every restaurant chain makes it: a broker walks Tobias through their book, Tobias visits a handful, intuition wins. That works at 14 stores. At 30, then 100 across a future Western-world franchise rollout, intuition stops scaling. Picking the next location should be a model, not a vibe.
What Tinystores is — at the principle level
Tinystores is a multi-criteria decision model. It takes any Swiss commercial-lease offering and returns a composite score from 0 to 100, plus an explanation of which dimensions drove that score up or down. The model is opinionated: a Tiny Fish point-of-sale is a grab-and-go funnel sustained by pedestrian density and lunchtime office overflow, not a destination restaurant. That principle shapes every weight.
The model evaluates a candidate location against several dimensions that map directly to Tiny Fish's unit economics: how many pedestrians actually walk past, who they are, what they can spend, how easy the location is to reach by public transit, how much office activity sustains the weekday lunch peak, what competition already exists, how visible the shopfront is, and how flexibly the lease is structured. Before any of that scoring runs, hard filters reject anything outside the operational envelope — the 3-hour Altstetten kitchen radius, the footprint size that fits the Tiny Fish format, and a minimum distance from any existing store to avoid cannibalization.
The data approach
The signal layer is built deliberately on a foundation of free, public, federal-grade Swiss data — combined with a small number of paid APIs we already operate. The result is a model that runs at a marginal cost in the low double-digit Swiss francs per month, with no broker fees, no Swisscom Mobility subscription, no CBRE quarterly reports. The expensive paid alternatives buy 70–80% of the same signal at 100–250x the cost — they make sense for retail giants ranking thousands of micro-sites, not for a thoughtfully-paced brand picking carefully.
Critically, every signal is auditable. The model doesn't make decisions; it makes recommendations the human eye can interrogate and override. When a candidate is rejected by a hard filter, the user sees which filter. When a candidate scores 78 instead of 85, the user sees which dimension is pulling it down. The system is designed as a tool for Tobias and Diego, not as an oracle.
How it validates
The honest validation method is not revenue (Tobias holds per-store revenue close, and rightly so) but agreement with judgement that has already been made. If the model scores existing Tiny Fish stores well — if Bleicherweg, Löwenstrasse, Hardstrasse, Bern Spitalgasse come up in the top quartile of the model's distribution — then the model has earned the right to be trusted on locations not yet chosen. If it disagrees with Tobias on those, the disagreement itself is information: either the model is wrong, or the model has caught a signal Tobias's intuition encodes implicitly but hasn't articulated.
The model also validates from below. Locations that are obviously poor candidates — industrial zones, vacation towns outside the 3-hour radius, residential suburbs with thin daytime population — must score poorly. They do.
The vision: Switzerland is the blueprint, the world is the canvas
Tiny Fish's roadmap is 30 stores in Switzerland by 2030, then one new country every two years through the 2030s via asset-light franchise partnerships. The location intelligence built here — the dimensions, the principles, the data architecture — transfers directly to those future markets. The same model that scores Lausanne in 2026 can score Munich in 2030, Paris in 2032, Manchester in 2034. The Swiss build is the proof; the international rollout is the leverage.
By 2040, if execution holds, Tiny Fish is a Western-world brand recognised in 30 countries for high-quality sushi at a fair price. By then, Tobias is around 50. That would be a proud moment to look back on.
What's gated, and why
The demo shows 17 of the 20 candidate locations the model evaluated. The top 3 — the ones we'd actually recommend Tobias and Diego visit next — are currently under internal audit and not yet published. Once each is hand-validated against Tiny Fish's operational filters and the underlying signals confirmed, they'll be revealed in gated form (available by request). The bottom 10 visible serve as proof the model rejects what it should; the middle 7 show the gradient; the top 3 are reserved for the conversation that will actually act on them.
Picking a location for a CHF 20 lunch is not a CHF 20 decision. It's a CHF 1m, 5-year, brand-shaping decision. It deserves to be modeled, not vibed.← Back to demo