Kohiaido: private analytics for public places

Foot traffic, orders, queues and table occupancy for a café — measured from sound and the till, with no cameras, no stored audio, and no record that could identify a single customer.

Kohiaido gives the owner of a café or a shop the kind of analytics a large chain takes for granted — how busy the room is through the day, what sells and what it sells alongside, how long the queue runs, which tables turn — without watching anyone. It listens rather than looks: one microphone at the counter, and the point-of-sale. Everything that could name a person is discarded before it leaves the building.

The interior of No More Mondays, a coffee shop, with its counter and seating. — Kohiaido runs at No More Mondays, the coffee shop where it was first deployed.

What the owner opens is a dashboard; what the owner tends to read first, most mornings, is the advice sitting on top of it.

Fig. 1The advisors. Five specialists — operations, finance, product, facilities and demand — each watch one part of the business, flag what needs attention on a three-tier scale, and can be questioned in plain language. Every reply quotes the same figures shown on the tiles behind it; here the operations advisor hands the conversation to finance when the question turns to cost.

What it measures, and what it never keeps

The sensor is sound, not vision. A microphone at the counter records the room in ten-second fragments and passes each to a small server on the premises, where light models pull the fragment apart — speech from the hiss of the espresso machine from the music underneath — and count what matters: how many people, how much movement, the shape of a queue.

Speech is the hazardous part. Where a voice is found it is stripped of anything personal and reduced to something coarse and garbled; the fragment is discarded inside its ten-second window and never written to disk. The till supplies the other half — what was ordered, and when — as timestamps with no card and no name attached. The two streams are fused into figures blunt enough that no individual survives in them, and only those figures travel to the cloud. Three properties hold by design rather than by policy: the system does not identify people, does not store audio, and keeps no private data.

The dashboard

The opening view is a day at a glance: revenue against target, customers, average check, occupancy, the queue, and a reading of ambient noise that stands in for how full the room feels. A single control recomputes every figure for the day, the week, the month or the year.

Fig. 2The shop at a glance. Eight headline figures with their trend, the day’s footfall against a typical curve, the order mix, and the takings by part of day. Switching the period from today to the week or the month re-reads every tile.

The floor, in real time

A second view concerns the next hour rather than the last quarter: how many are in the queue and the wait that implies, how many seats are taken, how quickly tables turn, and where staff cover sits against demand. The floor plan colours each table by how heavily it is used, and the ambient-noise meter reports the felt busyness of the room without recording a word of it.

Fig. 3Operations. The queue drawn as waiting figures with a projected wait, occupancy as a dial, a heat-map of the tables, footfall by hour, and staffing set against demand — the view an owner glances at during a rush.

What customers do

Below the live view sit the order-level patterns: the split of drinks, food and retail; the items rising and falling week on week; the things habitually bought together; how long people linger; and the busiest hours of the week. None of it rests on knowing who anyone is — only on what was ordered and how the room moved.

Fig. 4Ordering behaviour. The composition and temperature of the order mix, a ranked table of movers, the pairings that recur, the distribution of dwell time, and an hour-by-weekday heat-map of demand.

The longer view

The final view steps back to the year: revenue by month, this year against last, the pull of weather on footfall, the swing between hot and iced drinks across the seasons, the mark left by public holidays, and a short forecast of the week ahead. It is the register in which a small business plans — when to prepare for a rush, when to let a quiet afternoon be quiet.

Fig. 5Season and weather. Twelve months of revenue with a year-on-year comparison, footfall against temperature, the hot-and-cold drink balance across the year, the effect of public holidays, and a forecast for the days ahead.