Skip to content
Spice · Hospitality & Retail

An intelligence layer over your operation.

No new till. No second dashboard. A system that joins your data from every corner of the operation into one model, and returns not metrics but decisions.

StockStaffOpening hoursOne core, many operation types
Café & Restaurant
Bakery
Hotel
Canteen
Retail
Multi-site
Starting point

You have enough metrics.

Most operations do not suffer from missing numbers. The till and the BWA supply plenty. What is missing is the step after that.

"While reading, I kept asking myself which concrete decisions I would actually make differently because of these additional reports."
Operations manager in hospitality, from an advisory conversation

That is the right question, and it is the benchmark for this product. A metric describes a state. A decision needs three more things: an expectation of what comes next. An estimate of what a measure achieves. And a ranking, because nobody implements twelve recommendations at once.

The model

A digital model of your operation.

At the core sits not a database but a knowledge graph: the objects your operation consists of, and the links between them. Those links are the product, not the individual table.

Objects

Site, item, ingredient, supplier, shift: the things your operation consists of.

Links

An item consists of ingredients. Sales consume stock. Shifts cost hours.

Signals

Weather, events and bookings act on all of it, together and not separately.

Release

Every change is logged, reversible, and waits for you.

Only because the model knows a cappuccino contains milk from a particular dairy with three days' lead time can a demand forecast become an order proposal. Till, inventory, rostering and bookings stay where they are. The layer sits above them, not beside them.

Computation path

One decision, worked through.

Rather than claiming the system computes, here is the whole path from raw data to measure. Every step names its source.

Till export
Sales, last 8 weeks

Croissant, Saturday, per part of day, read from your export and never estimated.

Series
Weather service
Saturday forecast

Rain in the morning. The model knows the relationship from your own history.

Signal
Calendar
Local event

A date that has already fallen on a Saturday twice in your history.

Signal
Forecast
Expected demand

As a range, never as a single number. Order quantity and staffing derive from it.

P10–P90
Recipe
Converted to ingredients

Through the stored recipe to flour, butter, yeast, and from there to pack sizes.

Quantity
Stock + supplier
Order proposal

Stock netted off, supplier lead time accounted for, separated per supplier.

Proposal
Result

"Order one pack size more flour from the mill supplier. Reason: expected higher Saturday-morning demand, two-day lead time, current stock otherwise runs out Friday evening."

That is the difference from a report. A report tells you cost of goods has risen. Here there is a named measure, on a named item, with a named supplier, and every step behind it can be read back.

Computed by

Code. Every value above comes from a verified computation path.

Phrased by

The AI layer. It retrieves the result and phrases it. It does not recompute it.

Schematic illustration of the computation path. Item, quantities and supplier are placeholders for explanation. They do not come from a real operation. In use, every value is formed from that site's own data. Forecasts are issued as ranges; where the data is insufficient, the statement is withheld rather than estimated.

Boundaries

Where the boundaries are drawn deliberately.

Most vendors treat law and co-determination as a side issue. Here they are part of the product, and the reason the system survives a works council.

Till

Only exports are read. Nothing is written back, no signed transaction data is altered. Your KassenSichV obligations remain untouched.

Tax law

Business analysis, not tax advice. Your Steuerbüro stays responsible, and gets better numbers.

Staffing

Team level is the default. Named planning stays technically locked until co-determination is settled. No absence prediction, not even on request.

Tenancy

The separation of your data is enforced in the database, not programmed in the application, and backed by an automated test.

AI processing

EU endpoints only for employee and guest data, with no retention. No training on your data.

Working time

Rest periods and maximum hours are hard constraints. Every plan is re-checked after computation.

Getting started

How an operation begins.

Check the data

After the first call and an NDA, you send a till export. You get an honest assessment, including when it is "not yet".

A few days
Record the model

Recipes, purchase prices, fixed costs, staffing structure. The only step that genuinely needs your involvement.

One to two weeks
Backtest

The forecast is run backwards against your actual sales before you have to believe anything.

Straight after
Run alongside

You plan as you do today and get the proposal next to it. You switch only once it beats your experience.

Four to eight weeks
How you should measure us

Cost-of-goods ratio. Write-offs and spoilage. Staffing ratio.

Three figures from your own accounts. If they have not moved after a quarter, the system has not earned its price. We deliberately name no percentage in advance: every operation starts from a different level, and a number we cannot evidence against your data would be an assertion.

FAQ

Frequently asked.

Start with a conversation.

Twenty minutes to establish whether this is worth it for your operation. Only once that is settled do we talk about data. You send us nothing before you know who we are and what happens to it.