What Iriska AI does
Seven questions a restaurant owner asks — and what the product can answer
Every block below works the same way: the data that already exists about your place → what it means for your business → a recommendation or a scenario → the next step → a check. Every number you see carries its kind, its confidence, its date and its source. And every block says, beside its answers, what it cannot conclude — that column is not a footnote, it is the product's argument.
Nothing to upload to begin: the product starts from what is public — your website, the map, the forecast, the calendar — and every finding names the source it stands on. What you tell it is stored as your own estimate, never overwritten: a figure you restate is a new row beside the old one.
Business performance
“how is my place doing and what should I fix first”
- On public data alone
- What the public sees of your place: whether your website and your hours are findable, how your hours compare with the neighbours' on the map, and the demand context of the next two weeks — the forecast and the public holidays.
- When you add a few figures
- Your revenue range, average ticket, covers per day, seats, team and opening pattern give covers per seat and per opening hour and a purchases share of revenue, each with its formula written out; a figure that does not add up is put to you as a question, not a verdict.
- When you add documents
- A document you upload — a PDF, a spreadsheet, a photo — is read into text and kept with its page cited. Reading monthly sales and purchase summaries into a plan-versus-actual is not built yet; the product says so rather than guessing.
- What it cannot conclude
- Your actual margin or profit: an operating result appears only when revenue and the four cost lines are stated as single figures, never from a range. Exact demand. Anything about a competitor's revenue.
Costs
“where is my money going and what changed”
- On public data alone
- The consumer food price index of your country — Eurostat, monthly, food and seven of its classes — with its twelve-month change, labelled as a benchmark, the month it is for named (never "current"), the population it describes stated. It compares a population to a population; it says nothing about what you pay.
- When you add a few figures
- Your four cost lines as monthly figures or ranges and your main suppliers with their share of purchases give each line's share of revenue, all lines together against revenue (a question when they overshoot), the largest supplier's share and how many suppliers reach eighty per cent.
- When you add documents
- Invoices, delivery notes and price lists you choose to upload are read into text and kept. Comparing supplier prices on comparable items across documents is not built yet.
- What it cannot conclude
- Cost of goods sold from purchases alone — purchases are not cost of goods sold. Dish profitability. A saving without an alternative quote.
Promotions
“should I run an offer, which, and did it work”
- On public data alone
- A fourteen-day calendar of occasions composed from what is on file: public holidays with their scope (a regional one is not a national one), days unusually warm, wet or cold against the fortnight's own mean, produce coming into season, and the offers visible on your own website. Peers' offers are never stored.
- When you add a few figures
- Your goal, your weak days, your capacity and your current offers give three or four options, each with its assumptions shown before any number — and "run no promotion" is always one of them, with its reason. Your own discount and uplift replace the product's assumptions when you state them.
- When you add documents
- Sales by day and the cost of an item would let a result be measured and a special be costed. Neither is read from documents yet.
- What it cannot conclude
- A competitor's promotion effectiveness. A conversion rate. An exact incremental sale.
Menu / assortment
“what to change on the menu”
- On public data alone
- Your own public menu, read from your website's pages: sections, items and prices, each an inference cited to the line it was read from, which you can confirm with one click. The price band of your mains with how many items it counts. The seasonal calendar of your country for this month and the next.
- When you add a few figures
- Your bestsellers and slow movers as you name them and a target food-cost share give findings: a slow mover on produce nothing else uses, a bestseller on what is in season, a section priced away from the mains — and the target as a scenario with its assumptions first.
- When you add documents
- Item sales mix and recipe costs would give dish contribution. Neither is read from documents yet.
- What it cannot conclude
- Dish profitability without recipe costs and sales mix. Allergen facts without a declared source. No recipe, cost, ingredient list or allergen is ever invented.
Week and staff organisation
“what does next week look like and who works when”
- On public data alone
- The demand context of the next fourteen days: the forecast day by day, the public holidays, the season — every day a fact with its source, and a finding when a source did not answer.
- When you add a few figures
- Your figures give three what-ifs with their assumptions before the number — a purchases cut, a weekday promotion, extra opening hours — and the parameters are yours to set. A team roster and a draft week are not built yet.
- When you add documents
- Past covers by day and past schedules would calibrate coverage. Not read yet.
- What it cannot conclude
- The exact staffing need. Any employee decision — the product proposes, a person decides.
Competitors / market context
“who is around me and what changed”
- On public data alone
- The peers within a kilometre of your confirmed place, from OpenStreetMap — name, category, cuisine, hours, distance — stored with attribution and refreshed daily. Your own Google place, when you confirm it through Google with a key: its id is what is kept, its details are shown live and never stored. A neighbour's Google details are not read at all.
- When you add a few figures
- When your place is matched on the map, the First Look compares your hours with the neighbours'. Naming which peers matter to you and your own positioning is not built yet.
- When you add documents
- Your own sales trends read against the context. Not read yet.
- What it cannot conclude
- A competitor's revenue, margins or terms. Review trends from Google — not storable.
Supplier search and communication
“find me a supplier of X and handle the conversation”
- On public data alone
- Finding supplier candidates from a register, a catalogue or the web is not built: no directory, no search, no candidate is named.
- When you add a few figures
- Your need — product, quality, volume, delivery area — in your own words, the suppliers you already buy from read from the Costs block, and a checklist of what to ask any supplier, composed from your own figures: reach, minimum order, cold chain, shelf life, the season of the product, each line naming the fact it came from.
- When you add documents
- Reading a supplier's reply or price list with provenance, preparing a draft and sending it with your consent are not built: no message is written or sent on your behalf.
- What it cannot conclude
- Availability or a price without a quote. Any commitment on your behalf. And today: finding candidates and handling the correspondence.
Supplier search and communication
The half that nothing can invalidate is built: the need, your current suppliers and what to ask. Finding candidates and handling the correspondence are not — no directory, no draft, no message sent on your behalf.
Free at the base level, no card
Iriska AI's base level is free, with no end date and no card. Separate features may become paid later; prices are not published, and this page will not name a number it does not have.
The assistant inside the product is an AI and says so wherever it speaks (EU AI Act, art. 50). The findings, the digest and the review are composed by code from the data on file — not by a model.