Drafted at 02:00.
Sent by you at 07:15.
Every night, ARES reads your till's sales, forecasts tomorrow with machine learning, and drafts your supplier order with a reason on every line. You review it, adjust it, and send it yourself.
- Flour 002 x 25 kg€43.00
Why: Weekend forecast sits in the high band. Two day lead time, rounded up to whole bags.
- Whole milk36 L€33.10
Why: Your kitchen uses 8% more than the recipes say, learned from your counts. Shelf life caps this at five days.
- Fior di latte3 x 2 kg€41.70
Why: Tuesday pizza spike. Cases of 2 kg, and without this line the basket misses the €100 order minimum.
Kitchens with a focused menu: burger bars, pizzerias, brunch cafés, pancake houses, snack bars. Running a small chain? We start with one venue and prove it there.
One service ends.
Another shift clocks in.
The till closes.
Your POS now holds everything ARES needs for tonight: what sold, when, and how much. Nobody stays late counting stock. Nobody uploads a spreadsheet. The kitchen goes dark, and the night shift takes over.
Four jobs run, in order.
Each labeled for what it really is.
Two of these are machine learning. One is plain optimization math, and we say so. The last one keeps the other three honest. The clock times are illustrative: the nightly run is scheduled around your closing hours, on the schedule you choose.
The forecast
Demand for every menu item, as a range rather than a single number, trained on your sales only. Weekday rhythm, weather, holidays, recent trend, promotions you've planned. You choose how careful each ingredient should be: generous on the mozzarella, lean on the truffle oil.
The reality check
Your recipes say one thing. Your kitchen does another. ARES compares theoretical usage against your real stock counts and learns the gap per ingredient, the heavy pours, the trim, the staff meals, so orders fit the kitchen you actually run, not the one on paper. The correction sharpens every time you count.
The basket
An optimizer compares your suppliers' offers and packs the cheapest compliant basket: unit prices and volume tiers, delivery fees, order minimums, case sizes, cutoff times, shelf life. It writes down every constraint it respected, line by line, so you can see why 14 kg and not 10.
The self-audit
ARES also checks itself: would ordering like last week have beaten the forecast? It backtests against four naive baselines every night and keeps the score. When it loses, you can see it lose. We think that beats a black box that always claims it won.
06:30. The draft is waiting
with the coffee.
Read it.
Every line carries its reason in plain language: the forecast band, the shelf life cap, the case rounding, the order minimum it had to clear. No black box. You can argue with it, ingredient by ingredient.
Change it.
Think 10 kg is enough where it says 14? Type 10. Your judgement wins, every time, and the adjustment is recorded next to the original suggestion. The system proposes. It never overrules.
Send it. Yourself.
To your own supplier contacts, from your own phone or email, in your own words if you like. Then the day starts, and the ordering is already behind you.
The kitchen gets its prep sheet.
Alongside the order: recommended batch quantities for the day, portioned from the same overnight forecast by plain rules and delivered on the schedule you choose. Your team preps for the day you're actually about to have, not the one you fear.
ARES never contacts your suppliers. Every order is reviewed, adjusted, and sent by your manager. We think that's how it should be: the system does the math and shows its work; your people keep the relationships and the final word.
We only quote numbers
we measured.
We follow one rule most software companies don't: we only quote numbers we measured. Before your venue goes live we capture its baseline: waste, emergency purchases, stockouts, ordering time. The pilot measures the change against that baseline. Until then, we show you the workflow and tell you exactly which parts are machine learning and which are plain rules.
That's why these tiles say what they say. A dashboard that refuses to bluff is worth more than one that always has an answer.
Needs your week zero baseline first. We capture it before go live, then the pilot measures the change.
Starts reporting after 60 to 90 days of your sales history. Until then, your dashboard says so, in writing.
- Food purchased that becomes waste4 to 10%USDA ERS · NRDC food waste study
- Revenue eaten by food costs28 to 35%National Restaurant Association, 2025
- Average net margin left over3 to 5%Toast restaurant report, 2025
EU hosted
Your data lives in EU data centers and does not leave the Union.
Encrypted
TLS for everything in transit, AES-256 for everything at rest.
Scoped access
Every query is scoped to your organization and location, with owner, manager, and staff roles.
Yours to take
Export everything at any time. Leave, and we delete your data on request.
Questions operators actually ask us.
Accuracy depends on how much history you have and how stable your operation is. We publish each forecast as a range rather than a single number, so you can choose how aggressive to be for each ingredient: conservative for perishables, leaner for shelf stable items. During onboarding we typically need 60 to 90 days of sales history to reach meaningful accuracy, and your dashboard tracks the accuracy openly from day one.
A little, and we are upfront about it. The system needs deliveries confirmed when they arrive, a stock count now and then, and five minutes of your attention on the draft each morning. What disappears is the weekly count-everything-and-phone-everyone marathon. We tell you exactly what the routine is before you sign, not after.
The simplest way to start is a sales export. You export your history, we load it, and forecasts start running within a day. For the long term we work with you on a direct POS connection. Tell us which POS you use and we will walk you through the options.
It stays yours. We never sell or share it, and we never use it to train forecasts for other clients. Each restaurant's forecasts are built only from its own data. You can export everything at any time, and we delete your data on request if you leave.
Honest answer: any forecast is wrong sometimes. That is why you get a range instead of a single number, and you set the safety buffer for each ingredient. The system also scores its own forecasts nightly against simple baselines and shows you the result, so you always know how much to trust it.
The first usable draft appears about a day after we have your menu, recipes, suppliers, and at least 60 days of sales history loaded. A realistic timeline from first call to a working nightly routine is one to two weeks, most of which is collecting and cleaning your existing data.
Pricing is per location, scaled to your menu size. We do not publish a number because the right price depends on your menu, recipes, suppliers, and number of locations. Book a call and you will have a straight answer early in the conversation, not on slide forty.
Tonight, the order
is still on you.
Twenty minutes on a call. We show you the workflow on demo data, wearing its demo badge, and give you a straight price early. You decide if it is worth a pilot.
Already working with us? Set up your restaurant yourself