UA Talk to a human
Purchasing

Demand forecasting and ready supplier orders

The system forecasts demand for every item and calculates how much to order, taking into account stock in transit, batch multiples and shelf life. The buyer makes the final decision.

Status: In production at a restaurant group with catering (daily supplier requests) and at a veterinary and pharmacy products brand (monthly orders to manufacturers).

The problem

  • The buyer calculates orders by hand in spreadsheets, and the logic lives in their head.
  • Some items are overstocked, and cash sits frozen in the warehouse. Others run out, and sales are lost.
  • An order takes months to arrive, so today's mistake only shows up a quarter later.
  • Seasonality, promotions and new items are estimated by eye.
How we solve it

Data and a metric first, then the model.

We build the model on your data and test it on a “blind” period before we promise anything.

  1. Brief

    :

    How you buy today, from whom, with what lead times, minimum batches and constraints.

  2. Data audit

    :

    The history of sales, stock and supplier orders from your accounting system. The key question is whether stockouts are visible. If they aren't, the model learns an empty warehouse instead of demand.

  3. Metric before we start

    :

    Together we agree how to measure accuracy: horizon, per item or per group, in units or in money. The baseline is how your buyer calculates today.

  4. Blind-period test

    :

    The model forecasts a period it hasn't seen, and we compare the result with actual sales before trusting it with money.

  5. An app for the buyer

    :

    A ready order, edits with a reason, a history of decisions.

  6. Controlled retraining

    :

    A new model version goes live only if it tests no worse than the previous one.

What you get

A ready order, with the reasons behind it.

The system calculates, the buyer decides. The model also learns from the buyer's edits and the reasons behind them.

  • A demand forecast for every item with an explanation: season, promotions, trend.
  • A ready supplier order that accounts for stock in transit, batch multiples, safety stock and shelf life.
  • A monthly purchasing budget built into the calculation as a limit.
  • A monthly forecast vs. actual report.
  • Buyer edits with a reason, which the system learns from.
Channels and integrations

Where we get the data and where the order goes.

  • livein production
  • donebuilt in another project
  • per projectfirst time in your project, API checked
  • 1C / BAS accounting: reading sales and stock: per project: connected via API or file exchange; checked against the documentation
  • Excel and Google Sheets: order export: per project: we've worked with Google Sheets in other projects
  • Telegram alerts: done: in our systems in production
  • Restaurant POS: daily sales from every location: live: in production at a restaurant group with catering
  • Forecasting model: live: in production at a restaurant group and at a veterinary and pharmacy products brand
Timeline and price

Timeline after the data audit. Price after the brief.

First we look at what your data holds, and only then do we promise anything.

Timeline and accuracy

After the data audit

Before the audit, we don't name a timeline or an accuracy target.

Timeline
After the data audit.
Price
After the brief.
Brief
Free.
Case studies and demo

What it looks like in practice.

Two case studies in production: daily supplier requests for a restaurant group with catering, and a monthly order to manufacturers for a veterinary and pharmacy products brand.

FAQ

Questions about forecasting.

What accuracy do you guarantee?

We set the target after testing on a blind period of your data. High accuracy is usually achievable on fast-moving items, but not on rare ones.

How much history do you need?

Quality matters more than the number of years: stockout flags, a promotions calendar, a dated history of supplier orders. We define the minimum during the audit.

Who decides on the order?

The buyer. The system gives a ready calculation with an explanation, and a person approves or edits it.

Does the model retrain on its own?

At the press of a button or on a schedule, but a new version goes live only if testing shows it is no worse.

Aren't sales to retail chains different from shopper demand?

Yes, shipments to retail chains and marketplaces are not the same as sales to end customers. We account for this in the data audit and in the model.

Next step

Write to us. A human replies.

Machine view: this page as markdownOpen index.md
Loading markdown…