AI demand forecasting
Demand forecasting uses past sales and the factors that move demand, such as promotions, prices, holidays and seasons, to predict how much of each product will sell in the coming period. Corrplus builds forecasting models on your data at product and warehouse level and delivers the result as a range into the ERP or sheet your purchasing and production planners already use.
Book a free callSymptoms
- Order quantities are set from last year's numbers and experience.
- Best sellers run out while slow movers sit in the warehouse.
- Promotions and holidays surprise you every time.
- Only one person can update the forecast sheet, and planning stops when they are away.
- Sales, purchasing and production plan the same period with different numbers.
What waiting costs
Every day out of stock is lost sales, and every unit of excess stock is cash and warehouse space tied up. When prices change often, both mistakes cost more.
Our approach
Data review
We combine your sales, stock, price and promotion history and report plainly whether it is enough to forecast from, and what is missing.
Model comparison
We start from a simple baseline and compare statistical and machine learning models on the same historical period.
Connect to planning
The chosen model's forecast, with lower and upper bounds, is written on a schedule to your ERP, planning sheet or a dashboard.
Monitor and retrain
Forecast error against actual sales is tracked; when drift grows, the model is updated and your team is told.
What to measure
- Forecast error by product
- Days out of stock
- Excess stock held
- Time spent on each planning cycle
Our own products
We use the tools and checklists from client projects on our own products too.
All products →Why do forecasts miss?
In most companies, demand forecasts come from the same period last year plus the planner’s experience. That works when things are stable. In Türkiye, the factors behind demand shift often: holidays that move every year, sale periods on marketplaces such as Trendyol and Hepsiburada, frequent price updates, and tourist seasons that stretch or shrink. European businesses face their own version with moving holidays and promotions.
There is also a hidden problem. On days you were out of stock, sales show zero, but demand was not zero. Any method that does not separate those days will under-forecast your best sellers. It is one of the first things we look for in the data review.
Which method is enough, and when?
| Method | What it accounts for | Enough when | Limit |
|---|---|---|---|
| Historical average (spreadsheet) | Past sales only | Few products, steady demand | Misses promotions and calendar effects |
| Statistical models (ETS, ARIMA, Prophet) | Trend and seasonality | Products with a regular history | Struggles with many outside factors |
| Machine learning (e.g. LightGBM) | Promotions, price, calendar, product attributes | Many products, rich data | Needs more data and upkeep |
In every project we compare all three on the same historical period. Sometimes a simple statistical model is enough and a more complex system is not worth it. We say so when that happens.
How we start
The first step is an AI consulting engagement: we review your data and report which product group would gain most from better forecasts and whether the data supports it. If it does, the first model covers one product group or one warehouse.
The model runs close to your data, on your own servers or the cloud infrastructure you prefer. Writing the forecast to your ERP, spreadsheet or dashboard on a schedule, and alerting the team when it drifts, is built with workflow automation. Sales data rarely contains personal data; if customer-level data is used, we assess it under GDPR and Türkiye’s data protection law, KVKK, during design.
AI demand forecasting: where to start in your business
Tell us where you are in a free first call; we will say plainly where it makes sense to start.
Related services and use cases
Frequently asked questions
How much history do we need?
To see seasonality you need sales by product and by day or week covering at least a few seasons. The exact threshold depends on your products and how regular demand is; the data review says so plainly.
How is this different from our spreadsheet forecast?
Spreadsheets usually average the past. A model learns the separate effects of promotions, price, holidays and weekday, runs across thousands of products at once and gives a likely range instead of a single number.
Can it handle holidays that move every year?
Yes. Moving holidays such as Ramadan and Eid in Türkiye or Easter in Europe, school breaks and marketplace sale days are given to the model as calendar features, so it compares demand with the right period, not the same calendar date.
Can you forecast new products?
Forecasts without history are weak. We use approaches that borrow from the early weeks of similar products, but those forecasts come with wider ranges and the planner's judgement matters more.
How will we know the forecast is right?
Before going live we forecast a past period as if it were unknown and compare it with what actually sold. We run the same test on your current method, so the decision rests on a direct comparison.
Let’s talk about your transformation
Tell us which work repeats every week. We will tell you honestly whether it is worth automating.
How we start
- A free first callWe pin down the problem and the goal together.
- A 2-week discovery sprintWe look at the data, the process and the tools, and decide where to start.
- A written proposalScope, architecture and timeline, in writing.