How to forecast demand and stock in a small online store: simple formulas, seasonality, data for models and the situations where AI really makes sense.
About us
For demand and stock forecasting, a small online store usually needs nothing more than sales history, seasonal coefficients and simple formulas: average sales, lead time, safety stock. AI models become useful when the assortment has many items, seasonality is noticeable and a clean history has accumulated. Order in your records comes first; automation is the second step.
A forecast answers two practical questions: how much of a product will sell over a period, and when and how much to order. The second depends on lead time, minimum order quantity and available cash.
Sales show how much people bought. Demand shows how much they wanted to buy. When an item was out of stock, sales were zero, and based on such data a model will treat demand as low. Mark the days with no stock and account for them when calculating the average.
A forecast can be built by product, category or brand. For items with infrequent sales, a forecast per individual SKU is unreliable. In that case you forecast the category and distribute the result among the items according to their share.
A small store can manage with a few methods that can be kept in a spreadsheet.
| Method | Essence | When it fits |
|---|---|---|
| Average sales and safety stock | Average daily sales multiplied by the lead time, plus a buffer for fluctuations | Stable products with regular demand |
| Seasonal coefficients | Comparison with the same period last year | Products with pronounced seasonality |
| Promotions calendar | Separate tracking of mailings, sales and advertising that change demand | Stores with regular promotions |
| Groups by importance | Items with the highest revenue are checked more often and purchased more carefully | A wide assortment and limited time for record-keeping |
The reorder point equals average daily sales multiplied by the lead time in days, plus safety stock. When stock reaches this point, it is time to place an order. The less stable the demand and the longer the delivery, the larger the safety stock.
For example, a product sells 4 units a day on average, delivery takes 10 days and the safety stock is 12 units. The reorder point equals 4 times 10 plus 12, that is, 52 units. As soon as stock drops to this level, you need to order. This is a hypothetical example with rounded numbers; it only shows the logic. In a real store the numbers are checked against sales data and revised when the season, the supplier or a promotion changes. If deliveries are sometimes delayed, the safety stock is increased.
Seasonality comes in several kinds: yearly (seasons of the year, holidays), weekly (weekdays and weekends) and event-driven (sales, advertising, mailings). For each kind it helps to have a separate coefficient or a mark in the calendar. Compare the current period with the same period last year. For a new store, rely on the category and general market patterns.
One-off spikes should not enter the calculation as the norm. A large wholesale order or a rush caused by advertising is better marked as an event and excluded from the regular average.
If data is kept in several places, consolidate it in one. Different SKUs for the same item, returns recorded as sales and manual stock corrections are what most often spoil a forecast.
Time series forecasting models, including machine learning methods, take seasonality, prices, promotions and other factors into account at the same time. They make sense under these conditions:
New products have no history. In that case you use similar products from the same category, a small trial batch and a quick review after the first sales. Some inventory systems and services for stores already have a forecasting module, so check whether your system has one before buying a separate tool. Choose a tool that explains its forecast and lets you correct it manually.
A forecast serves decisions about money, so check it against what the business can afford. Excess stock ties up cash, takes up warehouse space and risks going stale. Insufficient stock leaves shoppers without the product. Priority goes to items with stable demand and good margins; for uncertain products, order smaller batches more often if the supplier's terms allow it.
Advertising and the website are connected to the forecast. Planned promotions and ad campaigns increase demand, so coordinate them with purchasing in advance. In product ads, turn off out-of-stock items, otherwise the budget is spent on clicks without sales. On the website, show current availability and the expected delivery date so the customer knows what to expect.
Simple methods and tidy record-keeping are often enough. AI models make sense when there are many items, seasonality is pronounced and a long history has accumulated.
It depends on the product. To see seasonality, you need a history that covers a full seasonal cycle. For fast-moving products, shorter periods are also useful. The less data you have, the more cautiously you should treat the model's result.
Rely on similar products from the same category, order a small trial batch and review the stock after the first weeks of sales.
In dropshipping, the product is stored at the supplier's, so the main risk is a mismatch in availability data. Sync stock levels with the supplier and show customers current availability. A demand forecast helps agree on volumes with the supplier.
Compare the forecast with actual sales: average deviation by product, share of out-of-stock days, surpluses and dead stock. Improvement in these indicators matters more than the accuracy of any single forecast.
Demand forecasting in an online store starts with order in your records; algorithms come as the second step. A small store can manage with average sales, seasonal coefficients and safety stock; AI models are worth trying on the part of the assortment with a long history. Compare the forecast with actual results every week and adjust the rules. Start with one product group and expand the set when the forecast matches actual results. The result depends on the niche and the stability of supply, so rely on your own data. LightSEO is a digital agency from Mukachevo that has been working with online stores and advertising since 2015, so it can help with launching an online store and Google Ads, where up-to-date product availability matters.
