What a forecast is
A forecast is a best guess of how many units of an item will sell at a location in each coming week. Insights looks eight weeks ahead.
It is based only on your own past sales of that item at that location. It is a guide, not a promise.
Open the forecast
- Go to Insights → Forecast.
- Choose the Location.
- Choose Sort by: Highest forecast, Lowest cover, Slowest movers or Best accuracy.
- Click an item row, or its Chart button, to see the chart for that item.
The table
On hand: stock at the location now. Avg / day: average units sold per day over the last 28 days. Next 28 d: units expected to sell in the next 28 days.
Cover (d): about how many days the current stock will last at the expected rate. A cover under 7 days is shown in red.
Model: the method used for this item. Error: how far off that method was when tested on the last four weeks. "not enough history" means no error can be measured yet.
The chart
Bars (Sold per week) show what really sold in recent weeks. The line (Forecast) shows the expected sales for each coming week. The "today" mark separates the past from the future.
The shaded area is the Likely range. Most weeks should land inside it. A wide range means sales of this item jump around, so keep a little extra safety stock. Point at a dot to see the expected figure and its range.
Below the chart you see Next 28 days, On hand, Days of cover, Avg per day (28 d), Last sale and History (how many days of sales Insights has for this item).
How the method is chosen
Insights tries several methods on every item: a 28-day average, a weekday pattern, a trend with a weekly season, and two methods for items that sell only now and then (shown as Intermittent).
Each method pretends it is four weeks ago and predicts the last four weeks. The one that came closest to what really sold is used for the forecast. Its miss is shown as the error: 20 % means the weekly totals were off by about a fifth.
Why history matters
An item with fewer than four weeks of sales shows the model "Too new to model". Its forecast is a simple average and no accuracy is claimed.
The more weeks of sales an item has, the better the pattern can be learnt. A trend with a weekly season needs about two months of sales before it can be used.
If an item shows "No forecast yet for this item", it will appear after the next refresh.
Poya days and holidays
Sri Lankan Poya days, public holidays and weekends are known to Insights. When your history has enough of them, it learns how your sales change on those days and adjusts the weeks that contain them.
When a Poya day or holiday falls inside the chart, it is listed under the chart as "Poya & holidays in view".
Promotions and price changes are not yet part of the forecast. After a big promotion, expect the next weeks' forecast to run a little high.