Guide
What Skuvelo uses to plan
What connections, imports, history, roles, and devices contribute to planning, and where the limits are.
No verified update date is available
Skuvelo works from the information your team connects or imports. This guide sets out what each part contributes so you can start with the right expectations.
Open Integrations to see the current status of each supported connection and to start an available connection.
Sales history
- Skuvelo uses: imports and supported channel connections bring order history into the forecast. Around 12 months gives the forecast a useful baseline.
- Your decision or input: the forecast reads the latest 110 weeks, so an item with two full years of its own sales uses its own seasons when both years show the same strong pattern. An item with one full year uses that year's shape when the season in it is clear. Otherwise, below two full years, an item's seasonality comes from like items. Once the item has eight weeks of its own sales, Skuvelo takes the closest group where at least two other items have a full season (its style, then family, then brand within its category, then category) and borrows that group's shape only when the shape is strong. A seasonal shape is never chosen by how well it scored on past weeks: it is used because the pattern is clear. Otherwise shape a season yourself with the seasonal swing on the demand grid. Shorter history can still produce conservative recommendations. For an item with no history yet, see how to forecast demand for a new product.
How the forecast is calculated
The math behind the forecast numbers runs on your own shipped units, in four steps:
- The weekly history: for each item, the forecast counts the units shipped in every completed week, Monday to Sunday in your workspace time zone, over the latest 110 weeks. A week when the item was out of stock or unlisted is filled in from the days it could sell, or from its recent normal weeks, and is only ever raised, so a stockout never pulls the forecast down.
- The method: each forecast first sorts the item by its latest 52 weeks: a frequent seller, which sells in most weeks, or an occasional one. Every item starts on the default: its average week over the latest 52 weeks (every week since a sales level you confirmed changed, or since its first sale when it has less than a year), times its group's growth, with half of any seasonal shape. The group is its product family, else its main supplier, else your whole business, and its growth is its latest 13 weeks against the same 13 weeks a year earlier, counting only the items that sold in both, held between -50% and +50%. The item page shows it as "growth applied +X%", with "(capped)" when the limit held it. For a frequent seller the default is challenged by exponential smoothing at four settings, plus trend smoothing when its recent weeks show a clear trend; for an occasional seller, by a method for slow, lumpy demand. By rule rather than by test, an item that is Winding down or has sold nothing for 13 weeks is not held on the default, and an occasional one then takes a method that lets the forecast fade; a launching occasional seller keeps the slow-demand method. Each setting forecasts weeks it has not yet seen and is scored on its error over your buy window: the total sold from the end of the item's lead time to the end of the weeks an order covers (your target cover, 4 weeks unless you set it). The smallest error wins, and when the runner-up's error is within 10% of it the two are averaged. A winner replaces the default only when it clearly does better: at least 10% less error, and closer in most of the windows scored, on two monthly forecasts in a row; a method in use that no longer clearly does better than the default goes back to it. Between two other methods, a new winner replaces the one in use after beating it by at least 5% on two monthly forecasts in a row. Either way the first month blends the two; a second forecast in the same month counts as no new evidence. A change of planning phase or seasonal shape, or a change in how often the item sells, starts the comparison afresh, back on the default; a seasonal shape is put on or taken off only when two monthly forecasts in a row agree. When a forecast would move an item's buy-window total from the previous one without new evidence (a week's sales outside that forecast's range, a change of planning phase or confirmed sales level, of lead time or cover, or a correction to past sales), the move is held to the smaller of 20% and what the method's smoothing weight times the new weeks' misses would move it, and the rest follows over the next forecasts. When the item has a seasonal shape, as set out under sales history, the method forecasts its level with the season taken out and the shape is put back on. Last week's level, a 13-week average and the same week last year are scored beside the winner as yardsticks and are never picked.
- When it runs: a new forecast is made once a month, in the first week of the month. Between those, a new one is made only when you confirm something the forecast reads: a new sales level, a "yes, demand changed" answer, a one-off week or a market event. An ordinary week of sales does not change it. Your own numbers, events and group changes reach the plan as soon as you save them, without waiting for the next forecast.
- The weekly number: each week's forecast is the winning method's forecast of units shipped that week. The low and high around it widen with that method's measured error. Over the buy window the range is sized from the method's past buy-window errors so the actual total lands inside it about 8 times in 10; an item without enough history borrows the spread of similar items, and a new item gets half its expected units either way.
- The horizon: the forecast runs week by week out to your near-term fence, the longest lead time plus review period across your active suppliers, and never fewer than 13 weeks, plus 6 more so a month-old forecast still covers the fence week by week. Alongside them it runs 24 calendar months starting the month after the one the current week began in. Replenishment covers the longer of those weekly rows or the lead time plus your review period, and spreads each month's forecast evenly across its days past the last weekly row.
Stock
- Skuvelo uses: supported connections can bring stock observations with product and order data.
- Your decision or input: your physical count is authoritative. Set opening balances or import a stock count before treating on-hand stock as established.
Sync timing
- Skuvelo does: queued work is processed about every 60 seconds and connected stores are checked for changes about every 15 minutes.
- Your decision or input: a store change can take roughly 15 minutes to appear. A healthy connection with no update for over an hour needs attention.
Spreadsheet orders
- Skuvelo uses: order history accepts a
statuscolumn and uses fulfilled orders as shipment demand. - Your decision or input:
statusis required. A blank value is rejected; it is never treated as an open order.
Permissions and phone access
- Skuvelo does: Planners manage inventory and catalog work, raise purchase orders, and can decide workflow approvals such as count variances. Purchasing, Owners, and Admins approve purchase orders. Phone views support routine review, follow-up, approvals available to the role, and shipment work.
- Your decision or input: Owners and Admins set approval policy and manage the team. Creating or editing suppliers needs a tablet or desktop.
Every new connection begins in Observe mode. Skuvelo reads data and shows what it sees; it does not write to your store or create a purchase order. After onboarding, recommendations and draft purchases still need a person to review and approve them.
For field-by-field import requirements, see bringing your data in from a spreadsheet. For a store that seems delayed, see why your store can look out of sync.
Try the plan on your own stock.
Start a trial with your own sales and stock, or look around the live demo first.