What the forecasting engine is

The forecasting engine predicts each line's daily sales. Source updates, forecast calculation and confidence have separate run times. Monday's accuracy pass at 06:30 UTC scores the seven completed local days; the 06:45 UTC re-score selects methods from each line's history. Its page is Demand forecasting, on every plan.

When the forecast is built and refreshed

Forecast engine status shows the cadence. With daily scheduling, a full run is due at 04:00 UTC; intervening syncs update source data and dependent figures without rebuilding every forecast. Source pauses and manual or historical import controls still apply.

Recalculate now uses stored sales and existing methods; it fetches no source data and does not rerun the re-score. An explicit incremental recovery also keeps the existing curves.

Approved baselines retain their quantities, dates and references. Supported evidence changes refresh execution without another approval. Missing or conflicting evidence holds the affected figures; it never becomes zero demand.

How far ahead the forecast reaches

The forecast reaches 58 weeks ahead, with daily quantities per variant, location and channel. Demand forecasting adds those days into periods by the window you pick: a window of twelve days or less reads by day, under twelve weeks by week, and anything longer by month. A line's Forecast range, in its drawer, is the band its sales are expected to land inside about eight weeks in ten, drawn from how far the methods missed on your own past weeks.

How the forecast is made for each line

A line with enough history is forecast from its own sales by the method the Monday re-score chose for it. A line too short for a re-score keeps the method chosen for lines like it, still reading its own sales. A product that has never sold is forecast from look-alikes, past products that sold like it, on Essentials+ and Pro, which include Analogs. On Lite and Essentials none is proposed, so such a product has none to follow and says so.

• Forecast methods and the Monday re-score: which methods exist and how one is chosen.
• Forecast accuracy: how the forecast is scored against what sold.
• Confidence as a word: how each line is graded.
• The forecast's words: what each chip on a line means.
• Manual forecast: typing your own figure over the system's.

Within a commitment, active forecasting and replenishment use eligible DTC consumer demand. Customer orders remain separate buying requirements. Supported account sell-out forecasts belong to the account’s Plan in Connect; a customer-order channel alone does not supply consumer evidence.

What each plan sees of the forecast

Forecast methods, confidence and accuracy are the same on every plan. Pro adds season and commitment scope and re-scores declared styles within their selling dates. The Analogs page is on Pro; other plans with look-alike entitlement use automatically chosen sources.

The forecast's as-of and state sentences

The as-of marks the forecast run; Details holds the source sync. Incremental updates do not reset the full forecast's age. Daily forecasts become Late after 26 hours and Stale after 48 hours without a successful full run. Source freshness retains its four-hour promise.

The engine supplies these state messages:

The forecast has not run for this tenant yet. Forecast as of {as_of}; a run was due {due}. The last run rewrote {written} of {expected} lines; the rest are older. Today's forecast run did not finish; figures are as of {as_of}. No sales have synced yet, so there is nothing to forecast. Forecasting is off for this tenant. Paused with the connection; the forecast waits for the sync. This tenant is not on the forecast run.

On Demand forecasting that sentence is in the Forecast performance drawer.

What you can do about the forecast

Recalculate now stands beside the engine's sentence in the Forecast performance drawer. A seat that can write may press it, once every two hours for your tenant, and it is refused while a run is already going. A paused forecast offers Integrations instead, because the sync is what stopped. To change a figure, type a Manual forecast on Demand forecasting. Ask Tightly can explain the forecast and prepare a change for exact variants, channels and dates. It saves only after you confirm, then provides the same scoped forecast to check. It cannot press Recalculate now for you. Nobody picks or pins a line's method: the Monday re-score decides it from your own sales.

Saved assumptions

A declared look-alike takes precedence over automatic sources for its channel and leaves the source product active. Other channels keep their own source. Check the result after recalculation; saving the mapping alone does not mean the forecast has refreshed. Reference evidence remains distinct from the new product's own measured history.

A manual forecast adjustment belongs to its selected products, accounts or channels and dates. Clearing it restores the current calculated basis. Historical sales imports do not create forecasts that were never recorded; historical comparisons require recorded forecast vintages. Constructed demo comparisons illustrate behavior; they are not a customer's measured forecast record. A forecast change places no order.

Glossary line

What the forecasting engine expects each line to sell, with its basis and confidence.

Accuracy is shown only for a measured selection. Where the selected commitment, SKU or account has no scored result, accuracy remains unavailable.