What a borrowed forecast is

A product that has never sold has no record of its own, so the forecasting engine borrows one: it reads how comparable past styles sold from their launch week on, blends them into one curve, and gives the new product that curve. The engine's own sentence for it is:

Forecast borrowed from comparable past styles.

The line reads From an analog, the forecasting engine's basis word, and the forecasting engine treats the curve like any other forecast. The moment the product sells, its own sales take over. This is the forecasting engine's work, not the baseline engine's; the baseline engine reads the result as its last rung.

Which products get a borrowed forecast

Every product the forecast has no sale for inside the last year is looked at, on every sync. Most are refused, each with its reason, before any borrowing starts:

• A new colour or size on a style whose other options already sell is refused: those options do not say what this one will sell.
• A product listed a long time and never sold is refused, unless a live Commitment or line plan declares a complete window for it that has not closed: a planner putting a style into a live plan says it is coming, which outranks an inference from its creation date.
• A product that sold in the past and not in over a year is refused, and is forecast again from its own sales once it sells.

What remains, a genuinely new product, is offered to its analogs.

How the comparable past styles are chosen

Analogs are proposed each night: image matching reads the product pictures at 03:30 UTC, and the analog proposals run at 04:30 UTC. For each new product the engine ranks past products in the same category by text, keeps up to twenty above a similarity floor, and weights each by how far it clears that floor. A product in no category, or in the default one, is matched to nothing. How a product looks is not part of the ranking; a matching image is still shown as a reason beside the proposal. The blend reads each analog's whole style, not one colourway, over its first seventy-eight weeks, lined up on the analog's own launch week. The level is then split across the new product's sizes by its size curve.

Proposals are made only for a tenant whose own subscription names Analogs; Analogs is included with Tightly Essentials+ and Tightly Pro. Where nothing is proposed, as on Lite and Essentials, a never-sold product finds no comparable past style and reads the no-comparable-style sentence below.

Where the borrowed curve lands in time

The curve is laid over the first window that is declared: an active Commitment's window covering the product, then a line plan item's selling window, then the product's own selling window. With none, it runs twelve weeks from the day the product arrived in the catalogue.

The borrowed forecast's sentences

For every product it looked at, the engine records one sentence. These are the ones a never-sold product most often carries, character for character:

This is a new option on a style that already sells. It has no sales of its own, and the style's other options do not show what this one will sell, so no forecast is offered for it.

This product has been listed for a long time and has never sold, so there is no demand to forecast from.

This product sold in the past and has not sold in over a year, so there is no recent demand to forecast from. A product that sells again is forecast from those new sales.

No comparable past style has enough sales history to borrow a forecast from.

Comparable past styles were found, but they were out of stock for the whole period we can see, so their sales do not show what demand actually was.

Comparable past styles can show how much this product will sell but not when, so no week-by-week forecast is offered.

How much this product will sell is borrowed from comparable past styles. None of them could show when it will sell, so the weekly split follows the seasonal pattern of past trade rather than a comparable style's own curve.

This product's selling window is shorter than a week, which is too short to place a week-by-week forecast into.

A product with no sentence was never looked at, which is a different fact from one that was looked at and refused.

What you can change in a borrowed forecast

• Where proposals are made, a product given a category can be matched; one in no category, or in the default one, cannot.
• On Essentials+ and Pro, Analogs holds the proposals: Not this one takes a proposed analog out of the blend for good, and Use as the analog keeps it and puts it in place.
• On Pro, a Commitment's window, or a line plan item's selling window, moves where the curve lands.

Nothing is pressed to rerun it: the next sync reads the change. A viewer reads; a member or an admin makes these changes.

Glossary line

A borrowed forecast is the curve a never-sold product takes from comparable past styles until its own sales take over.