— The Vocabulary
The terms the engine uses, in plain English.
We refuse to dress up the engine as more than it is. Every term used anywhere on the platform — in a brief, in a chart, in a depth-of-buy line — is defined here, in the same plain English a senior buyer would use across a desk.
21 terms · A → Z
A floor-length, fitted-bodice, flared-skirt silhouette — built for festive and wedding-guest dressing.
e.g. A pre-Diwali anarkali in monsoon-mint chanderi sits at the heart of every September runway.
The hand-curated index of 247 colour, fabric, silhouette, print, and occasion tags the engine uses to read every garment. India-first by design — anarkali is a silhouette, not "a kind of dress".
A tie-and-dye resist print, predominantly Gujarati and Rajasthani. Reads on the runway as heritage occasion-wear.
A forecast range where the stated confidence (e.g. 80%) actually holds up against historical outcomes. We re-check calibration on every account.
e.g. An 80% interval that captures the real sell-through 8 times out of 10 over the last twelve windows.
A featherweight silk-cotton blend from Madhya Pradesh — translucent, lightly woven, a summer-festive staple.
A matched two-piece set (top + bottom or kurta + pant) sold as one assortment unit. The fastest-growing party-wear silhouette on Indian D2C floors.
A statistical wrapper around a forecast that turns a point prediction into an honest range. Wider band, thinner signal.
The order quantity that maximises expected gross profit, given the demand band, the unit margin, the markdown risk, and the leftover salvage value.
e.g. On a ₹2,400 lehenga with 45% margin and a wide demand band, the cost-optimal depth is often 20–30% under the gut-feel buy.
How many units of a single SKU you order. The defining lever of Indian D2C buying — buy deep on the wrong one, ruin your quarter.
A way of finding garments that look similar to each other, not by matching tags but by matching the engine's visual vocabulary. Used to surface runway-look analogues for your assortment.
Combining several forecasting techniques into one prediction. The strength of the ensemble is in the disagreement — when models agree, we're confident; when they don't, the band widens honestly.
End-of-season sale — the twice-yearly markdown window (typically January and July) that flushes the assortment before the next book lands.
A dated period (Diwali, Karva Chauth, Navratri, Eid, Christmas, etc.) where category demand multiplies. The engine prices forty-two of these into the forecast.
Mean Absolute Percentage Error. The simplest accuracy yardstick — how far off the forecast was, as a percent of actual sales. Lower is better.
e.g. A MAPE of 18% on chanderi anarkalis means the forecast missed the real number by ~18% on average.
An anonymised roll-up of similar Indian D2C brands' performance. Used to benchmark your account without ever surfacing any individual peer's SKU data.
The range within which actual sales are likely to fall, at a stated confidence level. Forecasts without intervals are guesses dressed up.
e.g. Forecast 1,200 units; 80% interval 950–1,460. The buyer plans the depth-of-buy on the band, not the midpoint.
A look from a tracked designer's runway that closely resembles a SKU in your assortment. Used to attach a cultural reference to a buying call.
The percentage of units sold against units bought, within a defined window. The honest measure of whether the buy worked.
Weighted Absolute Percentage Error. Like MAPE, but weighted by units — large-volume SKUs count more. The better yardstick for assortment-level accuracy.
The single largest Indian D2C occasion category — anything you wear to someone else's wedding. Mehendi, sangeet, reception, cocktail; not the bride.
Metallic thread (traditionally gold or silver, increasingly synthetic) woven or embroidered into festive cloth. Banarasi zari is the heritage gold standard.
— Back to the claim
Read it in context · the FAQ