Walk into most planning meetings in the spring and you will hear the same sentence. "We did four million in the category last year, we are targeting plus eight, so plan the buy at four-three-two." Heads nod. The number goes in the sheet. The buy gets placed. Everyone treats it as a plan.
It is not a plan. It is arithmetic performed on a memory. Last year plus a percentage takes whatever happened last season, good decisions and bad, lucky weeks and stockouts, and multiplies all of it by a single growth factor. Whatever you got wrong last year, you have just bought again, only bigger.
The comfort of the method is exactly the problem. It feels grounded because it starts from real history. But it treats that history as if it were demand, when most of it is a record of your own past constraints: what you happened to have in stock, where you happened to allocate it, and how the weather and the promo calendar happened to fall.
The percentage buries every mistake it should surface
One growth factor cannot tell a winner from a stockout.
Think about what a flat uplift does to a category that had a rough year. Say mediums sold out in week three across half the doors and never got replenished. In the sales history, that style looks like a modest performer. Apply plus eight and you buy a modest amount again, plus a sliver, and you stock it out again in week three. The method has laundered a stockout into a demand signal and told you to repeat it.
Now take the style that got over-bought and clearanced. Its sales history looks strong, because you moved a lot of units, at 40 percent off. Plus eight buys it deep again, and you discount it again. The winner and the loser both got the same treatment, because a single percentage cannot tell them apart. It has no idea which units sold because customers wanted them and which sold because you cut the price to move them.
This is the quiet cost of the method: it is not just imprecise, it is systematically biased toward repeating whatever you were forced to do last year. It has no way to distinguish demand from availability, price from desire, or a genuine trend from a one-off week.
And it gets worse the faster the business changes. A flat uplift assumes next year looks like last year with everything scaled up in lockstep, which is only true for a brand that never launches anything new, never shifts channel mix, never changes price and never faces a competitor doing any of those things. In other words, it is true for no brand at all. The moment your DTC channel grows faster than wholesale, or a category catches fire, or a hero silhouette runs its course, the single percentage is not just slightly off. It is applying last year's shape to this year's business, and the shape is the part that matters.
The tell is in how these plans get defended. When someone asks why the buy is what it is, the honest answer is "it is last year plus eight," which is not a reason, it is a procedure. Nobody in the room can point to the demand evidence for the number, because there is none. There is only the memory, grown.
improvement in forecast accuracy (wMAPE) on hero SKUs when the buy is planned by a trained model at SKU by size by channel instead of a last-year baseline.
Fourteen points is not a rounding difference. It is the gap between a buy you have to hedge against everywhere and a buy you can place with confidence where it counts. The uplift comes from the model reading the parts of the signal a percentage flattens: which sizes actually cleared, which doors actually pulled, and which weeks were real demand versus a promo spike.
A trained model reasons at the level you actually buy
You order a style, a size, a channel. Plan there, not one number up.
Here is the difference that matters. Last year plus a percentage lives at the category, because that is the only level a single number can describe. But nobody places an order for a category. You order a specific style, in a specific size curve, for a specific channel, on a specific lead time. The plan and the buy are speaking different languages.
A trained model closes that gap by forecasting at SKU by size by channel: the level the order is actually placed. It learns the medium sells three times as fast as the extra-large in this silhouette, that the website skews smaller than wholesale, that this price point pulls harder in the fourth quarter. None of that is visible in a category total, and none of it survives a flat uplift.
It also carries a confidence range on every prediction, which the percentage never does. A growth factor gives you a point estimate and hides the uncertainty completely, so you cannot tell a forecast you should trust from one you should not. The model shows you the range, so you know exactly which parts of the buy carry risk and which are close to certain. You stop insuring against the whole order and start covering only the parts that are genuinely uncertain.
There is a second thing a trained model does that arithmetic cannot: it uses more than your own history. A percentage can only grow the past you personally recorded, so a new style, a new door or a new category starts from nothing and gets the crudest possible guess. A model learns from the behavior of comparable products across the whole assortment, so a launch inherits the demand shape of the styles most like it rather than a category average or a manual hunch. The forecast is informed even where your own file is thin, which is precisely where a percentage is at its worst.
What a flat uplift leaves stranded versus a trained plan
The overhang is not spread evenly. It sits in the styles and sizes a percentage bought again without asking why they underperformed.
Ten points of inventory is a lot of cash tied up in stock that will only ever clear at markdown. It is also the cash you need for the next buy. That is how the method perpetuates itself: last year's stranded stock funds this year's caution, and the percentage keeps everyone from asking why the same styles keep ending up on the clearance rack.
Stop growing the past and start predicting the season
Cold-start the new, re-forecast the old, and buy against demand you can actually see.
The move off last year plus a percentage is not a bigger spreadsheet. It is putting the demand forecast on a trained model that reads the size curve, the channel mix and the price effect the merchants already see in real life. Carryover styles get re-forecast against their real history, cleaned of the weeks where availability, not demand, set the number. New styles cold-start from their nearest historical cluster, matched on category, price point and silhouette, so a launch ships with a real forecast instead of the category mean plus eight.
It also stops being a once-a-year event. A percentage is set in the spring and then defended for a year, because recomputing it means reopening the whole exercise. A model re-forecasts continuously against live sell-through, returns and channel mix, so the number is never more than a few hours behind the business. When a category the plan grew at eight percent turns out to be running at plus twenty, you find out in-season, while there is still time to chase it, not at the post-mortem where you agree to plan it better next year.
A percentage is a wish about the total. A forecast is a prediction about the thing you are actually going to order.
Get that right and the spring meeting changes. You stop arguing about the growth number and start looking at where the model is confident, where it is not, and where the demand has genuinely moved since last year. The buy comes down in the places you were over-insuring and up in the places you were quietly starving, and none of it depends on last year having been a good year.